def __init__( self, root_dir=None, scm=None, rev=None, subrepos=False, uninitialized=False, config=None, url=None, repo_factory=None, ): from dvc.config import Config from dvc.data_cloud import DataCloud from dvc.fs.local import LocalFileSystem from dvc.lock import LockNoop, make_lock from dvc.objects.db import ODBManager from dvc.repo.live import Live from dvc.repo.metrics import Metrics from dvc.repo.params import Params from dvc.repo.plots import Plots from dvc.repo.stage import StageLoad from dvc.scm import SCM from dvc.stage.cache import StageCache from dvc.state import State, StateNoop self.url = url self._fs_conf = {"repo_factory": repo_factory} if rev and not scm: scm = SCM(root_dir or os.curdir) self.root_dir, self.dvc_dir, self.tmp_dir = self._get_repo_dirs( root_dir=root_dir, scm=scm, rev=rev, uninitialized=uninitialized ) if scm: self._fs = scm.get_fs(rev) else: self._fs = LocalFileSystem(url=self.root_dir) self.config = Config(self.dvc_dir, fs=self.fs, config=config) self._uninitialized = uninitialized self._scm = scm # used by RepoFileSystem to determine if it should traverse subrepos self.subrepos = subrepos self.cloud = DataCloud(self) self.stage = StageLoad(self) if scm or not self.dvc_dir: self.lock = LockNoop() self.state = StateNoop() self.odb = ODBManager(self) else: self.lock = make_lock( os.path.join(self.tmp_dir, "lock"), tmp_dir=self.tmp_dir, hardlink_lock=self.config["core"].get("hardlink_lock", False), friendly=True, ) # NOTE: storing state and link_state in the repository itself to # avoid any possible state corruption in 'shared cache dir' # scenario. self.state = State(self.root_dir, self.tmp_dir, self.dvcignore) self.odb = ODBManager(self) self.stage_cache = StageCache(self) self._ignore() self.metrics = Metrics(self) self.plots = Plots(self) self.params = Params(self) self.live = Live(self) self.stage_collection_error_handler: Optional[ Callable[[str, Exception], None] ] = None self._lock_depth = 0
class Repo(object): DVC_DIR = ".dvc" from dvc.repo.destroy import destroy from dvc.repo.install import install from dvc.repo.add import add from dvc.repo.remove import remove from dvc.repo.lock import lock as lock_stage from dvc.repo.move import move from dvc.repo.run import run from dvc.repo.imp import imp from dvc.repo.imp_url import imp_url from dvc.repo.reproduce import reproduce from dvc.repo.checkout import checkout from dvc.repo.push import push from dvc.repo.fetch import fetch from dvc.repo.pull import pull from dvc.repo.status import status from dvc.repo.gc import gc from dvc.repo.commit import commit from dvc.repo.diff import diff from dvc.repo.brancher import brancher from dvc.repo.get import get from dvc.repo.get_url import get_url from dvc.repo.update import update def __init__(self, root_dir=None): from dvc.state import State from dvc.lock import Lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.repo.metrics import Metrics from dvc.scm.tree import WorkingTree from dvc.repo.tag import Tag root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir, repo=self) self.tree = WorkingTree(self.root_dir) self.lock = Lock(self.dvc_dir) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self, self.config.config) core = self.config.config[Config.SECTION_CORE] level = core.get(Config.SECTION_CORE_LOGLEVEL) if level: logger.setLevel(level.upper()) self.cache = Cache(self) self.cloud = DataCloud(self) self.metrics = Metrics(self) self.tag = Tag(self) self._ignore() def __repr__(self): return "Repo: '{root_dir}'".format(root_dir=self.root_dir) @classmethod def find_root(cls, root=None): if root is None: root = os.getcwd() else: root = os.path.abspath(os.path.realpath(root)) while True: dvc_dir = os.path.join(root, cls.DVC_DIR) if os.path.isdir(dvc_dir): return root if os.path.ismount(root): break root = os.path.dirname(root) raise NotDvcRepoError(root) @classmethod def find_dvc_dir(cls, root=None): root_dir = cls.find_root(root) return os.path.join(root_dir, cls.DVC_DIR) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False): from dvc.repo.init import init init(root_dir=root_dir, no_scm=no_scm, force=force) return Repo(root_dir) def unprotect(self, target): return self.cache.local.unprotect(PathInfo(target)) def _ignore(self): from dvc.updater import Updater updater = Updater(self.dvc_dir) flist = [ self.lock.lock_file, self.config.config_local_file, updater.updater_file, updater.lock.lock_file, ] + self.state.files if self.cache.local.cache_dir.startswith(self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def check_dag(self, stages): """Generate graph including the new stage to check for errors""" self.graph(stages=stages) @staticmethod def _check_cyclic_graph(graph): import networkx as nx from dvc.exceptions import CyclicGraphError cycles = list(nx.simple_cycles(graph)) if cycles: raise CyclicGraphError(cycles[0]) def _get_pipeline(self, node): pipelines = [i for i in self.pipelines() if i.has_node(node)] assert len(pipelines) == 1 return pipelines[0] def collect(self, target, with_deps=False, recursive=False): import networkx as nx from dvc.stage import Stage if not target or (recursive and os.path.isdir(target)): return self.active_stages(target) stage = Stage.load(self, target) if not with_deps: return [stage] node = relpath(stage.path, self.root_dir) G = self._get_pipeline(node) ret = [] for n in nx.dfs_postorder_nodes(G, node): ret.append(G.node[n]["stage"]) return ret def used_cache( self, target=None, all_branches=False, active=True, with_deps=False, all_tags=False, remote=None, force=False, jobs=None, recursive=False, ): """Get the stages related to the given target and collect the `info` of its outputs. This is useful to know what files from the cache are _in use_ (namely, a file described as an output on a stage). The scope is, by default, the working directory, but you can use `all_branches` or `all_tags` to expand scope. Returns: A dictionary with Schemes (representing output's location) as keys, and a list with the outputs' `dumpd` as values. """ cache = {} cache["local"] = [] cache["s3"] = [] cache["gs"] = [] cache["hdfs"] = [] cache["ssh"] = [] cache["azure"] = [] for branch in self.brancher(all_branches=all_branches, all_tags=all_tags): if target: if recursive and os.path.isdir(target): stages = self.stages(target) else: stages = self.collect(target, with_deps=with_deps) elif active: stages = self.active_stages() else: stages = self.stages() for stage in stages: if active and not target and stage.locked: logger.warning( "DVC-file '{path}' is locked. Its dependencies are" " not going to be pushed/pulled/fetched.".format( path=stage.relpath)) for out in stage.outs: scheme = out.path_info.scheme used_cache = out.get_used_cache(remote=remote, force=force, jobs=jobs) cache[scheme].extend( dict(entry, branch=branch) for entry in used_cache) return cache def graph(self, stages=None, from_directory=None): """Generate a graph by using the given stages on the given directory The nodes of the graph are the stage's path relative to the root. Edges are created when the output of one stage is used as a dependency in other stage. The direction of the edges goes from the stage to its dependency: For example, running the following: $ dvc run -o A "echo A > A" $ dvc run -d A -o B "echo B > B" $ dvc run -d B -o C "echo C > C" Will create the following graph: ancestors <-- | C.dvc -> B.dvc -> A.dvc | | | --> descendants | ------- pipeline ------> | v (weakly connected components) Args: stages (list): used to build a graph, if None given, use the ones on the `from_directory`. from_directory (str): directory where to look at for stages, if None is given, use the current working directory Raises: OutputDuplicationError: two outputs with the same path StagePathAsOutputError: stage inside an output directory OverlappingOutputPathsError: output inside output directory CyclicGraphError: resulting graph has cycles """ import networkx as nx from dvc.exceptions import ( OutputDuplicationError, StagePathAsOutputError, OverlappingOutputPathsError, ) G = nx.DiGraph() G_active = nx.DiGraph() stages = stages or self.stages(from_directory, check_dag=False) stages = [stage for stage in stages if stage] outs = {} for stage in stages: for out in stage.outs: if out.path_info in outs: stages = [stage.relpath, outs[out.path_info].stage.relpath] raise OutputDuplicationError(str(out), stages) outs[out.path_info] = out for stage in stages: for out in stage.outs: for p in out.path_info.parents: if p in outs: raise OverlappingOutputPathsError(outs[p], out) for stage in stages: stage_path_info = PathInfo(stage.path) for p in chain([stage_path_info], stage_path_info.parents): if p in outs: raise StagePathAsOutputError(stage.wdir, stage.relpath) for stage in stages: node = relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) G_active.add_node(node, stage=stage) for dep in stage.deps: if dep.path_info is None: continue for out in outs: if (out == dep.path_info or dep.path_info.isin(out) or out.isin(dep.path_info)): dep_stage = outs[out].stage dep_node = relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) if not stage.locked: G_active.add_node(dep_node, stage=dep_stage) G_active.add_edge(node, dep_node) self._check_cyclic_graph(G) return G, G_active def pipelines(self, from_directory=None): import networkx as nx G, G_active = self.graph(from_directory=from_directory) return [ G.subgraph(c).copy() for c in nx.weakly_connected_components(G) ] @staticmethod def _filter_out_dirs(dirs, outs, root_dir): def filter_dirs(dname): path = os.path.join(root_dir, dname) for out in outs: if path == os.path.normpath(out): return False return True return list(filter(filter_dirs, dirs)) def stages(self, from_directory=None, check_dag=True): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories tracked by DVC (e.g. `dvc add data` would skip `data/`) NOTE: For large repos, this could be an expensive operation. Consider using some memoization. """ from dvc.stage import Stage if not from_directory: from_directory = self.root_dir elif not os.path.isdir(from_directory): raise TargetNotDirectoryError(from_directory) stages = [] outs = [] for root, dirs, files in self.tree.walk(from_directory, dvcignore=self.dvcignore): for fname in files: path = os.path.join(root, fname) if not Stage.is_valid_filename(path): continue stage = Stage.load(self, path) for out in stage.outs: if out.scheme == "local": outs.append(out.fspath + out.sep) stages.append(stage) dirs[:] = self._filter_out_dirs(dirs, outs, root) if check_dag: self.check_dag(stages) return stages def active_stages(self, from_directory=None): import networkx as nx stages = [] for G in self.pipelines(from_directory): stages.extend(list(nx.get_node_attributes(G, "stage").values())) return stages def find_outs_by_path(self, path, outs=None, recursive=False): if not outs: # there is no `from_directory=path` argument because some data # files might be generated to an upper level, and so it is # needed to look at all the files (project root_dir) stages = self.stages() outs = [out for stage in stages for out in stage.outs] abs_path = os.path.abspath(path) is_dir = self.tree.isdir(abs_path) def func(out): if out.scheme == "local" and out.fspath == abs_path: return True if is_dir and recursive and out.path_info.isin(abs_path): return True return False matched = list(filter(func, outs)) if not matched: raise OutputNotFoundError(path) return matched def find_out_by_relpath(self, relpath): path = os.path.join(self.root_dir, relpath) out, = self.find_outs_by_path(path) return out def is_dvc_internal(self, path): path_parts = os.path.normpath(path).split(os.path.sep) return self.DVC_DIR in path_parts def open(self, path, remote=None, mode="r", encoding=None): """Opens a specified resource as a file descriptor""" out, = self.find_outs_by_path(path) if out.isdir(): raise ValueError("Can't open a dir") cache_file = self.cache.local.checksum_to_path_info(out.checksum) cache_file = fspath_py35(cache_file) with self.state: cache_info = out.get_used_cache(remote=remote) self.cloud.pull(cache_info, remote=remote) # Since pull may just skip with a warning, we need to check it here if not os.path.exists(cache_file): raise OutputFileMissingError(relpath(path, self.root_dir)) return _open(cache_file, mode=mode, encoding=encoding) @cached_property def dvcignore(self): return DvcIgnoreFilter(self.root_dir)
class Repo: DVC_DIR = ".dvc" from dvc.repo.destroy import destroy from dvc.repo.install import install from dvc.repo.add import add from dvc.repo.remove import remove from dvc.repo.ls import ls from dvc.repo.freeze import freeze, unfreeze from dvc.repo.move import move from dvc.repo.run import run from dvc.repo.imp import imp from dvc.repo.imp_url import imp_url from dvc.repo.reproduce import reproduce from dvc.repo.checkout import _checkout from dvc.repo.push import push from dvc.repo.fetch import _fetch from dvc.repo.pull import pull from dvc.repo.status import status from dvc.repo.gc import gc from dvc.repo.commit import commit from dvc.repo.diff import diff from dvc.repo.brancher import brancher from dvc.repo.get import get from dvc.repo.get_url import get_url from dvc.repo.update import update def __init__(self, root_dir=None, scm=None, rev=None): from dvc.state import State, StateNoop from dvc.lock import make_lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.repo.metrics import Metrics from dvc.repo.plots import Plots from dvc.repo.params import Params from dvc.scm.tree import WorkingTree from dvc.utils.fs import makedirs from dvc.stage.cache import StageCache if scm: # use GitTree instead of WorkingTree as default repo tree instance tree = scm.get_tree(rev) self.root_dir = self.find_root(root_dir, tree) self.scm = scm self.tree = tree self.state = StateNoop() else: root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) if not scm: no_scm = self.config["core"].get("no_scm", False) self.scm = SCM(self.root_dir, no_scm=no_scm) self.tree = WorkingTree(self.root_dir) self.tmp_dir = os.path.join(self.dvc_dir, "tmp") self.index_dir = os.path.join(self.tmp_dir, "index") makedirs(self.index_dir, exist_ok=True) hardlink_lock = self.config["core"].get("hardlink_lock", False) self.lock = make_lock( os.path.join(self.tmp_dir, "lock"), tmp_dir=self.tmp_dir, hardlink_lock=hardlink_lock, friendly=True, ) self.cache = Cache(self) self.cloud = DataCloud(self) if not scm: # NOTE: storing state and link_state in the repository itself to # avoid any possible state corruption in 'shared cache dir' # scenario. self.state = State(self.cache.local) self.stage_cache = StageCache(self) self.metrics = Metrics(self) self.plots = Plots(self) self.params = Params(self) self._ignore() @property def tree(self): return self._tree @tree.setter def tree(self, tree): if is_working_tree(tree) or tree.tree_root == self.root_dir: root = None else: root = self.root_dir self._tree = ( tree if isinstance(tree, CleanTree) else CleanTree(tree, root) ) # Our graph cache is no longer valid, as it was based on the previous # tree. self._reset() def __repr__(self): return f"{self.__class__.__name__}: '{self.root_dir}'" @classmethod def find_root(cls, root=None, tree=None): root_dir = os.path.realpath(root or os.curdir) if tree: if tree.isdir(os.path.join(root_dir, cls.DVC_DIR)): return root_dir raise NotDvcRepoError(f"'{root}' does not contain DVC directory") if not os.path.isdir(root_dir): raise NotDvcRepoError(f"directory '{root}' does not exist") while True: dvc_dir = os.path.join(root_dir, cls.DVC_DIR) if os.path.isdir(dvc_dir): return root_dir if os.path.ismount(root_dir): break root_dir = os.path.dirname(root_dir) message = ( "you are not inside of a DVC repository " "(checked up to mount point '{}')" ).format(root_dir) raise NotDvcRepoError(message) @classmethod def find_dvc_dir(cls, root=None): root_dir = cls.find_root(root) return os.path.join(root_dir, cls.DVC_DIR) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False, subdir=False): from dvc.repo.init import init init(root_dir=root_dir, no_scm=no_scm, force=force, subdir=subdir) return Repo(root_dir) def unprotect(self, target): return self.cache.local.tree.unprotect(PathInfo(target)) def _ignore(self): flist = [self.config.files["local"], self.tmp_dir] if path_isin(self.cache.local.cache_dir, self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def get_stage(self, path=None, name=None): if not path: path = PIPELINE_FILE logger.debug("Assuming '%s' to be a stage inside '%s'", name, path) dvcfile = Dvcfile(self, path) return dvcfile.stages[name] def get_stages(self, path=None, name=None): if not path: path = PIPELINE_FILE logger.debug("Assuming '%s' to be a stage inside '%s'", name, path) if name: return [self.get_stage(path, name)] dvcfile = Dvcfile(self, path) return list(dvcfile.stages.values()) def check_modified_graph(self, new_stages): """Generate graph including the new stage to check for errors""" # Building graph might be costly for the ones with many DVC-files, # so we provide this undocumented hack to skip it. See [1] for # more details. The hack can be used as: # # repo = Repo(...) # repo._skip_graph_checks = True # repo.add(...) # # A user should care about not duplicating outs and not adding cycles, # otherwise DVC might have an undefined behaviour. # # [1] https://github.com/iterative/dvc/issues/2671 if not getattr(self, "_skip_graph_checks", False): self._collect_graph(self.stages + new_stages) def _collect_inside(self, path, graph): import networkx as nx stages = nx.dfs_postorder_nodes(graph) return [stage for stage in stages if path_isin(stage.path, path)] def collect( self, target=None, with_deps=False, recursive=False, graph=None ): if not target: return list(graph) if graph else self.stages if recursive and os.path.isdir(target): return self._collect_inside( os.path.abspath(target), graph or self.graph ) path, name = parse_target(target) stages = self.get_stages(path, name) if not with_deps: return stages res = set() for stage in stages: res.update(self._collect_pipeline(stage, graph=graph)) return res def _collect_pipeline(self, stage, graph=None): import networkx as nx pipeline = get_pipeline(get_pipelines(graph or self.graph), stage) return nx.dfs_postorder_nodes(pipeline, stage) def _collect_from_default_dvcfile(self, target): dvcfile = Dvcfile(self, PIPELINE_FILE) if dvcfile.exists(): return dvcfile.stages.get(target) def collect_granular( self, target=None, with_deps=False, recursive=False, graph=None ): """ Priority is in the order of following in case of ambiguity: - .dvc file or .yaml file - dir if recursive and directory exists - stage_name - output file """ if not target: return [(stage, None) for stage in self.stages] file, name = parse_target(target) stages = [] # Optimization: do not collect the graph for a specific target if not file: # parsing is ambiguous when it does not have a colon # or if it's not a dvcfile, as it can be a stage name # in `dvc.yaml` or, an output in a stage. logger.debug( "Checking if stage '%s' is in '%s'", target, PIPELINE_FILE ) if not (recursive and os.path.isdir(target)): stage = self._collect_from_default_dvcfile(target) if stage: stages = ( self._collect_pipeline(stage) if with_deps else [stage] ) elif not with_deps and is_valid_filename(file): stages = self.get_stages(file, name) if not stages: if not (recursive and os.path.isdir(target)): try: (out,) = self.find_outs_by_path(target, strict=False) filter_info = PathInfo(os.path.abspath(target)) return [(out.stage, filter_info)] except OutputNotFoundError: pass try: stages = self.collect(target, with_deps, recursive, graph) except StageFileDoesNotExistError as exc: # collect() might try to use `target` as a stage name # and throw error that dvc.yaml does not exist, whereas it # should say that both stage name and file does not exist. if file and is_valid_filename(file): raise raise NoOutputOrStageError(target, exc.file) from exc except StageNotFound as exc: raise NoOutputOrStageError(target, exc.file) from exc return [(stage, None) for stage in stages] def used_cache( self, targets=None, all_branches=False, with_deps=False, all_tags=False, all_commits=False, remote=None, force=False, jobs=None, recursive=False, used_run_cache=None, ): """Get the stages related to the given target and collect the `info` of its outputs. This is useful to know what files from the cache are _in use_ (namely, a file described as an output on a stage). The scope is, by default, the working directory, but you can use `all_branches`/`all_tags`/`all_commits` to expand the scope. Returns: A dictionary with Schemes (representing output's location) mapped to items containing the output's `dumpd` names and the output's children (if the given output is a directory). """ from dvc.cache import NamedCache cache = NamedCache() for branch in self.brancher( all_branches=all_branches, all_tags=all_tags, all_commits=all_commits, ): targets = targets or [None] pairs = cat( self.collect_granular( target, recursive=recursive, with_deps=with_deps ) for target in targets ) suffix = f"({branch})" if branch else "" for stage, filter_info in pairs: used_cache = stage.get_used_cache( remote=remote, force=force, jobs=jobs, filter_info=filter_info, ) cache.update(used_cache, suffix=suffix) if used_run_cache: used_cache = self.stage_cache.get_used_cache( used_run_cache, remote=remote, force=force, jobs=jobs, ) cache.update(used_cache) return cache def _collect_graph(self, stages): """Generate a graph by using the given stages on the given directory The nodes of the graph are the stage's path relative to the root. Edges are created when the output of one stage is used as a dependency in other stage. The direction of the edges goes from the stage to its dependency: For example, running the following: $ dvc run -o A "echo A > A" $ dvc run -d A -o B "echo B > B" $ dvc run -d B -o C "echo C > C" Will create the following graph: ancestors <-- | C.dvc -> B.dvc -> A.dvc | | | --> descendants | ------- pipeline ------> | v (weakly connected components) Args: stages (list): used to build a graph, if None given, collect stages in the repository. Raises: OutputDuplicationError: two outputs with the same path StagePathAsOutputError: stage inside an output directory OverlappingOutputPathsError: output inside output directory CyclicGraphError: resulting graph has cycles """ import networkx as nx from pygtrie import Trie from dvc.exceptions import ( OutputDuplicationError, StagePathAsOutputError, OverlappingOutputPathsError, ) G = nx.DiGraph() stages = stages or self.stages outs = Trie() # Use trie to efficiently find overlapping outs and deps for stage in filter(bool, stages): # bug? not using it later for out in stage.outs: out_key = out.path_info.parts # Check for dup outs if out_key in outs: dup_stages = [stage, outs[out_key].stage] raise OutputDuplicationError(str(out), dup_stages) # Check for overlapping outs if outs.has_subtrie(out_key): parent = out overlapping = first(outs.values(prefix=out_key)) else: parent = outs.shortest_prefix(out_key).value overlapping = out if parent and overlapping: msg = ( "Paths for outs:\n'{}'('{}')\n'{}'('{}')\n" "overlap. To avoid unpredictable behaviour, " "rerun command with non overlapping outs paths." ).format( str(parent), parent.stage.addressing, str(overlapping), overlapping.stage.addressing, ) raise OverlappingOutputPathsError(parent, overlapping, msg) outs[out_key] = out for stage in stages: out = outs.shortest_prefix(PathInfo(stage.path).parts).value if out: raise StagePathAsOutputError(stage, str(out)) # Building graph G.add_nodes_from(stages) for stage in stages: for dep in stage.deps: if dep.path_info is None: continue dep_key = dep.path_info.parts overlapping = list(n.value for n in outs.prefixes(dep_key)) if outs.has_subtrie(dep_key): overlapping.extend(outs.values(prefix=dep_key)) G.add_edges_from((stage, out.stage) for out in overlapping) check_acyclic(G) return G @cached_property def graph(self): return self._collect_graph(self.stages) @cached_property def pipelines(self): return get_pipelines(self.graph) @cached_property def stages(self): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories tracked by DVC (e.g. `dvc add data` would skip `data/`) NOTE: For large repos, this could be an expensive operation. Consider using some memoization. """ return self._collect_stages() @cached_property def plot_templates(self): from .plots.template import PlotTemplates return PlotTemplates(self.dvc_dir) def _collect_stages(self): stages = [] outs = set() for root, dirs, files in self.tree.walk(self.root_dir): for file_name in filter(is_valid_filename, files): new_stages = self.get_stages(os.path.join(root, file_name)) stages.extend(new_stages) outs.update( out.fspath for stage in new_stages for out in stage.outs if out.scheme == "local" ) dirs[:] = [d for d in dirs if os.path.join(root, d) not in outs] return stages def find_outs_by_path(self, path, outs=None, recursive=False, strict=True): if not outs: outs = [out for stage in self.stages for out in stage.outs] abs_path = os.path.abspath(path) path_info = PathInfo(abs_path) match = path_info.__eq__ if strict else path_info.isin_or_eq def func(out): if out.scheme == "local" and match(out.path_info): return True if recursive and out.path_info.isin(path_info): return True return False matched = list(filter(func, outs)) if not matched: raise OutputNotFoundError(path, self) return matched def find_out_by_relpath(self, relpath): path = os.path.join(self.root_dir, relpath) (out,) = self.find_outs_by_path(path) return out def is_dvc_internal(self, path): path_parts = os.path.normpath(path).split(os.path.sep) return self.DVC_DIR in path_parts @contextmanager def open_by_relpath(self, path, remote=None, mode="r", encoding=None): """Opens a specified resource as a file descriptor""" tree = RepoTree(self, stream=True) path = os.path.join(self.root_dir, path) try: with self.state: with tree.open( os.path.join(self.root_dir, path), mode=mode, encoding=encoding, remote=remote, ) as fobj: yield fobj except FileNotFoundError as exc: raise FileMissingError(path) from exc except IsADirectoryError as exc: raise DvcIsADirectoryError(f"'{path}' is a directory") from exc def close(self): self.scm.close() @locked def checkout(self, *args, **kwargs): return self._checkout(*args, **kwargs) @locked def fetch(self, *args, **kwargs): return self._fetch(*args, **kwargs) def _reset(self): self.__dict__.pop("graph", None) self.__dict__.pop("stages", None) self.__dict__.pop("pipelines", None) self.__dict__.pop("dvcignore", None)
def test_init_none(tmp_dir): assert isinstance(SCM(os.fspath(tmp_dir), no_scm=True), NoSCM)
def test_init_no_git(tmp_dir): with pytest.raises(SCMError): SCM(os.fspath(tmp_dir))
class Project(object): DVC_DIR = '.dvc' def __init__(self, root_dir): from dvc.logger import Logger from dvc.config import Config from dvc.state import LinkState, State from dvc.lock import Lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.updater import Updater self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir) self.lock = Lock(self.dvc_dir) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self) self.link_state = LinkState(self) core = self.config._config[Config.SECTION_CORE] self.logger = Logger(core.get(Config.SECTION_CORE_LOGLEVEL, None)) self.cache = Cache(self) self.cloud = DataCloud(self, config=self.config._config) self.updater = Updater(self.dvc_dir) self._ignore() self.updater.check() @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False): """ Initiate dvc project in directory. Args: root_dir: Path to project's root directory. Returns: Project instance. Raises: KeyError: Raises an exception. """ import shutil from dvc.scm import SCM, Base from dvc.config import Config root_dir = os.path.abspath(root_dir) dvc_dir = os.path.join(root_dir, Project.DVC_DIR) scm = SCM(root_dir) if type(scm) == Base and not no_scm: msg = "{} is not tracked by any supported scm tool(e.g. git)." raise InitError(msg.format(root_dir)) if os.path.isdir(dvc_dir): if not force: msg = "'{}' exists. Use '-f' to force." raise InitError(msg.format(os.path.relpath(dvc_dir))) shutil.rmtree(dvc_dir) os.mkdir(dvc_dir) config = Config.init(dvc_dir) proj = Project(root_dir) scm.add([config.config_file]) if scm.ignore_file(): scm.add([os.path.join(dvc_dir, scm.ignore_file())]) return proj def destroy(self): import shutil for stage in self.stages(): stage.remove() shutil.rmtree(self.dvc_dir) def _ignore(self): flist = [self.state.state_file, self.state._lock_file.lock_file, self.link_state.state_file, self.link_state._lock_file.lock_file, self.lock.lock_file, self.config.config_local_file, self.updater.updater_file] if self.cache.local.cache_dir.startswith(self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def install(self): self.scm.install() def to_dvc_path(self, path): return os.path.relpath(path, self.root_dir) def add(self, fname): stage = Stage.loads(project=self, outs=[fname], add=True) stage.save() stage.dump() return stage def remove(self, target, outs_only=False): if not Stage.is_stage_file(target): raise NotDvcFileError(target) stage = Stage.load(self, target) if outs_only: stage.remove_outs() else: stage.remove() return stage def lock_stage(self, target, unlock=False): if not Stage.is_stage_file(target): raise NotDvcFileError(target) stage = Stage.load(self, target) stage.locked = False if unlock else True stage.dump() return stage def move(self, from_path, to_path): import dvc.output as Output from_out = Output.loads_from(Stage(self, cwd=os.curdir), [from_path])[0] found = False for stage in self.stages(): for out in stage.outs: if out.path != from_out.path: continue if not stage.is_data_source: msg = 'Dvcfile \'{}\' is not a data source.' raise DvcException(msg.format(stage.rel_path)) found = True to_out = Output.loads_from(out.stage, [to_path], out.cache, out.metric)[0] out.move(to_out) stage_base = os.path.basename(stage.path) stage_base = stage_base.rstrip(Stage.STAGE_FILE_SUFFIX) stage_dir = os.path.dirname(stage.path) from_base = os.path.basename(from_path) to_base = os.path.basename(to_path) if stage_base == from_base: os.unlink(stage.path) path = to_base + Stage.STAGE_FILE_SUFFIX stage.path = os.path.join(stage_dir, path) stage.dump() if not found: msg = 'Unable to find dvcfile with output \'{}\'' raise DvcException(msg.format(from_path)) def run(self, cmd=None, deps=[], outs=[], outs_no_cache=[], metrics_no_cache=[], fname=Stage.STAGE_FILE, cwd=os.curdir, no_exec=False): stage = Stage.loads(project=self, fname=fname, cmd=cmd, cwd=cwd, outs=outs, outs_no_cache=outs_no_cache, metrics_no_cache=metrics_no_cache, deps=deps) if not no_exec: stage.run() stage.dump() return stage def imp(self, url, out): stage = Stage.loads(project=self, cmd=None, deps=[url], outs=[out]) stage.run() stage.dump() return stage def _reproduce_stage(self, stages, node, force): stage = stages[node] if stage.locked: msg = 'DVC file \'{}\' is locked. Its dependecies are not ' \ 'going to be reproduced.' self.logger.warn(msg.format(stage.relpath)) stage = stage.reproduce(force=force) if not stage: return [] stage.dump() return [stage] def reproduce(self, target, recursive=True, force=False): import networkx as nx G = self.graph()[1] stages = nx.get_node_attributes(G, 'stage') node = os.path.relpath(os.path.abspath(target), self.root_dir) if node not in stages: raise NotDvcFileError(target) if recursive: return self._reproduce_stages(G, stages, node, force) return self._reproduce_stage(stages, node, force) def _reproduce_stages(self, G, stages, node, force): import networkx as nx result = [] for n in nx.dfs_postorder_nodes(G, node): try: result += self._reproduce_stage(stages, n, force) except Exception as ex: raise ReproductionError(stages[n].relpath, ex) return result def _cleanup_unused_links(self, all_stages): used = [] for stage in all_stages: for out in stage.outs: used.append(out.path) self.link_state.remove_unused(used) def checkout(self, target=None): all_stages = self.active_stages() if target: if not Stage.is_stage_file(target): raise NotDvcFileError(target) stages = [Stage.load(self, target)] else: stages = all_stages self._cleanup_unused_links(all_stages) for stage in stages: if stage.locked: msg = 'DVC file \'{}\' is locked. Its dependecies are not ' \ 'going to be checked out.' self.logger.warn(msg.format(stage.relpath)) stage.checkout() def _used_cache(self, target=None, all_branches=False, active=True): cache = {} cache['local'] = [] cache['s3'] = [] cache['gs'] = [] cache['hdfs'] = [] cache['ssh'] = [] for branch in self.scm.brancher(all_branches=all_branches): if target: stages = [Stage.load(self, target)] elif active: stages = self.active_stages() else: stages = self.stages() for stage in stages: if active and not target and stage.locked: msg = 'DVC file \'{}\' is locked. Its dependecies are ' \ 'not going to be pushed/pulled/fetched.' self.logger.warn(msg.format(stage.relpath)) for out in stage.outs: if not out.use_cache or not out.info: continue info = out.dumpd() info['branch'] = branch cache[out.path_info['scheme']] += [info] return cache def gc(self, all_branches=False, cloud=False, remote=None): clist = self._used_cache(target=None, all_branches=all_branches, active=False) self.cache.local.gc(clist) if self.cache.s3: self.cache.s3.gc(clist) if self.cache.gs: self.cache.gs.gc(clist) if self.cache.ssh: self.cache.ssh.gc(clist) if self.cache.hdfs: self.cache.hdfs.gc(clist) if self.cache.azure: self.cache.azure.gc(clist) if cloud: self.cloud._get_cloud(remote).gc(clist) def push(self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False): self.cloud.push(self._used_cache(target, all_branches)['local'], jobs, remote=remote, show_checksums=show_checksums) def fetch(self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False): self.cloud.pull(self._used_cache(target, all_branches)['local'], jobs, remote=remote, show_checksums=show_checksums) def pull(self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False): self.fetch(target, jobs, remote=remote, all_branches=all_branches, show_checksums=show_checksums) self.checkout(target=target) def _local_status(self, target=None): status = {} if target: stages = [Stage.load(self, target)] else: stages = self.active_stages() for stage in stages: if stage.locked: msg = 'DVC file \'{}\' is locked. Its dependecies are not ' \ 'going to be shown in status output.' self.logger.warn(msg.format(stage.relpath)) status.update(stage.status()) return status def _cloud_status(self, target=None, jobs=1, remote=None, show_checksums=False): import dvc.remote.base as cloud status = {} for md5, ret in self.cloud.status(self._used_cache(target)['local'], jobs, remote=remote, show_checksums=show_checksums): if ret == cloud.STATUS_OK: continue prefix_map = { cloud.STATUS_DELETED: 'deleted', cloud.STATUS_NEW: 'new', } status[md5] = prefix_map[ret] return status def status(self, target=None, jobs=1, cloud=False, remote=None, show_checksums=False): if cloud: return self._cloud_status(target, jobs, remote=remote, show_checksums=show_checksums) return self._local_status(target) def _read_metric_json(self, fd, json_path): import json from jsonpath_rw import parse parser = parse(json_path) return [x.value for x in parser.find(json.load(fd))] def _do_read_metric_xsv(self, reader, row, col): if col is not None and row is not None: return [reader[row][col]] elif col is not None: return [r[col] for r in reader] elif row is not None: return reader[row] return None def _read_metric_hxsv(self, fd, hxsv_path, delimiter): import csv col, row = hxsv_path.split(',') row = int(row) reader = list(csv.DictReader(fd, delimiter=delimiter)) return self._do_read_metric_xsv(reader, row, col) def _read_metric_xsv(self, fd, xsv_path, delimiter): import csv col, row = xsv_path.split(',') row = int(row) col = int(col) reader = list(csv.reader(fd, delimiter=delimiter)) return self._do_read_metric_xsv(reader, row, col) def _read_metric(self, path, typ=None, xpath=None): ret = None if not os.path.exists(path): return ret try: with open(path, 'r') as fd: if typ == 'json': ret = self._read_metric_json(fd, xpath) elif typ == 'csv': ret = self._read_metric_xsv(fd, xpath, ',') elif typ == 'tsv': ret = self._read_metric_xsv(fd, xpath, '\t') elif typ == 'hcsv': ret = self._read_metric_hxsv(fd, xpath, ',') elif typ == 'htsv': ret = self._read_metric_hxsv(fd, xpath, '\t') else: ret = fd.read() except Exception as exc: self.logger.error('Unable to read metric in \'{}\''.format(path), exc) return ret def _find_output_by_path(self, path, outs=None): from dvc.exceptions import OutputDuplicationError if not outs: astages = self.active_stages() outs = [out for stage in astages for out in stage.outs] abs_path = os.path.abspath(path) matched = [out for out in outs if out.path == abs_path] stages = [out.stage.path for out in matched] if len(stages) > 1: raise OutputDuplicationError(path, stages) return matched[0] if matched else None def metrics_show(self, path=None, typ=None, xpath=None, all_branches=False): res = {} for branch in self.scm.brancher(all_branches=all_branches): astages = self.active_stages() outs = [out for stage in astages for out in stage.outs] if path: out = self._find_output_by_path(path, outs=outs) stage = out.stage.path if out else None if out and all([out.metric, not typ, isinstance(out.metric, dict)]): entries = [(path, out.metric.get(out.PARAM_METRIC_TYPE, None), out.metric.get(out.PARAM_METRIC_XPATH, None))] else: entries = [(path, typ, xpath)] else: metrics = filter(lambda o: o.metric, outs) stage = None entries = [] for o in metrics: if not typ and isinstance(o.metric, dict): t = o.metric.get(o.PARAM_METRIC_TYPE, None) x = o.metric.get(o.PARAM_METRIC_XPATH, None) else: t = typ x = xpath entries.append((o.path, t, x)) for fname, t, x in entries: if stage: self.checkout(stage) rel = os.path.relpath(fname) metric = self._read_metric(fname, typ=t, xpath=x) if not metric: continue if branch not in res: res[branch] = {} res[branch][rel] = metric for branch, val in res.items(): if all_branches: self.logger.info('{}:'.format(branch)) for fname, metric in val.items(): self.logger.info('\t{}: {}'.format(fname, metric)) if res: return res if path: msg = 'File \'{}\' does not exist'.format(path) else: msg = 'No metric files in this repository. ' \ 'Use \'dvc metrics add\' to add a metric file to track.' raise DvcException(msg) def _metrics_modify(self, path, typ=None, xpath=None, delete=False): out = self._find_output_by_path(path) if not out: msg = 'Unable to find file \'{}\' in the pipeline'.format(path) raise DvcException(msg) if out.path_info['scheme'] != 'local': msg = 'Output \'{}\' scheme \'{}\' is not supported for metrics' raise DvcException(msg.format(out.path, out.path_info['scheme'])) if out.use_cache: msg = 'Cached output \'{}\' is not supported for metrics' raise DvcException(msg.format(out.rel_path)) if typ: if not isinstance(out.metric, dict): out.metric = {} out.metric[out.PARAM_METRIC_TYPE] = typ if xpath: if not isinstance(out.metric, dict): out.metric = {} out.metric[out.PARAM_METRIC_XPATH] = xpath if delete: out.metric = None out._verify_metric() out.stage.dump() def metrics_modify(self, path=None, typ=None, xpath=None): self._metrics_modify(path, typ, xpath) def metrics_add(self, path, typ=None, xpath=None): if not typ: typ = 'raw' self._metrics_modify(path, typ, xpath) def metrics_remove(self, path): self._metrics_modify(path, delete=True) def graph(self): import networkx as nx G = nx.DiGraph() G_active = nx.DiGraph() stages = self.stages() outs = [] for stage in stages: outs += stage.outs # collect the whole DAG for stage in stages: node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) G_active.add_node(node, stage=stage) for dep in stage.deps: for out in outs: if out.path != dep.path \ and not dep.path.startswith(out.path + out.sep): continue dep_stage = out.stage dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) if not stage.locked: G_active.add_node(dep_node, stage=dep_stage) G_active.add_edge(node, dep_node) return G, G_active def pipelines(self): import networkx as nx G, G_active = self.graph() if len(G.nodes()) == 0: return [] # find pipeline ends aka "output stages" ends = [node for node, in_degree in G.in_degree() if in_degree == 0] # filter out subgraphs that didn't exist in original G pipelines = [] for c in nx.weakly_connected_components(G_active): H = G_active.subgraph(c) found = False for node in ends: if node in H: found = True break if found: pipelines.append(H) return pipelines def stages(self): stages = [] for root, dirs, files in os.walk(self.root_dir): for fname in files: path = os.path.join(root, fname) if not Stage.is_stage_file(path): continue stages.append(Stage.load(self, path)) return stages def active_stages(self): import networkx as nx stages = [] for G in self.pipelines(): stages.extend(list(nx.get_node_attributes(G, 'stage').values())) return stages
def setUp(self): super(TestGitSubmoduleTree, self).setUp() self.scm = SCM(self._root_dir) self.tree = GitTree(self.git, "master") self._pushd(self._root_dir)
def test_git(self): Repo.init(os.curdir) self.assertIsInstance(SCM(self._root_dir), Git)
class Repo(object): DVC_DIR = ".dvc" from dvc.repo.destroy import destroy from dvc.repo.install import install from dvc.repo.add import add from dvc.repo.remove import remove from dvc.repo.ls import ls from dvc.repo.lock import lock as lock_stage from dvc.repo.move import move from dvc.repo.run import run from dvc.repo.imp import imp from dvc.repo.imp_url import imp_url from dvc.repo.reproduce import reproduce from dvc.repo.checkout import _checkout from dvc.repo.push import push from dvc.repo.fetch import _fetch from dvc.repo.pull import pull from dvc.repo.status import status from dvc.repo.gc import gc from dvc.repo.commit import commit from dvc.repo.diff import diff from dvc.repo.brancher import brancher from dvc.repo.get import get from dvc.repo.get_url import get_url from dvc.repo.update import update def __init__(self, root_dir=None): from dvc.state import State from dvc.lock import make_lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.repo.metrics import Metrics from dvc.scm.tree import WorkingTree from dvc.repo.tag import Tag from dvc.utils.fs import makedirs root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) no_scm = self.config["core"].get("no_scm", False) self.scm = SCM(self.root_dir, no_scm=no_scm) self.tree = WorkingTree(self.root_dir) self.tmp_dir = os.path.join(self.dvc_dir, "tmp") makedirs(self.tmp_dir, exist_ok=True) hardlink_lock = self.config["core"].get("hardlink_lock", False) self.lock = make_lock( os.path.join(self.dvc_dir, "lock"), tmp_dir=os.path.join(self.dvc_dir, "tmp"), hardlink_lock=hardlink_lock, friendly=True, ) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self) self.cache = Cache(self) self.cloud = DataCloud(self) self.metrics = Metrics(self) self.tag = Tag(self) self._ignore() @property def tree(self): return self._tree @tree.setter def tree(self, tree): self._tree = tree if isinstance(tree, CleanTree) else CleanTree(tree) # Our graph cache is no longer valid, as it was based on the previous # tree. self._reset() def __repr__(self): return "{}: '{}'".format(self.__class__.__name__, self.root_dir) @classmethod def find_root(cls, root=None): root_dir = os.path.realpath(root or os.curdir) if not os.path.isdir(root_dir): raise NotDvcRepoError("directory '{}' does not exist".format(root)) while True: dvc_dir = os.path.join(root_dir, cls.DVC_DIR) if os.path.isdir(dvc_dir): return root_dir if os.path.ismount(root_dir): break root_dir = os.path.dirname(root_dir) message = ("you are not inside of a DVC repository " "(checked up to mount point '{}')").format(root_dir) raise NotDvcRepoError(message) @classmethod def find_dvc_dir(cls, root=None): root_dir = cls.find_root(root) return os.path.join(root_dir, cls.DVC_DIR) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False, subdir=False): from dvc.repo.init import init init(root_dir=root_dir, no_scm=no_scm, force=force, subdir=subdir) return Repo(root_dir) def unprotect(self, target): return self.cache.local.unprotect(PathInfo(target)) def _ignore(self): from dvc.updater import Updater updater = Updater(self.dvc_dir) flist = ([self.config.files["local"], updater.updater_file] + [self.lock.lockfile, updater.lock.lockfile, self.tmp_dir] + self.state.files) if path_isin(self.cache.local.cache_dir, self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def check_modified_graph(self, new_stages): """Generate graph including the new stage to check for errors""" self._collect_graph(self.stages + new_stages) def collect(self, target, with_deps=False, recursive=False, graph=None): import networkx as nx from dvc.stage import Stage G = graph or self.graph if not target: return list(G) target = os.path.abspath(target) if recursive and os.path.isdir(target): stages = nx.dfs_postorder_nodes(G) return [stage for stage in stages if path_isin(stage.path, target)] stage = Stage.load(self, target) if not with_deps: return [stage] pipeline = get_pipeline(get_pipelines(G), stage) return list(nx.dfs_postorder_nodes(pipeline, stage)) def collect_granular(self, target, *args, **kwargs): if not target: return [(stage, None) for stage in self.stages] try: (out, ) = self.find_outs_by_path(target, strict=False) filter_info = PathInfo(os.path.abspath(target)) return [(out.stage, filter_info)] except OutputNotFoundError: stages = self.collect(target, *args, **kwargs) return [(stage, None) for stage in stages] def used_cache( self, targets=None, all_branches=False, with_deps=False, all_tags=False, all_commits=False, remote=None, force=False, jobs=None, recursive=False, ): """Get the stages related to the given target and collect the `info` of its outputs. This is useful to know what files from the cache are _in use_ (namely, a file described as an output on a stage). The scope is, by default, the working directory, but you can use `all_branches`/`all_tags`/`all_commits` to expand the scope. Returns: A dictionary with Schemes (representing output's location) as keys, and a list with the outputs' `dumpd` as values. """ from dvc.cache import NamedCache cache = NamedCache() for branch in self.brancher( all_branches=all_branches, all_tags=all_tags, all_commits=all_commits, ): targets = targets or [None] pairs = cat( self.collect_granular( target, recursive=recursive, with_deps=with_deps) for target in targets) suffix = "({})".format(branch) if branch else "" for stage, filter_info in pairs: used_cache = stage.get_used_cache( remote=remote, force=force, jobs=jobs, filter_info=filter_info, ) cache.update(used_cache, suffix=suffix) return cache def _collect_graph(self, stages=None): """Generate a graph by using the given stages on the given directory The nodes of the graph are the stage's path relative to the root. Edges are created when the output of one stage is used as a dependency in other stage. The direction of the edges goes from the stage to its dependency: For example, running the following: $ dvc run -o A "echo A > A" $ dvc run -d A -o B "echo B > B" $ dvc run -d B -o C "echo C > C" Will create the following graph: ancestors <-- | C.dvc -> B.dvc -> A.dvc | | | --> descendants | ------- pipeline ------> | v (weakly connected components) Args: stages (list): used to build a graph, if None given, collect stages in the repository. Raises: OutputDuplicationError: two outputs with the same path StagePathAsOutputError: stage inside an output directory OverlappingOutputPathsError: output inside output directory CyclicGraphError: resulting graph has cycles """ import networkx as nx from pygtrie import Trie from dvc.exceptions import ( OutputDuplicationError, StagePathAsOutputError, OverlappingOutputPathsError, ) G = nx.DiGraph() stages = stages or self.stages stages = [stage for stage in stages if stage] outs = Trie() # Use trie to efficiently find overlapping outs and deps for stage in stages: for out in stage.outs: out_key = out.path_info.parts # Check for dup outs if out_key in outs: dup_stages = [stage, outs[out_key].stage] raise OutputDuplicationError(str(out), dup_stages) # Check for overlapping outs if outs.has_subtrie(out_key): parent = out overlapping = first(outs.values(prefix=out_key)) else: parent = outs.shortest_prefix(out_key).value overlapping = out if parent and overlapping: msg = ("Paths for outs:\n'{}'('{}')\n'{}'('{}')\n" "overlap. To avoid unpredictable behaviour, " "rerun command with non overlapping outs paths." ).format( str(parent), parent.stage.relpath, str(overlapping), overlapping.stage.relpath, ) raise OverlappingOutputPathsError(parent, overlapping, msg) outs[out_key] = out for stage in stages: out = outs.shortest_prefix(PathInfo(stage.path).parts).value if out: raise StagePathAsOutputError(stage, str(out)) # Building graph G.add_nodes_from(stages) for stage in stages: for dep in stage.deps: if dep.path_info is None: continue dep_key = dep.path_info.parts overlapping = list(n.value for n in outs.prefixes(dep_key)) if outs.has_subtrie(dep_key): overlapping.extend(outs.values(prefix=dep_key)) G.add_edges_from((stage, out.stage) for out in overlapping) check_acyclic(G) return G @cached_property def graph(self): return self._collect_graph() @cached_property def pipelines(self): return get_pipelines(self.graph) @cached_property def stages(self): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories tracked by DVC (e.g. `dvc add data` would skip `data/`) NOTE: For large repos, this could be an expensive operation. Consider using some memoization. """ from dvc.stage import Stage stages = [] outs = set() for root, dirs, files in self.tree.walk(self.root_dir): for fname in files: path = os.path.join(root, fname) if not Stage.is_valid_filename(path): continue stage = Stage.load(self, path) stages.append(stage) for out in stage.outs: if out.scheme == "local": outs.add(out.fspath) dirs[:] = [d for d in dirs if os.path.join(root, d) not in outs] return stages def find_outs_by_path(self, path, outs=None, recursive=False, strict=True): if not outs: outs = [out for stage in self.stages for out in stage.outs] abs_path = os.path.abspath(path) path_info = PathInfo(abs_path) match = path_info.__eq__ if strict else path_info.isin_or_eq def func(out): if out.scheme == "local" and match(out.path_info): return True if recursive and out.path_info.isin(path_info): return True return False matched = list(filter(func, outs)) if not matched: raise OutputNotFoundError(path, self) return matched def find_out_by_relpath(self, relpath): path = os.path.join(self.root_dir, relpath) (out, ) = self.find_outs_by_path(path) return out def is_dvc_internal(self, path): path_parts = os.path.normpath(path).split(os.path.sep) return self.DVC_DIR in path_parts @contextmanager def open_by_relpath(self, path, remote=None, mode="r", encoding=None): """Opens a specified resource as a file descriptor""" cause = None try: out = self.find_out_by_relpath(path) except OutputNotFoundError as exc: out = None cause = exc if out and out.use_cache: try: with self._open_cached(out, remote, mode, encoding) as fd: yield fd return except FileNotFoundError as exc: raise FileMissingError(path) from exc abs_path = os.path.join(self.root_dir, path) if os.path.exists(abs_path): with open(abs_path, mode=mode, encoding=encoding) as fd: yield fd return raise FileMissingError(path) from cause def _open_cached(self, out, remote=None, mode="r", encoding=None): if out.isdir(): raise ValueError("Can't open a dir") cache_file = self.cache.local.checksum_to_path_info(out.checksum) cache_file = fspath_py35(cache_file) if os.path.exists(cache_file): return open(cache_file, mode=mode, encoding=encoding) try: remote_obj = self.cloud.get_remote(remote) remote_info = remote_obj.checksum_to_path_info(out.checksum) return remote_obj.open(remote_info, mode=mode, encoding=encoding) except RemoteActionNotImplemented: with self.state: cache_info = out.get_used_cache(remote=remote) self.cloud.pull(cache_info, remote=remote) return open(cache_file, mode=mode, encoding=encoding) def close(self): self.scm.close() @locked def checkout(self, *args, **kwargs): return self._checkout(*args, **kwargs) @locked def fetch(self, *args, **kwargs): return self._fetch(*args, **kwargs) def _reset(self): self.__dict__.pop("graph", None) self.__dict__.pop("stages", None) self.__dict__.pop("pipelines", None) self.__dict__.pop("dvcignore", None)
def setUp(self): super().setUp() self.scm = SCM(self._root_dir) self._pushd(self._root_dir)
def test_none(self): self.assertIsInstance(SCM(self._root_dir), NoSCM)
def __init__(self, root_dir=None, scm=None, rev=None): from dvc.state import State, StateNoop from dvc.lock import make_lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.repo.metrics import Metrics from dvc.repo.plots import Plots from dvc.repo.params import Params from dvc.tree.local import LocalRemoteTree from dvc.utils.fs import makedirs from dvc.stage.cache import StageCache if scm: tree = scm.get_tree(rev) self.root_dir = self.find_root(root_dir, tree) self.scm = scm self.tree = scm.get_tree( rev, use_dvcignore=True, dvcignore_root=self.root_dir ) self.state = StateNoop() else: root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.tree = LocalRemoteTree( self, {"url": self.root_dir}, use_dvcignore=True, dvcignore_root=self.root_dir, ) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir, tree=self.tree) if not scm: no_scm = self.config["core"].get("no_scm", False) self.scm = SCM(self.root_dir, no_scm=no_scm) self.tmp_dir = os.path.join(self.dvc_dir, "tmp") self.index_dir = os.path.join(self.tmp_dir, "index") makedirs(self.index_dir, exist_ok=True) hardlink_lock = self.config["core"].get("hardlink_lock", False) self.lock = make_lock( os.path.join(self.tmp_dir, "lock"), tmp_dir=self.tmp_dir, hardlink_lock=hardlink_lock, friendly=True, ) self.cache = Cache(self) self.cloud = DataCloud(self) if not scm: # NOTE: storing state and link_state in the repository itself to # avoid any possible state corruption in 'shared cache dir' # scenario. self.state = State(self.cache.local) self.stage_cache = StageCache(self) self.metrics = Metrics(self) self.plots = Plots(self) self.params = Params(self) self._ignore()
class Project(object): DVC_DIR = '.dvc' def __init__(self, root_dir=None): from dvc.logger import logger from dvc.config import Config from dvc.state import State from dvc.lock import Lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.updater import Updater from dvc.prompt import Prompt root_dir = self._find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir, project=self) self.lock = Lock(self.dvc_dir) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self, self.config._config) core = self.config._config[Config.SECTION_CORE] self.logger = logger self.logger.set_level(core.get(Config.SECTION_CORE_LOGLEVEL, None)) self.cache = Cache(self) self.cloud = DataCloud(self, config=self.config._config) self.updater = Updater(self.dvc_dir) self.prompt = Prompt() self._files_to_git_add = [] self._ignore() self.updater.check() @staticmethod def _find_root(root=None): if root is None: root = os.getcwd() else: root = os.path.abspath(os.path.realpath(root)) while True: dvc_dir = os.path.join(root, Project.DVC_DIR) if os.path.isdir(dvc_dir): return root if os.path.ismount(root): break root = os.path.dirname(root) raise NotDvcProjectError(root) @staticmethod def _find_dvc_dir(root=None): root_dir = Project._find_root(root) return os.path.join(root_dir, Project.DVC_DIR) def _remind_to_git_add(self): if len(self._files_to_git_add) == 0: return msg = '\nTo track the changes with git run:\n\n' msg += '\tgit add ' + " ".join(self._files_to_git_add) self.logger.info(msg) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False): """ Initiate dvc project in directory. Args: root_dir: Path to project's root directory. Returns: Project instance. Raises: KeyError: Raises an exception. """ import colorama import shutil from dvc.scm import SCM, Base from dvc.config import Config from dvc.logger import logger root_dir = os.path.abspath(root_dir) dvc_dir = os.path.join(root_dir, Project.DVC_DIR) scm = SCM(root_dir) if type(scm) == Base and not no_scm: msg = "{} is not tracked by any supported scm tool(e.g. git)." raise InitError(msg.format(root_dir)) if os.path.isdir(dvc_dir): if not force: msg = "'{}' exists. Use '-f' to force." raise InitError(msg.format(os.path.relpath(dvc_dir))) shutil.rmtree(dvc_dir) os.mkdir(dvc_dir) config = Config.init(dvc_dir) proj = Project(root_dir) scm.add([config.config_file]) if scm.ignore_file(): scm.add([os.path.join(dvc_dir, scm.ignore_file())]) logger.info('\nYou can now commit the changes to git.') logger.info( "\n" "{yellow}What's next?{nc}\n" "{yellow}------------{nc}\n" "- Check out the documentation: {blue}https://dvc.org/doc{nc}\n" "- Get help and share ideas: {blue}https://dvc.org/chat{nc}\n" "- Star us on GitHub: {blue}https://github.com/iterative/dvc{nc}" .format(yellow=colorama.Fore.YELLOW, blue=colorama.Fore.BLUE, green=colorama.Fore.GREEN, nc=colorama.Fore.RESET) ) return proj @staticmethod def load_all(projects_paths): """ Instantiate all projects in the given list of paths. Args: projects_paths: List of paths to projects. Returns: List of Project instances in the same order of the given paths. """ return [Project(path) for path in projects_paths] def destroy(self): import shutil for stage in self.stages(): stage.remove() shutil.rmtree(self.dvc_dir) def _ignore(self): flist = [ self.state.state_file, self.lock.lock_file, self.config.config_local_file, self.updater.updater_file, self.updater.lock.lock_file, ] + self.state.temp_files if self.cache.local.cache_dir.startswith(self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def install(self): self.scm.install() def to_dvc_path(self, path): return os.path.relpath(path, self.root_dir) def _check_cwd_specified_as_output(self, cwd, stages): from dvc.exceptions import WorkingDirectoryAsOutputError cwd_path = os.path.abspath(os.path.normpath(cwd)) for stage in stages: for output in stage.outs: if os.path.isdir(output.path) and output.path == cwd_path: raise WorkingDirectoryAsOutputError(cwd, stage.relpath) def _check_output_duplication(self, outs, stages): from dvc.exceptions import OutputDuplicationError for stage in stages: for o in stage.outs: for out in outs: if o.path == out.path and o.stage.path != out.stage.path: stages = [o.stage.relpath, out.stage.relpath] raise OutputDuplicationError(o.path, stages) def _check_circular_dependency(self, deps, outs): from dvc.exceptions import CircularDependencyError circular_dependencies = ( set(file.path for file in deps) & set(file.path for file in outs) ) if circular_dependencies: raise CircularDependencyError(circular_dependencies.pop()) def add(self, fname, recursive=False): from dvc.stage import Stage fnames = [] if recursive and os.path.isdir(fname): fnames = [] for root, dirs, files in os.walk(fname): for f in files: path = os.path.join(root, f) if Stage.is_stage_file(path): continue if os.path.basename(path) == self.scm.ignore_file(): continue if self.scm.is_tracked(path): continue fnames.append(path) else: fnames = [fname] all_stages = self.stages() stages = [] self._files_to_git_add = [] with self.state: for f in fnames: stage = Stage.loads(project=self, outs=[f], add=True) if stage is None: stages.append(stage) continue self._check_output_duplication(stage.outs, all_stages) stage.save() stage.dump() stages.append(stage) self._remind_to_git_add() return stages def remove(self, target, outs_only=False): from dvc.stage import Stage stage = Stage.load(self, target) if outs_only: stage.remove_outs() else: stage.remove() return stage def lock_stage(self, target, unlock=False): from dvc.stage import Stage stage = Stage.load(self, target) stage.locked = False if unlock else True stage.dump() return stage def move(self, from_path, to_path): import dvc.output as Output from dvc.stage import Stage from_out = Output.loads_from(Stage(self, cwd=os.curdir), [from_path])[0] found = False self._files_to_git_add = [] with self.state: for stage in self.stages(): for out in stage.outs: if out.path != from_out.path: continue if not stage.is_data_source: raise MoveNotDataSourceError(stage.relpath) found = True to_out = Output.loads_from(out.stage, [to_path], out.cache, out.metric)[0] out.move(to_out) stage_base = os.path.basename(stage.path) stage_base = stage_base.rstrip(Stage.STAGE_FILE_SUFFIX) stage_dir = os.path.dirname(stage.path) from_base = os.path.basename(from_path) to_base = os.path.basename(to_path) if stage_base == from_base: os.unlink(stage.path) path = to_base + Stage.STAGE_FILE_SUFFIX stage.path = os.path.join(stage_dir, path) stage.dump() self._remind_to_git_add() if not found: msg = 'Unable to find dvcfile with output \'{}\'' raise DvcException(msg.format(from_path)) def _unprotect_file(self, path): import stat import uuid from dvc.system import System from dvc.utils import copyfile, move, remove if System.is_symlink(path) or System.is_hardlink(path): self.logger.debug("Unprotecting '{}'".format(path)) tmp = os.path.join(os.path.dirname(path), '.' + str(uuid.uuid4())) move(path, tmp) copyfile(tmp, path) remove(tmp) else: self.logger.debug("Skipping copying for '{}', since it is not " "a symlink or a hardlink.".format(path)) os.chmod(path, os.stat(path).st_mode | stat.S_IWRITE) def _unprotect_dir(self, path): for root, dirs, files in os.walk(path): for f in files: path = os.path.join(root, f) self._unprotect_file(path) def unprotect(self, path): if not os.path.exists(path): raise DvcException("Can't unprotect non-existing " "data '{}'".format(path)) if os.path.isdir(path): self._unprotect_dir(path) else: self._unprotect_file(path) def run(self, cmd=None, deps=[], outs=[], outs_no_cache=[], metrics_no_cache=[], fname=None, cwd=os.curdir, no_exec=False, overwrite=False, ignore_build_cache=False, remove_outs=False): from dvc.stage import Stage with self.state: stage = Stage.loads(project=self, fname=fname, cmd=cmd, cwd=cwd, outs=outs, outs_no_cache=outs_no_cache, metrics_no_cache=metrics_no_cache, deps=deps, overwrite=overwrite, ignore_build_cache=ignore_build_cache, remove_outs=remove_outs) if stage is None: return None all_stages = self.stages() self._check_cwd_specified_as_output(cwd, all_stages) self._check_output_duplication(stage.outs, all_stages) self._check_circular_dependency(stage.deps, stage.outs) self._files_to_git_add = [] with self.state: if not no_exec: stage.run() stage.dump() self._remind_to_git_add() return stage def imp(self, url, out): from dvc.stage import Stage stage = Stage.loads(project=self, cmd=None, deps=[url], outs=[out]) if stage is None: return None self._check_output_duplication(stage.outs, self.stages()) self._check_circular_dependency(stage.deps, stage.outs) self._files_to_git_add = [] with self.state: stage.run() stage.dump() self._remind_to_git_add() return stage def _reproduce_stage(self, stages, node, force, dry, interactive): stage = stages[node] if stage.locked: msg = 'DVC file \'{}\' is locked. Its dependencies are not ' \ 'going to be reproduced.' self.logger.warn(msg.format(stage.relpath)) stage = stage.reproduce(force=force, dry=dry, interactive=interactive) if not stage: return [] if not dry: stage.dump() return [stage] def reproduce(self, target=None, recursive=True, force=False, dry=False, interactive=False, pipeline=False, all_pipelines=False): from dvc.stage import Stage if target is None and not all_pipelines: raise ValueError() if not interactive: config = self.config core = config._config[config.SECTION_CORE] interactive = core.get(config.SECTION_CORE_INTERACTIVE, False) targets = [] if pipeline or all_pipelines: if pipeline: stage = Stage.load(self, target) node = os.path.relpath(stage.path, self.root_dir) pipelines = [self._get_pipeline(node)] else: pipelines = self.pipelines() for G in pipelines: for node in G.nodes(): if G.in_degree(node) == 0: targets.append(os.path.join(self.root_dir, node)) else: targets.append(target) self._files_to_git_add = [] ret = [] with self.state: for target in targets: stages = self._reproduce(target, recursive=recursive, force=force, dry=dry, interactive=interactive) ret.extend(stages) self._remind_to_git_add() return ret def _reproduce(self, target, recursive=True, force=False, dry=False, interactive=False): import networkx as nx from dvc.stage import Stage stage = Stage.load(self, target) G = self.graph()[1] stages = nx.get_node_attributes(G, 'stage') node = os.path.relpath(stage.path, self.root_dir) if recursive: ret = self._reproduce_stages(G, stages, node, force, dry, interactive) else: ret = self._reproduce_stage(stages, node, force, dry, interactive) return ret def _reproduce_stages(self, G, stages, node, force, dry, interactive): import networkx as nx result = [] for n in nx.dfs_postorder_nodes(G, node): try: result += self._reproduce_stage(stages, n, force, dry, interactive) except Exception as ex: raise ReproductionError(stages[n].relpath, ex) return result def _cleanup_unused_links(self, all_stages): used = [] for stage in all_stages: for out in stage.outs: used.append(out.path) self.state.remove_unused_links(used) def checkout(self, target=None, with_deps=False, force=False): all_stages = self.active_stages() stages = all_stages if target: stages = self._collect(target, with_deps=with_deps) with self.state: self._cleanup_unused_links(all_stages) for stage in stages: if stage.locked: msg = 'DVC file \'{}\' is locked. Its dependencies are ' \ 'not going to be checked out.' self.logger.warn(msg.format(stage.relpath)) stage.checkout(force=force) def _get_pipeline(self, node): pipelines = list(filter(lambda g: node in g.nodes(), self.pipelines())) assert len(pipelines) == 1 return pipelines[0] def _collect(self, target, with_deps=False): import networkx as nx from dvc.stage import Stage stage = Stage.load(self, target) if not with_deps: return [stage] node = os.path.relpath(stage.path, self.root_dir) G = self._get_pipeline(node) stages = nx.get_node_attributes(G, 'stage') ret = [stage] for n in nx.dfs_postorder_nodes(G, node): ret.append(stages[n]) return ret def _collect_dir_cache(self, out, branch=None, remote=None, force=False, jobs=None): info = out.dumpd() ret = [info] r = out.remote md5 = info[r.PARAM_MD5] if self.cache.local.changed_cache_file(md5): try: self.cloud.pull(ret, jobs=jobs, remote=remote, show_checksums=False) except DvcException as exc: msg = "Failed to pull cache for '{}': {}" self.logger.debug(msg.format(out, exc)) if self.cache.local.changed_cache_file(md5): msg = "Missing cache for directory '{}'. " \ "Cache for files inside will be lost. " \ "Would you like to continue? Use '-f' to force. " if not (force or self.prompt.prompt(msg, False)): raise DvcException("Unable to fully collect " "used cache without cache " "for directory '{}'".format(out)) else: return ret for i in self.cache.local.load_dir_cache(md5): i['branch'] = branch i[r.PARAM_PATH] = os.path.join(info[r.PARAM_PATH], i[r.PARAM_RELPATH]) ret.append(i) return ret def _collect_used_cache(self, out, branch=None, remote=None, force=False, jobs=None): if not out.use_cache or not out.info: return [] info = out.dumpd() info['branch'] = branch ret = [info] if out.path_info['scheme'] != 'local': return ret md5 = info[out.remote.PARAM_MD5] cache = self.cache.local.get(md5) if not out.remote.is_dir_cache(cache): return ret return self._collect_dir_cache(out, branch=branch, remote=remote, force=force, jobs=jobs) def _used_cache(self, target=None, all_branches=False, active=True, with_deps=False, all_tags=False, remote=None, force=False, jobs=None): cache = {} cache['local'] = [] cache['s3'] = [] cache['gs'] = [] cache['hdfs'] = [] cache['ssh'] = [] cache['azure'] = [] for branch in self.scm.brancher(all_branches=all_branches, all_tags=all_tags): if target: stages = self._collect(target, with_deps=with_deps) elif active: stages = self.active_stages() else: stages = self.stages() for stage in stages: if active and not target and stage.locked: msg = 'DVC file \'{}\' is locked. Its dependencies are ' \ 'not going to be pushed/pulled/fetched.' self.logger.warn(msg.format(stage.relpath)) for out in stage.outs: scheme = out.path_info['scheme'] cache[scheme] += self._collect_used_cache(out, branch=branch, remote=remote, force=force, jobs=jobs) return cache @staticmethod def merge_cache_lists(clists): merged_cache = collections.defaultdict(list) for cache_list in clists: for scheme, cache in cache_list.items(): for item in cache: if item not in merged_cache[scheme]: merged_cache[scheme].append(item) return merged_cache @staticmethod def load_all_used_cache(projects, target=None, all_branches=False, active=True, with_deps=False, all_tags=False, remote=None, force=False, jobs=None): clists = [] for project in projects: with project.state: project_clist = project._used_cache(target=None, all_branches=all_branches, active=False, with_deps=with_deps, all_tags=all_tags, remote=remote, force=force, jobs=jobs) clists.append(project_clist) return clists def _do_gc(self, typ, func, clist): removed = func(clist) if not removed: self.logger.info("No unused {} cache to remove.".format(typ)) def gc(self, all_branches=False, cloud=False, remote=None, with_deps=False, all_tags=False, force=False, jobs=None, projects=None): all_projects = [self] if projects is not None and len(projects) > 0: all_projects.extend(Project.load_all(projects)) all_clists = Project.load_all_used_cache(all_projects, target=None, all_branches=all_branches, active=False, with_deps=with_deps, all_tags=all_tags, remote=remote, force=force, jobs=jobs) if len(all_clists) > 1: clist = Project.merge_cache_lists(all_clists) else: clist = all_clists[0] with self.state: self._do_gc('local', self.cache.local.gc, clist) if self.cache.s3: self._do_gc('s3', self.cache.s3.gc, clist) if self.cache.gs: self._do_gc('gs', self.cache.gs.gc, clist) if self.cache.ssh: self._do_gc('ssh', self.cache.ssh.gc, clist) if self.cache.hdfs: self._do_gc('hdfs', self.cache.hdfs.gc, clist) if self.cache.azure: self._do_gc('azure', self.cache.azure.gc, clist) if cloud: self._do_gc('remote', self.cloud._get_cloud(remote, 'gc -c').gc, clist) def push(self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False, with_deps=False, all_tags=False): with self.state: used = self._used_cache(target, all_branches=all_branches, all_tags=all_tags, with_deps=with_deps, force=True, remote=remote, jobs=jobs)['local'] self.cloud.push(used, jobs, remote=remote, show_checksums=show_checksums) def fetch(self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False, with_deps=False, all_tags=False): with self.state: used = self._used_cache(target, all_branches=all_branches, all_tags=all_tags, with_deps=with_deps, force=True, remote=remote, jobs=jobs)['local'] self.cloud.pull(used, jobs, remote=remote, show_checksums=show_checksums) def pull(self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False, with_deps=False, all_tags=False, force=False): self.fetch(target, jobs, remote=remote, all_branches=all_branches, all_tags=all_tags, show_checksums=show_checksums, with_deps=with_deps) self.checkout(target=target, with_deps=with_deps, force=force) def _local_status(self, target=None, with_deps=False): status = {} if target: stages = self._collect(target, with_deps=with_deps) else: stages = self.active_stages() for stage in stages: if stage.locked: msg = 'DVC file \'{}\' is locked. Its dependencies are not ' \ 'going to be shown in the status output.' self.logger.warn(msg.format(stage.relpath)) status.update(stage.status()) return status def _cloud_status(self, target=None, jobs=1, remote=None, show_checksums=False, all_branches=False, with_deps=False, all_tags=False): import dvc.remote.base as cloud used = self._used_cache(target, all_branches=all_branches, all_tags=all_tags, with_deps=with_deps, force=True, remote=remote, jobs=jobs)['local'] status = {} for md5, ret in self.cloud.status(used, jobs, remote=remote, show_checksums=show_checksums): if ret == cloud.STATUS_OK: continue prefix_map = { cloud.STATUS_DELETED: 'deleted', cloud.STATUS_NEW: 'new', } status[md5] = prefix_map[ret] return status def status(self, target=None, jobs=1, cloud=False, remote=None, show_checksums=False, all_branches=False, with_deps=False, all_tags=False): with self.state: if cloud: return self._cloud_status(target, jobs, remote=remote, show_checksums=show_checksums, all_branches=all_branches, with_deps=with_deps, all_tags=all_tags) return self._local_status(target, with_deps=with_deps) def _read_metric_json(self, fd, json_path): import json from jsonpath_rw import parse parser = parse(json_path) return [x.value for x in parser.find(json.load(fd))] def _do_read_metric_xsv(self, reader, row, col): if col is not None and row is not None: return [reader[row][col]] elif col is not None: return [r[col] for r in reader] elif row is not None: return reader[row] return None def _read_metric_hxsv(self, fd, hxsv_path, delimiter): import csv col, row = hxsv_path.split(',') row = int(row) reader = list(csv.DictReader(fd, delimiter=delimiter)) return self._do_read_metric_xsv(reader, row, col) def _read_metric_xsv(self, fd, xsv_path, delimiter): import csv col, row = xsv_path.split(',') row = int(row) col = int(col) reader = list(csv.reader(fd, delimiter=delimiter)) return self._do_read_metric_xsv(reader, row, col) def _read_metric(self, path, typ=None, xpath=None): ret = None if not os.path.exists(path): return ret try: with open(path, 'r') as fd: if typ == 'json': ret = self._read_metric_json(fd, xpath) elif typ == 'csv': ret = self._read_metric_xsv(fd, xpath, ',') elif typ == 'tsv': ret = self._read_metric_xsv(fd, xpath, '\t') elif typ == 'hcsv': ret = self._read_metric_hxsv(fd, xpath, ',') elif typ == 'htsv': ret = self._read_metric_hxsv(fd, xpath, '\t') else: ret = fd.read() except Exception as exc: self.logger.error('Unable to read metric in \'{}\''.format(path), exc) return ret def _find_output_by_path(self, path, outs=None): from dvc.exceptions import OutputDuplicationError if not outs: astages = self.active_stages() outs = [out for stage in astages for out in stage.outs] abs_path = os.path.abspath(path) matched = [out for out in outs if out.path == abs_path] stages = [out.stage.relpath for out in matched] if len(stages) > 1: raise OutputDuplicationError(path, stages) return matched[0] if matched else None def metrics_show(self, path=None, typ=None, xpath=None, all_branches=False, all_tags=False): res = {} for branch in self.scm.brancher(all_branches=all_branches, all_tags=all_tags): astages = self.active_stages() outs = [out for stage in astages for out in stage.outs] if path: out = self._find_output_by_path(path, outs=outs) stage = out.stage.path if out else None if out and all([out.metric, not typ, isinstance(out.metric, dict)]): entries = [(path, out.metric.get(out.PARAM_METRIC_TYPE, None), out.metric.get(out.PARAM_METRIC_XPATH, None))] else: entries = [(path, typ, xpath)] else: metrics = filter(lambda o: o.metric, outs) stage = None entries = [] for o in metrics: if not typ and isinstance(o.metric, dict): t = o.metric.get(o.PARAM_METRIC_TYPE, typ) x = o.metric.get(o.PARAM_METRIC_XPATH, xpath) else: t = typ x = xpath entries.append((o.path, t, x)) for fname, t, x in entries: if stage: self.checkout(stage, force=True) rel = os.path.relpath(fname) metric = self._read_metric(fname, typ=t, xpath=x) if not metric: continue if branch not in res: res[branch] = {} res[branch][rel] = metric for branch, val in res.items(): if all_branches or all_tags: self.logger.info('{}:'.format(branch)) for fname, metric in val.items(): self.logger.info('\t{}: {}'.format(fname, metric)) if res: return res if path: msg = 'File \'{}\' does not exist'.format(path) else: msg = 'No metric files in this repository. ' \ 'Use \'dvc metrics add\' to add a metric file to track.' raise DvcException(msg) def _metrics_modify(self, path, typ=None, xpath=None, delete=False): out = self._find_output_by_path(path) if not out: msg = 'Unable to find file \'{}\' in the pipeline'.format(path) raise DvcException(msg) if out.path_info['scheme'] != 'local': msg = 'Output \'{}\' scheme \'{}\' is not supported for metrics' raise DvcException(msg.format(out.path, out.path_info['scheme'])) if out.use_cache: msg = 'Cached output \'{}\' is not supported for metrics' raise DvcException(msg.format(out.rel_path)) if typ: if not isinstance(out.metric, dict): out.metric = {} out.metric[out.PARAM_METRIC_TYPE] = typ if xpath: if not isinstance(out.metric, dict): out.metric = {} out.metric[out.PARAM_METRIC_XPATH] = xpath if delete: out.metric = None out._verify_metric() out.stage.dump() def metrics_modify(self, path=None, typ=None, xpath=None): self._metrics_modify(path, typ, xpath) def metrics_add(self, path, typ=None, xpath=None): if not typ: typ = 'raw' self._metrics_modify(path, typ, xpath) def metrics_remove(self, path): self._metrics_modify(path, delete=True) def graph(self): import networkx as nx from dvc.exceptions import OutputDuplicationError G = nx.DiGraph() G_active = nx.DiGraph() stages = self.stages() outs = [] outs_by_path = {} for stage in stages: for o in stage.outs: existing = outs_by_path.get(o.path, None) if existing is not None: stages = [o.stage.relpath, existing.stage.relpath] raise OutputDuplicationError(o.path, stages) outs.append(o) outs_by_path[o.path] = o # collect the whole DAG for stage in stages: node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) G_active.add_node(node, stage=stage) for dep in stage.deps: for out in outs: if out.path != dep.path \ and not dep.path.startswith(out.path + out.sep) \ and not out.path.startswith(dep.path + dep.sep): continue dep_stage = out.stage dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) if not stage.locked: G_active.add_node(dep_node, stage=dep_stage) G_active.add_edge(node, dep_node) return G, G_active def pipelines(self): import networkx as nx G, G_active = self.graph() if len(G.nodes()) == 0: return [] # find pipeline ends aka "output stages" ends = [node for node, in_degree in G.in_degree() if in_degree == 0] # filter out subgraphs that didn't exist in original G pipelines = [] for c in nx.weakly_connected_components(G_active): H = G_active.subgraph(c) found = False for node in ends: if node in H: found = True break if found: pipelines.append(H) return pipelines def stages(self): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`) or DVC itself (`.dvc`). NOTE: For large projects, this could be an expensive operation. Consider using some memorization. """ from dvc.stage import Stage stages = [] outs = [] for root, dirs, files in os.walk(self.root_dir): for fname in files: path = os.path.join(root, fname) if not Stage.is_stage_file(path): continue stage = Stage.load(self, path) for out in stage.outs: outs.append(out.path + out.sep) stages.append(stage) def filter_dirs(dname): path = os.path.join(root, dname) if path == self.dvc_dir or path == self.scm.dir: return False for out in outs: if path == os.path.normpath(out) or path.startswith(out): return False return True dirs[:] = list(filter(filter_dirs, dirs)) return stages def active_stages(self): import networkx as nx stages = [] for G in self.pipelines(): stages.extend(list(nx.get_node_attributes(G, 'stage').values())) return stages
class Repo(object): DVC_DIR = ".dvc" from dvc.repo.destroy import destroy from dvc.repo.install import install from dvc.repo.add import add from dvc.repo.remove import remove from dvc.repo.lock import lock as lock_stage from dvc.repo.move import move from dvc.repo.run import run from dvc.repo.imp import imp from dvc.repo.reproduce import reproduce from dvc.repo.checkout import checkout from dvc.repo.push import push from dvc.repo.fetch import fetch from dvc.repo.pull import pull from dvc.repo.status import status from dvc.repo.gc import gc from dvc.repo.commit import commit from dvc.repo.diff import diff from dvc.repo.brancher import brancher def __init__(self, root_dir=None): from dvc.config import Config from dvc.state import State from dvc.lock import Lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.updater import Updater from dvc.repo.metrics import Metrics from dvc.scm.tree import WorkingTree from dvc.repo.tag import Tag from dvc.repo.pkg import Pkg root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.tree = WorkingTree(self.root_dir) self.scm = SCM(self.root_dir, repo=self) self.lock = Lock(self.dvc_dir) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self, self.config.config) core = self.config.config[Config.SECTION_CORE] level = core.get(Config.SECTION_CORE_LOGLEVEL) if level: logger.setLevel(level.upper()) self.cache = Cache(self) self.cloud = DataCloud(self, config=self.config.config) self.updater = Updater(self.dvc_dir) self.metrics = Metrics(self) self.tag = Tag(self) self.pkg = Pkg(self) self._ignore() self.updater.check() def __repr__(self): return "Repo: '{root_dir}'".format(root_dir=self.root_dir) @staticmethod def find_root(root=None): if root is None: root = os.getcwd() else: root = os.path.abspath(os.path.realpath(root)) while True: dvc_dir = os.path.join(root, Repo.DVC_DIR) if os.path.isdir(dvc_dir): return root if os.path.ismount(root): break root = os.path.dirname(root) raise NotDvcRepoError(root) @staticmethod def find_dvc_dir(root=None): root_dir = Repo.find_root(root) return os.path.join(root_dir, Repo.DVC_DIR) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False): from dvc.repo.init import init init(root_dir=root_dir, no_scm=no_scm, force=force) return Repo(root_dir) def unprotect(self, target): path_info = {"schema": "local", "path": target} return self.cache.local.unprotect(path_info) def _ignore(self): flist = [ self.state.state_file, self.lock.lock_file, self.config.config_local_file, self.updater.updater_file, self.updater.lock.lock_file, ] + self.state.temp_files if self.cache.local.cache_dir.startswith(self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def check_dag(self, stages): """Generate graph including the new stage to check for errors""" self.graph(stages=stages) @staticmethod def _check_cyclic_graph(graph): import networkx as nx from dvc.exceptions import CyclicGraphError cycles = list(nx.simple_cycles(graph)) if cycles: raise CyclicGraphError(cycles[0]) def _get_pipeline(self, node): pipelines = [i for i in self.pipelines() if i.has_node(node)] assert len(pipelines) == 1 return pipelines[0] def collect(self, target, with_deps=False, recursive=False): import networkx as nx from dvc.stage import Stage if not target or recursive: return self.active_stages(target) stage = Stage.load(self, target) if not with_deps: return [stage] node = os.path.relpath(stage.path, self.root_dir) G = self._get_pipeline(node) ret = [] for n in nx.dfs_postorder_nodes(G, node): ret.append(G.node[n]["stage"]) return ret def _collect_dir_cache(self, out, branch=None, remote=None, force=False, jobs=None): info = out.dumpd() ret = [info] r = out.remote md5 = info[r.PARAM_CHECKSUM] if self.cache.local.changed_cache_file(md5): try: self.cloud.pull(ret, jobs=jobs, remote=remote, show_checksums=False) except DvcException as exc: msg = "Failed to pull cache for '{}': {}" logger.debug(msg.format(out, exc)) if self.cache.local.changed_cache_file(md5): msg = ("Missing cache for directory '{}'. " "Cache for files inside will be lost. " "Would you like to continue? Use '-f' to force. ") if not force and not prompt.confirm(msg): raise DvcException( "unable to fully collect used cache" " without cache for directory '{}'".format(out)) else: return ret for i in out.dir_cache: i["branch"] = branch i[r.PARAM_PATH] = os.path.join(info[r.PARAM_PATH], i[r.PARAM_RELPATH]) ret.append(i) return ret def _collect_used_cache(self, out, branch=None, remote=None, force=False, jobs=None): if not out.use_cache or not out.info: if not out.info: logger.warning("Output '{}'({}) is missing version " "info. Cache for it will not be collected. " "Use dvc repro to get your pipeline up to " "date.".format(out, out.stage)) return [] info = out.dumpd() info["branch"] = branch ret = [info] if out.scheme != "local": return ret if not out.is_dir_checksum: return ret return self._collect_dir_cache(out, branch=branch, remote=remote, force=force, jobs=jobs) def used_cache( self, target=None, all_branches=False, active=True, with_deps=False, all_tags=False, remote=None, force=False, jobs=None, recursive=False, ): cache = {} cache["local"] = [] cache["s3"] = [] cache["gs"] = [] cache["hdfs"] = [] cache["ssh"] = [] cache["azure"] = [] for branch in self.brancher(all_branches=all_branches, all_tags=all_tags): if target: if recursive: stages = self.stages(target) else: stages = self.collect(target, with_deps=with_deps) elif active: stages = self.active_stages() else: stages = self.stages() for stage in stages: if active and not target and stage.locked: logger.warning( "DVC file '{path}' is locked. Its dependencies are" " not going to be pushed/pulled/fetched.".format( path=stage.relpath)) for out in stage.outs: scheme = out.path_info["scheme"] cache[scheme].extend( self._collect_used_cache( out, branch=branch, remote=remote, force=force, jobs=jobs, )) return cache def graph(self, stages=None, from_directory=None): """Generate a graph by using the given stages on the given directory The nodes of the graph are the stage's path relative to the root. Edges are created when the output of one stage is used as a dependency in other stage. The direction of the edges goes from the stage to its dependency: For example, running the following: $ dvc run -o A "echo A > A" $ dvc run -d A -o B "echo B > B" $ dvc run -d B -o C "echo C > C" Will create the following graph: ancestors <-- | C.dvc -> B.dvc -> A.dvc | | | --> descendants | ------- pipeline ------> | v (weakly connected components) Args: stages (list): used to build a graph, if None given, use the ones on the `from_directory`. from_directory (str): directory where to look at for stages, if None is given, use the current working directory Raises: OutputDuplicationError: two outputs with the same path StagePathAsOutputError: stage inside an output directory OverlappingOutputPathsError: output inside output directory CyclicGraphError: resulting graph has cycles """ import networkx as nx from dvc.exceptions import ( OutputDuplicationError, StagePathAsOutputError, OverlappingOutputPathsError, ) G = nx.DiGraph() G_active = nx.DiGraph() stages = stages or self.stages(from_directory, check_dag=False) stages = [stage for stage in stages if stage] outs = [] for stage in stages: for out in stage.outs: existing = [] for o in outs: if o.path == out.path: existing.append(o.stage) in_o_dir = out.path.startswith(o.path + o.sep) in_out_dir = o.path.startswith(out.path + out.sep) if in_o_dir or in_out_dir: raise OverlappingOutputPathsError(o, out) if existing: stages = [stage.relpath, existing[0].relpath] raise OutputDuplicationError(out.path, stages) outs.append(out) for stage in stages: path_dir = os.path.dirname(stage.path) + os.sep for out in outs: if path_dir.startswith(out.path + os.sep): raise StagePathAsOutputError(stage.wdir, stage.relpath) for stage in stages: node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) G_active.add_node(node, stage=stage) for dep in stage.deps: for out in outs: if (out.path != dep.path and not dep.path.startswith(out.path + out.sep) and not out.path.startswith(dep.path + dep.sep)): continue dep_stage = out.stage dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) if not stage.locked: G_active.add_node(dep_node, stage=dep_stage) G_active.add_edge(node, dep_node) self._check_cyclic_graph(G) return G, G_active def pipelines(self, from_directory=None): import networkx as nx G, G_active = self.graph(from_directory=from_directory) return [ G.subgraph(c).copy() for c in nx.weakly_connected_components(G) ] def stages(self, from_directory=None, check_dag=True): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories tracked by DVC (e.g. `dvc add data` would skip `data/`) NOTE: For large repos, this could be an expensive operation. Consider using some memoization. """ from dvc.stage import Stage if not from_directory: from_directory = self.root_dir elif not os.path.isdir(from_directory): raise TargetNotDirectoryError(from_directory) stages = [] outs = [] ignore_file_handler = DvcIgnoreFileHandler(self.tree) for root, dirs, files in self.tree.walk( from_directory, ignore_file_handler=ignore_file_handler): for fname in files: path = os.path.join(root, fname) if not Stage.is_valid_filename(path): continue stage = Stage.load(self, path) for out in stage.outs: outs.append(out.path + out.sep) stages.append(stage) def filter_dirs(dname, root=root): path = os.path.join(root, dname) if path in (self.dvc_dir, self.scm.dir): return False for out in outs: if path == os.path.normpath(out) or path.startswith(out): return False return True dirs[:] = list(filter(filter_dirs, dirs)) if check_dag: self.check_dag(stages) return stages def active_stages(self, from_directory=None): import networkx as nx stages = [] for G in self.pipelines(from_directory): stages.extend(list(nx.get_node_attributes(G, "stage").values())) return stages def find_outs_by_path(self, path, outs=None, recursive=False): if not outs: # there is no `from_directory=path` argument because some data # files might be generated to an upper level, and so it is # needed to look at all the files (project root_dir) stages = self.stages() outs = [out for stage in stages for out in stage.outs] abs_path = os.path.abspath(path) is_dir = self.tree.isdir(abs_path) def func(out): if out.path == abs_path: return True if is_dir and recursive and out.path.startswith(abs_path + os.sep): return True return False matched = list(filter(func, outs)) if not matched: raise OutputNotFoundError(path) return matched def is_dvc_internal(self, path): path_parts = os.path.normpath(path).split(os.path.sep) return self.DVC_DIR in path_parts
def __init__( self, root_dir=None, fs=None, rev=None, subrepos=False, uninitialized=False, config=None, url=None, repo_factory=None, ): from dvc.config import Config from dvc.data.db import ODBManager from dvc.data_cloud import DataCloud from dvc.fs.git import GitFileSystem from dvc.fs.local import localfs from dvc.lock import LockNoop, make_lock from dvc.repo.live import Live from dvc.repo.metrics import Metrics from dvc.repo.params import Params from dvc.repo.plots import Plots from dvc.repo.stage import StageLoad from dvc.scm import SCM from dvc.stage.cache import StageCache from dvc.state import State, StateNoop self.url = url self._fs_conf = {"repo_factory": repo_factory} self._fs = fs or localfs self._scm = None if rev and not fs: self._scm = SCM(root_dir or os.curdir) self._fs = GitFileSystem(scm=self._scm, rev=rev) self.root_dir, self.dvc_dir, self.tmp_dir = self._get_repo_dirs( root_dir=root_dir, fs=self.fs, uninitialized=uninitialized) self.config = Config(self.dvc_dir, fs=self.fs, config=config) self._uninitialized = uninitialized # used by RepoFileSystem to determine if it should traverse subrepos self.subrepos = subrepos self.cloud = DataCloud(self) self.stage = StageLoad(self) if isinstance(self.fs, GitFileSystem) or not self.dvc_dir: self.lock = LockNoop() self.state = StateNoop() self.odb = ODBManager(self) else: self.lock = make_lock( os.path.join(self.tmp_dir, "lock"), tmp_dir=self.tmp_dir, hardlink_lock=self.config["core"].get("hardlink_lock", False), friendly=True, ) state_db_dir = self._get_database_dir("state") self.state = State(self.root_dir, state_db_dir, self.dvcignore) self.odb = ODBManager(self) self.stage_cache = StageCache(self) self._ignore() self.metrics = Metrics(self) self.plots = Plots(self) self.params = Params(self) self.live = Live(self) self.stage_collection_error_handler: Optional[Callable[ [str, Exception], None]] = None self._lock_depth = 0
class Repo(object): DVC_DIR = ".dvc" from dvc.repo.destroy import destroy from dvc.repo.install import install from dvc.repo.add import add from dvc.repo.remove import remove from dvc.repo.lock import lock as lock_stage from dvc.repo.move import move from dvc.repo.run import run from dvc.repo.imp import imp from dvc.repo.imp_url import imp_url from dvc.repo.reproduce import reproduce from dvc.repo.checkout import _checkout from dvc.repo.push import push from dvc.repo.fetch import _fetch from dvc.repo.pull import pull from dvc.repo.status import status from dvc.repo.gc import gc from dvc.repo.commit import commit from dvc.repo.diff import diff from dvc.repo.brancher import brancher from dvc.repo.get import get from dvc.repo.get_url import get_url from dvc.repo.update import update def __init__(self, root_dir=None): from dvc.state import State from dvc.lock import Lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.repo.metrics import Metrics from dvc.scm.tree import WorkingTree from dvc.repo.tag import Tag root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir) self.tree = WorkingTree(self.root_dir) self.lock = Lock( os.path.join(self.dvc_dir, "lock"), tmp_dir=os.path.join(self.dvc_dir, "tmp"), ) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self, self.config.config) core = self.config.config[Config.SECTION_CORE] level = core.get(Config.SECTION_CORE_LOGLEVEL) if level: logger.setLevel(level.upper()) self.cache = Cache(self) self.cloud = DataCloud(self) self.metrics = Metrics(self) self.tag = Tag(self) self._ignore() @property def tree(self): return self._tree @tree.setter def tree(self, tree): self._tree = tree # Our graph cache is no longer valid, as it was based on the previous # tree. self._reset() def __repr__(self): return "Repo: '{root_dir}'".format(root_dir=self.root_dir) @classmethod def find_root(cls, root=None): if root is None: root = os.getcwd() else: root = os.path.abspath(os.path.realpath(root)) while True: dvc_dir = os.path.join(root, cls.DVC_DIR) if os.path.isdir(dvc_dir): return root if os.path.ismount(root): break root = os.path.dirname(root) raise NotDvcRepoError(root) @classmethod def find_dvc_dir(cls, root=None): root_dir = cls.find_root(root) return os.path.join(root_dir, cls.DVC_DIR) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False): from dvc.repo.init import init init(root_dir=root_dir, no_scm=no_scm, force=force) return Repo(root_dir) def unprotect(self, target): return self.cache.local.unprotect(PathInfo(target)) def _ignore(self): from dvc.updater import Updater updater = Updater(self.dvc_dir) flist = ( [self.config.config_local_file, updater.updater_file] + self.state.files + self.lock.files + updater.lock.files ) if self.cache.local.cache_dir.startswith(self.root_dir + os.sep): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def check_modified_graph(self, new_stages): """Generate graph including the new stage to check for errors""" self._collect_graph(self.stages + new_stages) def collect(self, target, with_deps=False, recursive=False, graph=None): import networkx as nx from dvc.stage import Stage G = graph or self.graph if not target: return get_stages(G) target = os.path.abspath(target) if recursive and os.path.isdir(target): attrs = nx.get_node_attributes(G, "stage") nodes = [node for node in nx.dfs_postorder_nodes(G)] ret = [] for node in nodes: stage = attrs[node] if stage.path.startswith(target + os.sep): ret.append(stage) return ret stage = Stage.load(self, target) if not with_deps: return [stage] node = relpath(stage.path, self.root_dir) pipeline = get_pipeline(get_pipelines(G), node) return [ pipeline.node[n]["stage"] for n in nx.dfs_postorder_nodes(pipeline, node) ] def used_cache( self, targets=None, all_branches=False, with_deps=False, all_tags=False, all_commits=False, remote=None, force=False, jobs=None, recursive=False, ): """Get the stages related to the given target and collect the `info` of its outputs. This is useful to know what files from the cache are _in use_ (namely, a file described as an output on a stage). The scope is, by default, the working directory, but you can use `all_branches` or `all_tags` to expand scope. Returns: A dictionary with Schemes (representing output's location) as keys, and a list with the outputs' `dumpd` as values. """ from dvc.cache import NamedCache cache = NamedCache() for branch in self.brancher( all_branches=all_branches, all_tags=all_tags, all_commits=all_commits, ): if targets: stages = [] for target in targets: collected = self.collect( target, recursive=recursive, with_deps=with_deps ) stages.extend(collected) else: stages = self.stages for stage in stages: if stage.is_repo_import: dep, = stage.deps cache.external[dep.repo_pair].add(dep.def_path) continue for out in stage.outs: used_cache = out.get_used_cache( remote=remote, force=force, jobs=jobs ) suffix = "({})".format(branch) if branch else "" cache.update(used_cache, suffix=suffix) return cache def _collect_graph(self, stages=None): """Generate a graph by using the given stages on the given directory The nodes of the graph are the stage's path relative to the root. Edges are created when the output of one stage is used as a dependency in other stage. The direction of the edges goes from the stage to its dependency: For example, running the following: $ dvc run -o A "echo A > A" $ dvc run -d A -o B "echo B > B" $ dvc run -d B -o C "echo C > C" Will create the following graph: ancestors <-- | C.dvc -> B.dvc -> A.dvc | | | --> descendants | ------- pipeline ------> | v (weakly connected components) Args: stages (list): used to build a graph, if None given, collect stages in the repository. Raises: OutputDuplicationError: two outputs with the same path StagePathAsOutputError: stage inside an output directory OverlappingOutputPathsError: output inside output directory CyclicGraphError: resulting graph has cycles """ import networkx as nx from dvc.exceptions import ( OutputDuplicationError, StagePathAsOutputError, OverlappingOutputPathsError, ) G = nx.DiGraph() stages = stages or self.collect_stages() stages = [stage for stage in stages if stage] outs = {} for stage in stages: for out in stage.outs: if out.path_info in outs: stages = [stage.relpath, outs[out.path_info].stage.relpath] raise OutputDuplicationError(str(out), stages) outs[out.path_info] = out for stage in stages: for out in stage.outs: for p in out.path_info.parents: if p in outs: raise OverlappingOutputPathsError(outs[p], out) for stage in stages: stage_path_info = PathInfo(stage.path) for p in chain([stage_path_info], stage_path_info.parents): if p in outs: raise StagePathAsOutputError(stage.wdir, stage.relpath) for stage in stages: node = relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) for dep in stage.deps: if dep.path_info is None: continue for out in outs: if ( out == dep.path_info or dep.path_info.isin(out) or out.isin(dep.path_info) ): dep_stage = outs[out].stage dep_node = relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) check_acyclic(G) return G @cached_property def graph(self): return self._collect_graph() @cached_property def pipelines(self): return get_pipelines(self.graph) @staticmethod def _filter_out_dirs(dirs, outs, root_dir): def filter_dirs(dname): path = os.path.join(root_dir, dname) for out in outs: if path == os.path.normpath(out): return False return True return list(filter(filter_dirs, dirs)) def collect_stages(self): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories tracked by DVC (e.g. `dvc add data` would skip `data/`) NOTE: For large repos, this could be an expensive operation. Consider using some memoization. """ from dvc.stage import Stage stages = [] outs = [] for root, dirs, files in self.tree.walk( self.root_dir, dvcignore=self.dvcignore ): for fname in files: path = os.path.join(root, fname) if not Stage.is_valid_filename(path): continue stage = Stage.load(self, path) for out in stage.outs: if out.scheme == "local": outs.append(out.fspath + out.sep) stages.append(stage) dirs[:] = self._filter_out_dirs(dirs, outs, root) return stages @cached_property def stages(self): return get_stages(self.graph) def find_outs_by_path(self, path, outs=None, recursive=False): if not outs: outs = [out for stage in self.stages for out in stage.outs] abs_path = os.path.abspath(path) is_dir = self.tree.isdir(abs_path) def func(out): if out.scheme == "local" and out.fspath == abs_path: return True if is_dir and recursive and out.path_info.isin(abs_path): return True return False matched = list(filter(func, outs)) if not matched: raise OutputNotFoundError(path) return matched def find_out_by_relpath(self, relpath): path = os.path.join(self.root_dir, relpath) out, = self.find_outs_by_path(path) return out def is_dvc_internal(self, path): path_parts = os.path.normpath(path).split(os.path.sep) return self.DVC_DIR in path_parts @contextmanager def open(self, path, remote=None, mode="r", encoding=None): """Opens a specified resource as a file descriptor""" cause = None try: out, = self.find_outs_by_path(path) except OutputNotFoundError as e: out = None cause = e if out and out.use_cache: try: with self._open_cached(out, remote, mode, encoding) as fd: yield fd return except FileNotFoundError as e: raise FileMissingError(relpath(path, self.root_dir), cause=e) if self.tree.exists(path): with self.tree.open(path, mode, encoding) as fd: yield fd return raise FileMissingError(relpath(path, self.root_dir), cause=cause) def _open_cached(self, out, remote=None, mode="r", encoding=None): if out.isdir(): raise ValueError("Can't open a dir") cache_file = self.cache.local.checksum_to_path_info(out.checksum) cache_file = fspath_py35(cache_file) if os.path.exists(cache_file): return _open(cache_file, mode=mode, encoding=encoding) try: remote_obj = self.cloud.get_remote(remote) remote_info = remote_obj.checksum_to_path_info(out.checksum) return remote_obj.open(remote_info, mode=mode, encoding=encoding) except RemoteActionNotImplemented: with self.state: cache_info = out.get_used_cache(remote=remote) self.cloud.pull(cache_info, remote=remote) return _open(cache_file, mode=mode, encoding=encoding) @cached_property def dvcignore(self): return DvcIgnoreFilter(self.root_dir, self.tree) def close(self): self.scm.close() @locked def checkout(self, *args, **kwargs): return self._checkout(*args, **kwargs) @locked def fetch(self, *args, **kwargs): return self._fetch(*args, **kwargs) def _reset(self): self.__dict__.pop("graph", None) self.__dict__.pop("stages", None) self.__dict__.pop("pipelines", None) self.__dict__.pop("dvcignore", None)
def setUp(self): super().setUp() self.scm = SCM(self._root_dir) self.tree = GitTree(self.git, "master") self._pushd(self._root_dir)
def init(root_dir=os.curdir, no_scm=False, force=False, subdir=False): """ Creates an empty repo on the given directory -- basically a `.dvc` directory with subdirectories for configuration and cache. It should be tracked by a SCM or use the `--no-scm` flag. If the given directory is not empty, you must use the `--force` flag to override it. Args: root_dir: Path to repo's root directory. Returns: Repo instance. Raises: KeyError: Raises an exception. """ if no_scm and subdir: raise InvalidArgumentError( "Cannot initialize repo with `--no-scm` and `--subdir`") root_dir = os.path.realpath(root_dir) dvc_dir = os.path.join(root_dir, Repo.DVC_DIR) try: scm = SCM(root_dir, search_parent_directories=subdir, no_scm=no_scm) except SCMError: raise InitError( f"{root_dir} is not tracked by any supported SCM tool (e.g. Git). " "Use `--no-scm` if you don't want to use any SCM or " "`--subdir` if initializing inside a subdirectory of a parent SCM " "repository.") if scm.is_ignored(dvc_dir): raise InitError(f"{dvc_dir} is ignored by your SCM tool. \n" "Make sure that it's tracked, " "for example, by adding '!.dvc' to .gitignore.") if os.path.isdir(dvc_dir): if not force: raise InitError(f"'{relpath(dvc_dir)}' exists. Use `-f` to force.") remove(dvc_dir) os.mkdir(dvc_dir) config = Config.init(dvc_dir) if no_scm: with config.edit() as conf: conf["core"]["no_scm"] = True dvcignore = init_dvcignore(root_dir) proj = Repo(root_dir) proj.plots.templates.init() with proj.scm_context(autostage=True) as context: files = [ config.files["repo"], dvcignore, proj.plots.templates.templates_dir, ] ignore_file = context.scm.ignore_file if ignore_file: files.extend([os.path.join(dvc_dir, ignore_file)]) proj.scm_context.track_file(files) logger.info("Initialized DVC repository.\n") if not no_scm: logger.info("You can now commit the changes to git.\n") return proj
class Project(object): DVC_DIR = ".dvc" def __init__(self, root_dir=None): from dvc.config import Config from dvc.state import State from dvc.lock import Lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.updater import Updater root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir, project=self) self.lock = Lock(self.dvc_dir) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self, self.config.config) core = self.config.config[Config.SECTION_CORE] logger.set_level(core.get(Config.SECTION_CORE_LOGLEVEL)) self.cache = Cache(self) self.cloud = DataCloud(self, config=self.config.config) self.updater = Updater(self.dvc_dir) self.files_to_git_add = [] self._ignore() self.updater.check() def __repr__(self): return "Project: '{root_dir}'".format(root_dir=self.root_dir) @staticmethod def find_root(root=None): if root is None: root = os.getcwd() else: root = os.path.abspath(os.path.realpath(root)) while True: dvc_dir = os.path.join(root, Project.DVC_DIR) if os.path.isdir(dvc_dir): return root if os.path.ismount(root): break root = os.path.dirname(root) raise NotDvcProjectError(root) @staticmethod def find_dvc_dir(root=None): root_dir = Project.find_root(root) return os.path.join(root_dir, Project.DVC_DIR) def _remind_to_git_add(self): if not self.files_to_git_add: return logger.info( "\n" "To track the changes with git run:\n" "\n" "\tgit add {files}".format(files=" ".join(self.files_to_git_add))) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False): """ Creates an empty project on the given directory -- basically a `.dvc` directory with subdirectories for configuration and cache. It should be tracked by a SCM or use the `--no-scm` flag. If the given directory is not empty, you must use the `--force` flag to override it. Args: root_dir: Path to project's root directory. Returns: Project instance. Raises: KeyError: Raises an exception. """ import shutil from dvc.scm import SCM, Base from dvc.config import Config root_dir = os.path.abspath(root_dir) dvc_dir = os.path.join(root_dir, Project.DVC_DIR) scm = SCM(root_dir) if type(scm) == Base and not no_scm: raise InitError( "{project} is not tracked by any supported scm tool" " (e.g. git). Use '--no-scm' if you don't want to use any scm." .format(project=root_dir)) if os.path.isdir(dvc_dir): if not force: raise InitError( "'{project}' exists. Use '-f' to force.".format( project=os.path.relpath(dvc_dir))) shutil.rmtree(dvc_dir) os.mkdir(dvc_dir) config = Config.init(dvc_dir) proj = Project(root_dir) scm.add([config.config_file]) if scm.ignore_file: scm.add([os.path.join(dvc_dir, scm.ignore_file)]) logger.info("\nYou can now commit the changes to git.\n") proj._welcome_message() return proj def destroy(self): import shutil for stage in self.stages(): stage.remove() shutil.rmtree(self.dvc_dir) def _ignore(self): flist = [ self.state.state_file, self.lock.lock_file, self.config.config_local_file, self.updater.updater_file, self.updater.lock.lock_file, ] + self.state.temp_files if self.cache.local.cache_dir.startswith(self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def install(self): self.scm.install() @staticmethod def _check_cyclic_graph(graph): import networkx as nx from dvc.exceptions import CyclicGraphError cycles = list(nx.simple_cycles(graph)) if cycles: raise CyclicGraphError(cycles[0]) def add(self, fname, recursive=False): from dvc.stage import Stage fnames = [] if recursive and os.path.isdir(fname): for root, dirs, files in os.walk(fname): for f in files: path = os.path.join(root, f) if Stage.is_stage_file(path): continue if os.path.basename(path) == self.scm.ignore_file: continue if self.scm.is_tracked(path): continue fnames.append(path) else: fnames = [fname] stages = [] self.files_to_git_add = [] with self.state: for f in fnames: stage = Stage.create(project=self, outs=[f], add=True) if stage is None: stages.append(stage) continue stage.save() stages.append(stage) self._check_dag(self.stages() + stages) for stage in stages: stage.dump() self._remind_to_git_add() return stages def _check_dag(self, stages): """Generate graph including the new stage to check for errors""" self.graph(stages=stages) def remove(self, target, outs_only=False): from dvc.stage import Stage stage = Stage.load(self, target) if outs_only: stage.remove_outs() else: stage.remove() return stage def lock_stage(self, target, unlock=False): from dvc.stage import Stage stage = Stage.load(self, target) stage.locked = False if unlock else True stage.dump() return stage def move(self, from_path, to_path): """ Renames an output file and modifies the stage associated to reflect the change on the pipeline. If the output has the same name as its stage, it would also rename the corresponding stage file. E.g. Having: (hello, hello.dvc) $ dvc move hello greetings Result: (greeting, greeting.dvc) It only works with outputs generated by `add` or `import`, also known as data sources. """ import dvc.output as Output from dvc.stage import Stage from_out = Output.loads_from(Stage(self, cwd=os.curdir), [from_path])[0] to_path = self._expand_target_path(from_path, to_path) try: stage, out = next((stage, out) for stage in self.stages() for out in stage.outs if from_out.path == out.path) except StopIteration: raise DvcException( "unable to find stage file with output '{path}'".format( path=from_path)) if not stage.is_data_source: raise MoveNotDataSourceError(stage.relpath) stage_name = os.path.splitext(os.path.basename(stage.path))[0] from_name = os.path.basename(from_out.path) if stage_name == from_name: os.unlink(stage.path) stage.path = os.path.join( os.path.dirname(to_path), os.path.basename(to_path) + Stage.STAGE_FILE_SUFFIX, ) stage.cwd = os.path.join(self.root_dir, os.path.dirname(to_path)) to_out = Output.loads_from(stage, [os.path.basename(to_path)], out.cache, out.metric)[0] with self.state: out.move(to_out) stage.dump() self._remind_to_git_add() def _unprotect_file(self, path): import stat import uuid from dvc.system import System from dvc.utils import copyfile, move, remove if System.is_symlink(path) or System.is_hardlink(path): logger.debug("Unprotecting '{}'".format(path)) tmp = os.path.join(os.path.dirname(path), "." + str(uuid.uuid4())) move(path, tmp) copyfile(tmp, path) remove(tmp) else: logger.debug("Skipping copying for '{}', since it is not " "a symlink or a hardlink.".format(path)) os.chmod(path, os.stat(path).st_mode | stat.S_IWRITE) def _unprotect_dir(self, path): for root, dirs, files in os.walk(path): for f in files: path = os.path.join(root, f) self._unprotect_file(path) def unprotect(self, path): if not os.path.exists(path): raise DvcException( "can't unprotect non-existing data '{}'".format(path)) if os.path.isdir(path): self._unprotect_dir(path) else: self._unprotect_file(path) def run( self, cmd=None, deps=[], outs=[], outs_no_cache=[], metrics_no_cache=[], fname=None, cwd=os.curdir, no_exec=False, overwrite=False, ignore_build_cache=False, remove_outs=False, ): from dvc.stage import Stage with self.state: stage = Stage.create( project=self, fname=fname, cmd=cmd, cwd=cwd, outs=outs, outs_no_cache=outs_no_cache, metrics_no_cache=metrics_no_cache, deps=deps, overwrite=overwrite, ignore_build_cache=ignore_build_cache, remove_outs=remove_outs, ) if stage is None: return None self._check_dag(self.stages() + [stage]) self.files_to_git_add = [] with self.state: if not no_exec: stage.run() stage.dump() self._remind_to_git_add() return stage def imp(self, url, out, resume=False): from dvc.stage import Stage stage = Stage.create(project=self, cmd=None, deps=[url], outs=[out]) if stage is None: return None self._check_dag(self.stages() + [stage]) self.files_to_git_add = [] with self.state: stage.run(resume=resume) stage.dump() self._remind_to_git_add() return stage def _reproduce_stage(self, stages, node, force, dry, interactive): stage = stages[node] if stage.locked: logger.warning( "DVC file '{path}' is locked. Its dependencies are" " not going to be reproduced.".format(path=stage.relpath)) stage = stage.reproduce(force=force, dry=dry, interactive=interactive) if not stage: return [] if not dry: stage.dump() return [stage] def reproduce( self, target=None, recursive=True, force=False, dry=False, interactive=False, pipeline=False, all_pipelines=False, ignore_build_cache=False, ): from dvc.stage import Stage if not target and not all_pipelines: raise ValueError() if not interactive: config = self.config core = config.config[config.SECTION_CORE] interactive = core.get(config.SECTION_CORE_INTERACTIVE, False) targets = [] if pipeline or all_pipelines: if pipeline: stage = Stage.load(self, target) node = os.path.relpath(stage.path, self.root_dir) pipelines = [self._get_pipeline(node)] else: pipelines = self.pipelines() for G in pipelines: for node in G.nodes(): if G.in_degree(node) == 0: targets.append(os.path.join(self.root_dir, node)) else: targets.append(target) self.files_to_git_add = [] ret = [] with self.state: for target in targets: stages = self._reproduce( target, recursive=recursive, force=force, dry=dry, interactive=interactive, ignore_build_cache=ignore_build_cache, ) ret.extend(stages) self._remind_to_git_add() return ret def _reproduce( self, target, recursive=True, force=False, dry=False, interactive=False, ignore_build_cache=False, ): import networkx as nx from dvc.stage import Stage stage = Stage.load(self, target) G = self.graph()[1] stages = nx.get_node_attributes(G, "stage") node = os.path.relpath(stage.path, self.root_dir) if recursive: ret = self._reproduce_stages(G, stages, node, force, dry, interactive, ignore_build_cache) else: ret = self._reproduce_stage(stages, node, force, dry, interactive) return ret def _reproduce_stages(self, G, stages, node, force, dry, interactive, ignore_build_cache): import networkx as nx result = [] for n in nx.dfs_postorder_nodes(G, node): try: ret = self._reproduce_stage(stages, n, force, dry, interactive) if len(ret) == 0 and ignore_build_cache: # NOTE: we are walking our pipeline from the top to the # bottom. If one stage is changed, it will be reproduced, # which tells us that we should force reproducing all of # the other stages down below, even if their direct # dependencies didn't change. force = True result += ret except Exception as ex: raise ReproductionError(stages[n].relpath, ex) return result def _cleanup_unused_links(self, all_stages): used = [] for stage in all_stages: for out in stage.outs: used.append(out.path) self.state.remove_unused_links(used) def checkout(self, target=None, with_deps=False, force=False, recursive=False): if target and not recursive: all_stages = self.active_stages() stages = self._collect(target, with_deps=with_deps) else: all_stages = self.active_stages(target) stages = all_stages with self.state: self._cleanup_unused_links(all_stages) for stage in stages: if stage.locked: logger.warning( "DVC file '{path}' is locked. Its dependencies are" " not going to be checked out.".format( path=stage.relpath)) stage.checkout(force=force) def _get_pipeline(self, node): pipelines = list(filter(lambda g: node in g.nodes(), self.pipelines())) assert len(pipelines) == 1 return pipelines[0] def _collect(self, target, with_deps=False): import networkx as nx from dvc.stage import Stage stage = Stage.load(self, target) if not with_deps: return [stage] node = os.path.relpath(stage.path, self.root_dir) G = self._get_pipeline(node) stages = nx.get_node_attributes(G, "stage") ret = [stage] for n in nx.dfs_postorder_nodes(G, node): ret.append(stages[n]) return ret def _collect_dir_cache(self, out, branch=None, remote=None, force=False, jobs=None): info = out.dumpd() ret = [info] r = out.remote md5 = info[r.PARAM_CHECKSUM] if self.cache.local.changed_cache_file(md5): try: self.cloud.pull(ret, jobs=jobs, remote=remote, show_checksums=False) except DvcException as exc: msg = "Failed to pull cache for '{}': {}" logger.debug(msg.format(out, exc)) if self.cache.local.changed_cache_file(md5): msg = ("Missing cache for directory '{}'. " "Cache for files inside will be lost. " "Would you like to continue? Use '-f' to force. ") if not force and not prompt.confirm(msg): raise DvcException( "unable to fully collect used cache" " without cache for directory '{}'".format(out)) else: return ret for i in self.cache.local.load_dir_cache(md5): i["branch"] = branch i[r.PARAM_PATH] = os.path.join(info[r.PARAM_PATH], i[r.PARAM_RELPATH]) ret.append(i) return ret def _collect_used_cache(self, out, branch=None, remote=None, force=False, jobs=None): if not out.use_cache or not out.info: if not out.info: logger.warning("Output '{}'({}) is missing version " "info. Cache for it will not be collected. " "Use dvc repro to get your pipeline up to " "date.".format(out, out.stage)) return [] info = out.dumpd() info["branch"] = branch ret = [info] if out.scheme != "local": return ret md5 = info[out.remote.PARAM_CHECKSUM] cache = self.cache.local.get(md5) if not out.remote.is_dir_cache(cache): return ret return self._collect_dir_cache(out, branch=branch, remote=remote, force=force, jobs=jobs) def _used_cache( self, target=None, all_branches=False, active=True, with_deps=False, all_tags=False, remote=None, force=False, jobs=None, recursive=False, ): cache = {} cache["local"] = [] cache["s3"] = [] cache["gs"] = [] cache["hdfs"] = [] cache["ssh"] = [] cache["azure"] = [] for branch in self.scm.brancher(all_branches=all_branches, all_tags=all_tags): if target: if recursive: stages = self.stages(target) else: stages = self._collect(target, with_deps=with_deps) elif active: stages = self.active_stages() else: stages = self.stages() for stage in stages: if active and not target and stage.locked: logger.warning( "DVC file '{path}' is locked. Its dependencies are" " not going to be pushed/pulled/fetched.".format( path=stage.relpath)) for out in stage.outs: scheme = out.path_info["scheme"] cache[scheme] += self._collect_used_cache( out, branch=branch, remote=remote, force=force, jobs=jobs, ) return cache @staticmethod def merge_cache_lists(clists): merged_cache = collections.defaultdict(list) for cache_list in clists: for scheme, cache in cache_list.items(): for item in cache: if item not in merged_cache[scheme]: merged_cache[scheme].append(item) return merged_cache @staticmethod def load_all_used_cache( projects, target=None, all_branches=False, active=True, with_deps=False, all_tags=False, remote=None, force=False, jobs=None, ): clists = [] for project in projects: with project.state: project_clist = project._used_cache( target=None, all_branches=all_branches, active=False, with_deps=with_deps, all_tags=all_tags, remote=remote, force=force, jobs=jobs, ) clists.append(project_clist) return clists def _do_gc(self, typ, func, clist): removed = func(clist) if not removed: logger.info("No unused {} cache to remove.".format(typ)) def gc( self, all_branches=False, cloud=False, remote=None, with_deps=False, all_tags=False, force=False, jobs=None, projects=None, ): all_projects = [self] if projects: all_projects.extend(Project(path) for path in projects) all_clists = Project.load_all_used_cache( all_projects, target=None, all_branches=all_branches, active=False, with_deps=with_deps, all_tags=all_tags, remote=remote, force=force, jobs=jobs, ) if len(all_clists) > 1: clist = Project.merge_cache_lists(all_clists) else: clist = all_clists[0] with self.state: self._do_gc("local", self.cache.local.gc, clist) if self.cache.s3: self._do_gc("s3", self.cache.s3.gc, clist) if self.cache.gs: self._do_gc("gs", self.cache.gs.gc, clist) if self.cache.ssh: self._do_gc("ssh", self.cache.ssh.gc, clist) if self.cache.hdfs: self._do_gc("hdfs", self.cache.hdfs.gc, clist) if self.cache.azure: self._do_gc("azure", self.cache.azure.gc, clist) if cloud: self._do_gc("remote", self.cloud._get_cloud(remote, "gc -c").gc, clist) def push( self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False, with_deps=False, all_tags=False, recursive=False, ): with self.state: used = self._used_cache( target, all_branches=all_branches, all_tags=all_tags, with_deps=with_deps, force=True, remote=remote, jobs=jobs, recursive=recursive, )["local"] self.cloud.push(used, jobs, remote=remote, show_checksums=show_checksums) def fetch( self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False, with_deps=False, all_tags=False, recursive=False, ): with self.state: used = self._used_cache( target, all_branches=all_branches, all_tags=all_tags, with_deps=with_deps, force=True, remote=remote, jobs=jobs, recursive=recursive, )["local"] self.cloud.pull(used, jobs, remote=remote, show_checksums=show_checksums) def pull( self, target=None, jobs=1, remote=None, all_branches=False, show_checksums=False, with_deps=False, all_tags=False, force=False, recursive=False, ): self.fetch( target, jobs, remote=remote, all_branches=all_branches, all_tags=all_tags, show_checksums=show_checksums, with_deps=with_deps, recursive=recursive, ) self.checkout( target=target, with_deps=with_deps, force=force, recursive=recursive, ) def _local_status(self, target=None, with_deps=False): status = {} if target: stages = self._collect(target, with_deps=with_deps) else: stages = self.active_stages() for stage in stages: if stage.locked: logger.warning( "DVC file '{path}' is locked. Its dependencies are" " not going to be shown in the status output.".format( path=stage.relpath)) status.update(stage.status()) return status def _cloud_status( self, target=None, jobs=1, remote=None, show_checksums=False, all_branches=False, with_deps=False, all_tags=False, ): import dvc.remote.base as cloud used = self._used_cache( target, all_branches=all_branches, all_tags=all_tags, with_deps=with_deps, force=True, remote=remote, jobs=jobs, )["local"] ret = {} status_info = self.cloud.status(used, jobs, remote=remote, show_checksums=show_checksums) for md5, info in status_info.items(): name = info["name"] status = info["status"] if status == cloud.STATUS_OK: continue prefix_map = { cloud.STATUS_DELETED: "deleted", cloud.STATUS_NEW: "new", } ret[name] = prefix_map[status] return ret def status( self, target=None, jobs=1, cloud=False, remote=None, show_checksums=False, all_branches=False, with_deps=False, all_tags=False, ): with self.state: if cloud: return self._cloud_status( target, jobs, remote=remote, show_checksums=show_checksums, all_branches=all_branches, with_deps=with_deps, all_tags=all_tags, ) return self._local_status(target, with_deps=with_deps) def _read_metric_json(self, fd, json_path): import json from jsonpath_rw import parse parser = parse(json_path) return [x.value for x in parser.find(json.load(fd))] def _do_read_metric_xsv(self, reader, row, col): if col is not None and row is not None: return [reader[row][col]] elif col is not None: return [r[col] for r in reader] elif row is not None: return reader[row] return None def _read_metric_hxsv(self, fd, hxsv_path, delimiter): import csv col, row = hxsv_path.split(",") row = int(row) reader = list(csv.DictReader(fd, delimiter=builtin_str(delimiter))) return self._do_read_metric_xsv(reader, row, col) def _read_metric_xsv(self, fd, xsv_path, delimiter): import csv col, row = xsv_path.split(",") row = int(row) col = int(col) reader = list(csv.reader(fd, delimiter=builtin_str(delimiter))) return self._do_read_metric_xsv(reader, row, col) def _read_metric(self, path, typ=None, xpath=None): ret = None if not os.path.exists(path): return ret try: with open(path, "r") as fd: if typ == "json": ret = self._read_metric_json(fd, xpath) elif typ == "csv": ret = self._read_metric_xsv(fd, xpath, ",") elif typ == "tsv": ret = self._read_metric_xsv(fd, xpath, "\t") elif typ == "hcsv": ret = self._read_metric_hxsv(fd, xpath, ",") elif typ == "htsv": ret = self._read_metric_hxsv(fd, xpath, "\t") else: ret = fd.read().strip() except Exception: logger.error("unable to read metric in '{}'".format(path)) return ret def _find_output_by_path(self, path, outs=None, recursive=False): from dvc.exceptions import OutputDuplicationError if not outs: astages = self.active_stages() outs = [out for stage in astages for out in stage.outs] abs_path = os.path.abspath(path) if os.path.isdir(abs_path) and recursive: matched = [ out for out in outs if os.path.abspath(out.path).startswith(abs_path) ] else: matched = [out for out in outs if out.path == abs_path] stages = [out.stage.relpath for out in matched] if len(stages) > 1: raise OutputDuplicationError(path, stages) return matched if matched else [] def metrics_show( self, path=None, typ=None, xpath=None, all_branches=False, all_tags=False, recursive=False, ): res = {} for branch in self.scm.brancher(all_branches=all_branches, all_tags=all_tags): astages = self.active_stages() outs = [out for stage in astages for out in stage.outs] if path: outs = self._find_output_by_path(path, outs=outs, recursive=recursive) stages = [out.stage.path for out in outs] entries = [] for out in outs: if all([out.metric, not typ, isinstance(out.metric, dict)]): entries += [( out.path, out.metric.get(out.PARAM_METRIC_TYPE, None), out.metric.get(out.PARAM_METRIC_XPATH, None), )] else: entries += [(out.path, typ, xpath)] if not entries: if os.path.isdir(path): logger.warning( "Path '{path}' is a directory. " "Consider runnig with '-R'.".format(path=path)) return {} else: entries += [(path, typ, xpath)] else: metrics = filter(lambda o: o.metric, outs) stages = None entries = [] for o in metrics: if not typ and isinstance(o.metric, dict): t = o.metric.get(o.PARAM_METRIC_TYPE, typ) x = o.metric.get(o.PARAM_METRIC_XPATH, xpath) else: t = typ x = xpath entries.append((o.path, t, x)) for fname, t, x in entries: if stages: for stage in stages: self.checkout(stage, force=True) rel = os.path.relpath(fname) metric = self._read_metric(fname, typ=t, xpath=x) if not metric: continue if branch not in res: res[branch] = {} res[branch][rel] = metric for branch, val in res.items(): if all_branches or all_tags: logger.info("{}:".format(branch)) for fname, metric in val.items(): logger.info("\t{}: {}".format(fname, metric)) if res: return res if path and os.path.isdir(path): msg = "file '{}' does not exist".format(path) return res if path: msg = "file '{}' does not exist".format(path) else: msg = ("no metric files in this repository." " use 'dvc metrics add' to add a metric file to track.") raise DvcException(msg) def _metrics_modify(self, path, typ=None, xpath=None, delete=False): outs = self._find_output_by_path(path) if not outs: msg = "unable to find file '{}' in the pipeline".format(path) raise DvcException(msg) if len(outs) != 1: msg = ("-R not yet supported for metrics modify. " "Make sure only one metric is referred to by '{}'".format( path)) raise DvcException(msg) out = outs[0] if out.scheme != "local": msg = "output '{}' scheme '{}' is not supported for metrics" raise DvcException(msg.format(out.path, out.path_info["scheme"])) if out.use_cache: msg = "cached output '{}' is not supported for metrics" raise DvcException(msg.format(out.rel_path)) if typ: if not isinstance(out.metric, dict): out.metric = {} out.metric[out.PARAM_METRIC_TYPE] = typ if xpath: if not isinstance(out.metric, dict): out.metric = {} out.metric[out.PARAM_METRIC_XPATH] = xpath if delete: out.metric = None out._verify_metric() out.stage.dump() def metrics_modify(self, path=None, typ=None, xpath=None): self._metrics_modify(path, typ, xpath) def metrics_add(self, path, typ=None, xpath=None): if not typ: typ = "raw" self._metrics_modify(path, typ, xpath) def metrics_remove(self, path): self._metrics_modify(path, delete=True) def graph(self, stages=None, from_directory=None): import networkx as nx from dvc.exceptions import ( OutputDuplicationError, WorkingDirectoryAsOutputError, ) G = nx.DiGraph() G_active = nx.DiGraph() stages = stages or self.stages(from_directory) stages = [stage for stage in stages if stage] outs = [] for stage in stages: for out in stage.outs: existing = [o.stage for o in outs if o.path == out.path] if existing: stages = [stage.relpath, existing[0].relpath] raise OutputDuplicationError(out.path, stages) outs.append(out) for stage in stages: for out in outs: overlaps = stage.cwd == out.path or stage.cwd.startswith( out.path + os.sep) if overlaps: raise WorkingDirectoryAsOutputError( stage.cwd, stage.relpath) # collect the whole DAG for stage in stages: node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) G_active.add_node(node, stage=stage) for dep in stage.deps: for out in outs: if (out.path != dep.path and not dep.path.startswith(out.path + out.sep) and not out.path.startswith(dep.path + dep.sep)): continue dep_stage = out.stage dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) if not stage.locked: G_active.add_node(dep_node, stage=dep_stage) G_active.add_edge(node, dep_node) self._check_cyclic_graph(G) return G, G_active def pipelines(self, from_directory=None): import networkx as nx G, G_active = self.graph(from_directory=from_directory) return [ G.subgraph(c).copy() for c in nx.weakly_connected_components(G) ] def stages(self, from_directory=None): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories tracked by DVC (e.g. `dvc add data` would skip `data/`) NOTE: For large projects, this could be an expensive operation. Consider using some memoization. """ from dvc.stage import Stage if not from_directory: from_directory = self.root_dir stages = [] outs = [] for root, dirs, files in os.walk(from_directory): for fname in files: path = os.path.join(root, fname) if not Stage.is_stage_file(path): continue stage = Stage.load(self, path) for out in stage.outs: outs.append(out.path + out.sep) stages.append(stage) def filter_dirs(dname, root=root): path = os.path.join(root, dname) if path in (self.dvc_dir, self.scm.dir): return False for out in outs: if path == os.path.normpath(out) or path.startswith(out): return False return True dirs[:] = list(filter(filter_dirs, dirs)) return stages def active_stages(self, from_directory=None): import networkx as nx stages = [] for G in self.pipelines(from_directory): stages.extend(list(nx.get_node_attributes(G, "stage").values())) return stages def _welcome_message(self): import colorama logger.box( "DVC has enabled anonymous aggregate usage analytics.\n" "Read the analytics documentation (and how to opt-out) here:\n" "{blue}https://dvc.org/doc/user-guide/analytics{nc}".format( blue=colorama.Fore.BLUE, nc=colorama.Fore.RESET), border_color="red", ) logger.info( "{yellow}What's next?{nc}\n" "{yellow}------------{nc}\n" "- Check out the documentation: {blue}https://dvc.org/doc{nc}\n" "- Get help and share ideas: {blue}https://dvc.org/chat{nc}\n" "- Star us on GitHub: {blue}https://github.com/iterative/dvc{nc}". format( yellow=colorama.Fore.YELLOW, blue=colorama.Fore.BLUE, nc=colorama.Fore.RESET, )) def _expand_target_path(self, from_path, to_path): if os.path.isdir(to_path) and not os.path.isdir(from_path): return os.path.join(to_path, os.path.basename(from_path)) return to_path
class Repo(object): DVC_DIR = ".dvc" from dvc.repo.destroy import destroy from dvc.repo.install import install from dvc.repo.add import add from dvc.repo.remove import remove from dvc.repo.lock import lock as lock_stage from dvc.repo.move import move from dvc.repo.run import run from dvc.repo.imp import imp from dvc.repo.reproduce import reproduce from dvc.repo.checkout import checkout from dvc.repo.push import push from dvc.repo.fetch import fetch from dvc.repo.pull import pull from dvc.repo.status import status from dvc.repo.gc import gc from dvc.repo.commit import commit from dvc.repo.pkg import install_pkg def __init__(self, root_dir=None): from dvc.config import Config from dvc.state import State from dvc.lock import Lock from dvc.scm import SCM from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.updater import Updater from dvc.repo.metrics import Metrics root_dir = self.find_root(root_dir) self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir, repo=self) self.lock = Lock(self.dvc_dir) # NOTE: storing state and link_state in the repository itself to avoid # any possible state corruption in 'shared cache dir' scenario. self.state = State(self, self.config.config) core = self.config.config[Config.SECTION_CORE] logger.set_level(core.get(Config.SECTION_CORE_LOGLEVEL)) self.cache = Cache(self) self.cloud = DataCloud(self, config=self.config.config) self.updater = Updater(self.dvc_dir) self.metrics = Metrics(self) self._ignore() self.updater.check() def __repr__(self): return "Repo: '{root_dir}'".format(root_dir=self.root_dir) @staticmethod def find_root(root=None): if root is None: root = os.getcwd() else: root = os.path.abspath(os.path.realpath(root)) while True: dvc_dir = os.path.join(root, Repo.DVC_DIR) if os.path.isdir(dvc_dir): return root if os.path.ismount(root): break root = os.path.dirname(root) raise NotDvcRepoError(root) @staticmethod def find_dvc_dir(root=None): root_dir = Repo.find_root(root) return os.path.join(root_dir, Repo.DVC_DIR) @staticmethod def init(root_dir=os.curdir, no_scm=False, force=False): from dvc.repo.init import init init(root_dir=root_dir, no_scm=no_scm, force=force) return Repo(root_dir) @staticmethod def unprotect(target): from dvc.repo.unprotect import unprotect return unprotect(target) def _ignore(self): flist = [ self.state.state_file, self.lock.lock_file, self.config.config_local_file, self.updater.updater_file, self.updater.lock.lock_file, ] + self.state.temp_files if self.cache.local.cache_dir.startswith(self.root_dir): flist += [self.cache.local.cache_dir] self.scm.ignore_list(flist) def check_dag(self, stages): """Generate graph including the new stage to check for errors""" self.graph(stages=stages) @staticmethod def _check_cyclic_graph(graph): import networkx as nx from dvc.exceptions import CyclicGraphError cycles = list(nx.simple_cycles(graph)) if cycles: raise CyclicGraphError(cycles[0]) def _get_pipeline(self, node): pipelines = list(filter(lambda g: node in g.nodes(), self.pipelines())) assert len(pipelines) == 1 return pipelines[0] def collect(self, target, with_deps=False): import networkx as nx from dvc.stage import Stage stage = Stage.load(self, target) if not with_deps: return [stage] node = os.path.relpath(stage.path, self.root_dir) G = self._get_pipeline(node) stages = nx.get_node_attributes(G, "stage") ret = [stage] for n in nx.dfs_postorder_nodes(G, node): ret.append(stages[n]) return ret def _collect_dir_cache( self, out, branch=None, remote=None, force=False, jobs=None ): info = out.dumpd() ret = [info] r = out.remote md5 = info[r.PARAM_CHECKSUM] if self.cache.local.changed_cache_file(md5): try: self.cloud.pull( ret, jobs=jobs, remote=remote, show_checksums=False ) except DvcException as exc: msg = "Failed to pull cache for '{}': {}" logger.debug(msg.format(out, exc)) if self.cache.local.changed_cache_file(md5): msg = ( "Missing cache for directory '{}'. " "Cache for files inside will be lost. " "Would you like to continue? Use '-f' to force. " ) if not force and not prompt.confirm(msg): raise DvcException( "unable to fully collect used cache" " without cache for directory '{}'".format(out) ) else: return ret for i in self.cache.local.load_dir_cache(md5): i["branch"] = branch i[r.PARAM_PATH] = os.path.join( info[r.PARAM_PATH], i[r.PARAM_RELPATH] ) ret.append(i) return ret def _collect_used_cache( self, out, branch=None, remote=None, force=False, jobs=None ): if not out.use_cache or not out.info: if not out.info: logger.warning( "Output '{}'({}) is missing version " "info. Cache for it will not be collected. " "Use dvc repro to get your pipeline up to " "date.".format(out, out.stage) ) return [] info = out.dumpd() info["branch"] = branch ret = [info] if out.scheme != "local": return ret md5 = info[out.remote.PARAM_CHECKSUM] cache = self.cache.local.get(md5) if not out.remote.is_dir_cache(cache): return ret return self._collect_dir_cache( out, branch=branch, remote=remote, force=force, jobs=jobs ) def used_cache( self, target=None, all_branches=False, active=True, with_deps=False, all_tags=False, remote=None, force=False, jobs=None, recursive=False, ): cache = {} cache["local"] = [] cache["s3"] = [] cache["gs"] = [] cache["hdfs"] = [] cache["ssh"] = [] cache["azure"] = [] for branch in self.scm.brancher( all_branches=all_branches, all_tags=all_tags ): if target: if recursive: stages = self.stages(target) else: stages = self.collect(target, with_deps=with_deps) elif active: stages = self.active_stages() else: stages = self.stages() for stage in stages: if active and not target and stage.locked: logger.warning( "DVC file '{path}' is locked. Its dependencies are" " not going to be pushed/pulled/fetched.".format( path=stage.relpath ) ) for out in stage.outs: scheme = out.path_info["scheme"] cache[scheme] += self._collect_used_cache( out, branch=branch, remote=remote, force=force, jobs=jobs, ) return cache def graph(self, stages=None, from_directory=None): import networkx as nx from dvc.exceptions import ( OutputDuplicationError, StagePathAsOutputError, ) G = nx.DiGraph() G_active = nx.DiGraph() stages = stages or self.stages(from_directory, check_dag=False) stages = [stage for stage in stages if stage] outs = [] for stage in stages: for out in stage.outs: existing = [o.stage for o in outs if o.path == out.path] if existing: stages = [stage.relpath, existing[0].relpath] raise OutputDuplicationError(out.path, stages) outs.append(out) for stage in stages: path_dir = os.path.dirname(stage.path) + os.sep for out in outs: if path_dir.startswith(out.path + os.sep): raise StagePathAsOutputError(stage.wdir, stage.relpath) # collect the whole DAG for stage in stages: node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) G_active.add_node(node, stage=stage) for dep in stage.deps: for out in outs: if ( out.path != dep.path and not dep.path.startswith(out.path + out.sep) and not out.path.startswith(dep.path + dep.sep) ): continue dep_stage = out.stage dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) if not stage.locked: G_active.add_node(dep_node, stage=dep_stage) G_active.add_edge(node, dep_node) self._check_cyclic_graph(G) return G, G_active def pipelines(self, from_directory=None): import networkx as nx G, G_active = self.graph(from_directory=from_directory) return [ G.subgraph(c).copy() for c in nx.weakly_connected_components(G) ] def stages(self, from_directory=None, check_dag=True): """ Walks down the root directory looking for Dvcfiles, skipping the directories that are related with any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories tracked by DVC (e.g. `dvc add data` would skip `data/`) NOTE: For large repos, this could be an expensive operation. Consider using some memoization. """ from dvc.stage import Stage if not from_directory: from_directory = self.root_dir stages = [] outs = [] for root, dirs, files in os.walk(str(from_directory)): for fname in files: path = os.path.join(root, fname) if not Stage.is_stage_file(path): continue stage = Stage.load(self, path) for out in stage.outs: outs.append(out.path + out.sep) stages.append(stage) def filter_dirs(dname, root=root): path = os.path.join(root, dname) if path in (self.dvc_dir, self.scm.dir): return False for out in outs: if path == os.path.normpath(out) or path.startswith(out): return False return True dirs[:] = list(filter(filter_dirs, dirs)) if check_dag: self.check_dag(stages) return stages def active_stages(self, from_directory=None): import networkx as nx stages = [] for G in self.pipelines(from_directory): stages.extend(list(nx.get_node_attributes(G, "stage").values())) return stages def find_outs_by_path(self, path, outs=None, recursive=False): if not outs: outs = [out for stage in self.stages() for out in stage.outs] abs_path = os.path.abspath(path) is_dir = os.path.isdir(abs_path) def func(out): if out.path == abs_path: return True if is_dir and recursive and out.path.startswith(abs_path + os.sep): return True return False matched = list(filter(func, outs)) if not matched: raise OutputNotFoundError(path) return matched def is_dvc_internal(self, path): path_parts = os.path.normpath(path).split(os.path.sep) return self.DVC_DIR in path_parts
def scm(self): return SCM(self.root_dir)
def _scm_in_use(): try: scm = SCM(root_dir=Repo.find_root()) return type(scm).__name__ except NotDvcRepoError: pass
def test_init_git(tmp_dir): Repo.init(os.fspath(tmp_dir)) assert isinstance(SCM(os.fspath(tmp_dir)), Git)
class Project(object): DVC_DIR = '.dvc' def __init__(self, root_dir): self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.scm = SCM(self.root_dir) self.lock = Lock(self.dvc_dir) self.cache = Cache(self.dvc_dir) self.state = State(self.root_dir, self.dvc_dir) self.config = Config(self.dvc_dir) self.logger = Logger(self.config._config) self.cloud = DataCloud(self.cache, self.config._config) @staticmethod def init(root_dir=os.curdir): """ Initiate dvc project in directory. Args: root_dir: Path to project's root directory. Returns: Project instance. Raises: KeyError: Raises an exception. """ root_dir = os.path.abspath(root_dir) dvc_dir = os.path.join(root_dir, Project.DVC_DIR) os.mkdir(dvc_dir) config = Config.init(dvc_dir) cache = Cache.init(dvc_dir) state = State.init(root_dir, dvc_dir) lock = Lock(dvc_dir) scm = SCM(root_dir) scm.ignore_list([cache.cache_dir, state.state_file, lock.lock_file]) ignore_file = os.path.join(dvc_dir, scm.ignore_file()) scm.add([config.config_file, ignore_file]) return Project(root_dir) def to_dvc_path(self, path): return os.path.relpath(path, self.root_dir) def add(self, fname): out = os.path.basename(fname) stage_fname = out + Stage.STAGE_FILE_SUFFIX cwd = os.path.dirname(os.path.abspath(fname)) stage = Stage.loads(project=self, cmd=None, deps=[], outs=[out], fname=stage_fname, cwd=cwd) stage.save() stage.dump() return stage def remove(self, target): if not Stage.is_stage_file(target): raise StageNotFoundError(target) stage = Stage.load(self, target) for out in stage.outs: out.remove() return stage def run(self, cmd=None, deps=[], outs=[], outs_no_cache=[], fname=Stage.STAGE_FILE, cwd=os.curdir, no_exec=False): stage = Stage.loads(project=self, fname=fname, cmd=cmd, cwd=cwd, outs=outs, outs_no_cache=outs_no_cache, deps=deps) if not no_exec: stage.run() stage.dump() return stage def _reproduce_stage(self, stages, node, force): if not stages[node].changed(): return [] stages[node].reproduce(force=force) stages[node].dump() return [stages[node]] def reproduce(self, target, recursive=True, force=False): stages = nx.get_node_attributes(self.graph(), 'stage') node = os.path.relpath(os.path.abspath(target), self.root_dir) if node not in stages: raise StageNotFoundError(target) if recursive: return self._reproduce_stages(stages, node, force) return self._reproduce_stage(stages, node, force) def _reproduce_stages(self, stages, node, force): result = [] for n in nx.dfs_postorder_nodes(self.graph(), node): try: result += self._reproduce_stage(stages, n, force) except Exception as ex: raise ReproductionError(stages[n].relpath, ex) return result def _remove_untracked_hardlinks(self): untracked = self.scm.untracked_files() cache = dict((System.inode(c), c) for c in self.cache.all()) for file in untracked: inode = System.inode(file) if inode not in cache.keys(): continue Logger.info(u'Remove \'{}\''.format(file)) os.remove(file) dir = os.path.dirname(file) if len(dir) != 0 and not os.listdir(dir): Logger.info(u'Remove empty directory \'{}\''.format(dir)) os.removedirs(dir) def checkout(self): self._remove_untracked_hardlinks() for stage in self.stages(): stage.checkout() def _used_cache(self, target=None): cache_set = set() if target: stages = [Stage.load(self, target)] else: stages = self.stages() for stage in stages: for out in stage.outs: if not out.use_cache: continue cache_set |= set([out.cache]) if out.is_dir_cache(out.cache) and os.path.isfile(out.cache): dir_cache = out.dir_cache() cache_set |= set(dir_cache.values()) return list(cache_set) def gc(self): clist = self._used_cache() for cache in self.cache.all(): if cache in clist: continue os.unlink(cache) self.logger.info(u'\'{}\' was removed'.format( self.to_dvc_path(cache))) def push(self, target=None, jobs=1): return self.cloud.push(self._used_cache(target), jobs) def fetch(self, target=None, jobs=1): return self.cloud.pull(self._used_cache(target), jobs) def pull(self, target=None, jobs=1): ret = self.fetch(target, jobs) self.checkout() return ret def _local_status(self, target=None): status = {} if target: stages = [Stage.load(self, target)] else: stages = self.stages() for stage in self.stages(): status.update(stage.status()) return status def _cloud_status(self, target=None, jobs=1): status = {} for target, ret in self.cloud.status(self._used_cache(target), jobs): if ret == cloud.STATUS_UNKNOWN or ret == cloud.STATUS_OK: continue prefix_map = { cloud.STATUS_DELETED: 'deleted', cloud.STATUS_MODIFIED: 'modified', cloud.STATUS_NEW: 'new', } path = os.path.relpath(target, self.cache.cache_dir) status[path] = prefix_map[ret] return status def status(self, target=None, jobs=1, cloud=False): if cloud: return self._cloud_status(target, jobs) return self._local_status(target) def graph(self): G = nx.DiGraph() for stage in self.stages(): node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) for dep in stage.deps: dep_stage = dep.stage() if not dep_stage: continue dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) return G def stages(self): stages = [] for root, dirs, files in os.walk(self.root_dir): for fname in files: path = os.path.join(root, fname) if not Stage.is_stage_file(path): continue stages.append(Stage.load(self, path)) return stages def outs(self): outs = [] for stage in self.stages(): outs += stage.outs return outs
def test_git_submodule(self): self.assertIsInstance(SCM(os.curdir), Git)
def __init__( self, root_dir=None, scm=None, rev=None, subrepos=False, uninitialized=False, ): from dvc.cache import Cache from dvc.data_cloud import DataCloud from dvc.lock import make_lock from dvc.repo.experiments import Experiments from dvc.repo.metrics import Metrics from dvc.repo.params import Params from dvc.repo.plots import Plots from dvc.scm import SCM from dvc.stage.cache import StageCache from dvc.state import State, StateNoop from dvc.tree.local import LocalTree from dvc.utils.fs import makedirs if scm: tree = scm.get_tree(rev) self.root_dir = self.find_root(root_dir, tree) self.scm = scm self.tree = scm.get_tree(rev, use_dvcignore=True, dvcignore_root=self.root_dir) self.state = StateNoop() else: try: root_dir = self.find_root(root_dir) except NotDvcRepoError: if not uninitialized: raise root_dir = SCM(root_dir or os.curdir).root_dir self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.tree = LocalTree( self, {"url": self.root_dir}, use_dvcignore=True, dvcignore_root=self.root_dir, ) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir, tree=self.tree) if not scm: no_scm = self.config["core"].get("no_scm", False) self.scm = SCM(self.root_dir, no_scm=no_scm) self.tmp_dir = os.path.join(self.dvc_dir, "tmp") makedirs(self.tmp_dir, exist_ok=True) hardlink_lock = self.config["core"].get("hardlink_lock", False) self.lock = make_lock( os.path.join(self.tmp_dir, "lock"), tmp_dir=self.tmp_dir, hardlink_lock=hardlink_lock, friendly=True, ) # used by RepoTree to determine if it should traverse subrepos self.subrepos = subrepos self.cache = Cache(self) self.cloud = DataCloud(self) if not scm: # NOTE: storing state and link_state in the repository itself to # avoid any possible state corruption in 'shared cache dir' # scenario. self.state = State(self.cache.local) self.stage_cache = StageCache(self) self.metrics = Metrics(self) self.plots = Plots(self) self.params = Params(self) try: self.experiments = Experiments(self) except NotImplementedError: self.experiments = None self._ignore()
class Project(object): DVC_DIR = '.dvc' def __init__(self, root_dir): self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir) self.lock = Lock(self.dvc_dir) self.link_state = LinkState(self.root_dir, self.dvc_dir) self.logger = Logger(self.config._config[Config.SECTION_CORE].get( Config.SECTION_CORE_LOGLEVEL, None)) self.cache = Cache(self) self.cloud = DataCloud(self, config=self.config._config) self.updater = Updater(self.dvc_dir) self._ignore() self.updater.check() @staticmethod def init(root_dir=os.curdir, no_scm=False): """ Initiate dvc project in directory. Args: root_dir: Path to project's root directory. Returns: Project instance. Raises: KeyError: Raises an exception. """ root_dir = os.path.abspath(root_dir) dvc_dir = os.path.join(root_dir, Project.DVC_DIR) scm = SCM(root_dir) if type(scm) == Base and not no_scm: msg = "{} is not tracked by any supported scm tool(e.g. git).".format( root_dir) raise InitError(msg) os.mkdir(dvc_dir) config = Config.init(dvc_dir) proj = Project(root_dir) scm.add([config.config_file]) if scm.ignore_file(): scm.add([os.path.join(dvc_dir, scm.ignore_file())]) return proj def _ignore(self): l = [ self.link_state.state_file, self.link_state._lock_file.lock_file, self.lock.lock_file, self.config.config_local_file, self.updater.updater_file ] if self.cache.local.cache_dir.startswith(self.root_dir): l += [self.cache.local.cache_dir] self.scm.ignore_list(l) def install(self): self.scm.install() def to_dvc_path(self, path): return os.path.relpath(path, self.root_dir) def add(self, fname): out = os.path.basename(os.path.normpath(fname)) stage_fname = out + Stage.STAGE_FILE_SUFFIX cwd = os.path.dirname(os.path.abspath(fname)) stage = Stage.loads(project=self, cmd=None, deps=[], outs=[out], fname=stage_fname, cwd=cwd) stage.save() stage.dump() return stage def remove(self, target, outs_only=False): if not Stage.is_stage_file(target): raise NotDvcFileError(target) stage = Stage.load(self, target) if outs_only: stage.remove_outs() else: stage.remove() return stage def lock_stage(self, target, unlock=False): if not Stage.is_stage_file(target): raise NotDvcFileError(target) stage = Stage.load(self, target) stage.locked = False if unlock else True stage.dump() return stage def move(self, from_path, to_path): from_out = Output.loads_from(Stage(self, cwd=os.curdir), [from_path])[0] found = False for stage in self.stages(): for out in stage.outs: if out.path != from_out.path: continue if not stage.is_data_source: raise DvcException( 'Dvcfile \'{}\' is not a data source.'.format( stage.rel_path)) found = True to_out = Output.loads_from(out.stage, [to_path], out.cache, out.metric)[0] out.move(to_out) stage_base = os.path.basename(stage.path).rstrip( Stage.STAGE_FILE_SUFFIX) stage_dir = os.path.dirname(stage.path) from_base = os.path.basename(from_path) to_base = os.path.basename(to_path) if stage_base == from_base: os.unlink(stage.path) stage.path = os.path.join( stage_dir, to_base + Stage.STAGE_FILE_SUFFIX) stage.dump() if not found: raise DvcException( 'Unable to find dvcfile with output \'{}\''.format(from_path)) def run(self, cmd=None, deps=[], outs=[], outs_no_cache=[], metrics_no_cache=[], fname=Stage.STAGE_FILE, cwd=os.curdir, no_exec=False): stage = Stage.loads(project=self, fname=fname, cmd=cmd, cwd=cwd, outs=outs, outs_no_cache=outs_no_cache, metrics_no_cache=metrics_no_cache, deps=deps) if not no_exec: stage.run() stage.dump() return stage def imp(self, url, out): stage_fname = out + Stage.STAGE_FILE_SUFFIX cwd = os.path.dirname(os.path.abspath(out)) stage = Stage.loads(project=self, cmd=None, deps=[url], outs=[out], fname=stage_fname, cwd=cwd) stage.run() stage.dump() return stage def _reproduce_stage(self, stages, node, force): stage = stages[node].reproduce(force=force) if not stage: return [] stage.dump() return [stage] def reproduce(self, target, recursive=True, force=False): stages = nx.get_node_attributes(self.graph(), 'stage') node = os.path.relpath(os.path.abspath(target), self.root_dir) if node not in stages: raise NotDvcFileError(target) if recursive: return self._reproduce_stages(stages, node, force) return self._reproduce_stage(stages, node, force) def _reproduce_stages(self, stages, node, force): result = [] for n in nx.dfs_postorder_nodes(self.graph(), node): try: result += self._reproduce_stage(stages, n, force) except Exception as ex: raise ReproductionError(stages[n].relpath, ex) return result def checkout(self, target=None): if target: if not Stage.is_stage_file(target): raise NotDvcFileError(target) stages = [Stage.load(self, target)] else: self.link_state.remove_all() stages = self.stages() for stage in stages: stage.checkout() def _used_cache(self, target=None, all_branches=False): cache = {} cache['local'] = [] cache['s3'] = [] cache['gs'] = [] cache['hdfs'] = [] cache['ssh'] = [] for branch in self.scm.brancher(all_branches=all_branches): if target: stages = [Stage.load(self, target)] else: stages = self.stages() for stage in stages: for out in stage.outs: if not out.use_cache or not out.info: continue cache[out.path_info['scheme']] += [out.info] return cache def gc(self, all_branches=False): clist = self._used_cache(target=None, all_branches=all_branches) self.cache.local.gc(clist['local']) if self.cache.s3: self.cache.s3.gc(clist['s3']) if self.cache.gs: self.cache.gs.gc(clist['gs']) if self.cache.ssh: self.cache.ssh.gc(clist['ssh']) if self.cache.hdfs: self.cache.hdfs.gc(clist['hdfs']) def push(self, target=None, jobs=1, remote=None, all_branches=False): self.cloud.push(self._used_cache(target, all_branches)['local'], jobs, remote=remote) def fetch(self, target=None, jobs=1, remote=None, all_branches=False): self.cloud.pull(self._used_cache(target, all_branches)['local'], jobs, remote=remote) def pull(self, target=None, jobs=1, remote=None, all_branches=False): self.fetch(target, jobs, remote=remote, all_branches=all_branches) self.checkout(target=target) def _local_status(self, target=None): status = {} if target: stages = [Stage.load(self, target)] else: stages = self.stages() for stage in stages: status.update(stage.status()) return status def _cloud_status(self, target=None, jobs=1, remote=None): import dvc.remote.base as cloud status = {} for md5, ret in self.cloud.status(self._used_cache(target)['local'], jobs, remote=remote): if ret == cloud.STATUS_OK: continue prefix_map = { cloud.STATUS_DELETED: 'deleted', cloud.STATUS_NEW: 'new', } status[md5] = prefix_map[ret] return status def status(self, target=None, jobs=1, cloud=False, remote=None): if cloud: return self._cloud_status(target, jobs, remote=remote) return self._local_status(target) def _read_metric_json(self, fd, json_path): parser = parse(json_path) return [x.value for x in parser.find(json.load(fd))] def _do_read_metric_tsv(self, reader, row, col): if col != None and row != None: return [reader[row][col]] elif col != None: return [r[col] for r in reader] elif row != None: return reader[row] return None def _read_metric_htsv(self, fd, htsv_path): col, row = htsv_path.split(',') row = int(row) reader = list(csv.DictReader(fd, delimiter='\t')) return self._do_read_metric_tsv(reader, row, col) def _read_metric_tsv(self, fd, tsv_path): col, row = tsv_path.split(',') row = int(row) col = int(col) reader = list(csv.reader(fd, delimiter='\t')) return self._do_read_metric_tsv(reader, row, col) def _read_metric(self, path, json_path=None, tsv_path=None, htsv_path=None): ret = None if not os.path.exists(path): return ret try: with open(path, 'r') as fd: if json_path: ret = self._read_metric_json(fd, json_path) elif tsv_path: ret = self._read_metric_tsv(fd, tsv_path) elif htsv_path: ret = self._read_metric_htsv(fd, htsv_path) else: ret = fd.read() except Exception as exc: self.logger.debug('Unable to read metric in \'{}\''.format(path), exc) return ret def metrics_show(self, path=None, json_path=None, tsv_path=None, htsv_path=None, all_branches=False): res = {} for branch in self.scm.brancher(all_branches=all_branches): metrics = filter(lambda o: o.metric, self.outs()) fnames = [path] if path else map(lambda o: o.path, metrics) for fname in fnames: rel = os.path.relpath(fname) metric = self._read_metric(fname, json_path=json_path, tsv_path=tsv_path, htsv_path=htsv_path) if not metric: continue if branch not in res: res[branch] = {} res[branch][rel] = metric for branch, val in res.items(): self.logger.info('{}:'.format(branch)) for fname, metric in val.items(): self.logger.info('\t{}: {}'.format(fname, metric)) return res def _metrics_modify(self, path, val): apath = os.path.abspath(path) for stage in self.stages(): for out in stage.outs: if apath != out.path: continue if out.path_info['scheme'] != 'local': msg = 'Output \'{}\' scheme \'{}\' is not supported for metrics' raise DvcException( msg.format(out.path, out.path_info['scheme'])) if out.use_cache: msg = 'Cached output \'{}\' is not supported for metrics' raise DvcException(msg.format(out.rel_path)) out.metric = val stage.dump() def metrics_add(self, path): self._metrics_modify(path, True) def metrics_remove(self, path): self._metrics_modify(path, False) def graph(self): G = nx.DiGraph() stages = self.stages() outs = self.outs() for stage in stages: node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) if stage.locked: continue for dep in stage.deps: for out in outs: if out.path != dep.path and not dep.path.startswith( out.path + out.sep): continue dep_stage = out.stage dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) return G def stages(self): stages = [] for root, dirs, files in os.walk(self.root_dir): for fname in files: path = os.path.join(root, fname) if not Stage.is_stage_file(path): continue stages.append(Stage.load(self, path)) return stages def outs(self): outs = [] for stage in self.stages(): outs += stage.outs return outs
def init(root_dir=os.curdir, no_scm=False, force=False, subdir=False): """ Creates an empty repo on the given directory -- basically a `.dvc` directory with subdirectories for configuration and cache. It should be tracked by a SCM or use the `--no-scm` flag. If the given directory is not empty, you must use the `--force` flag to override it. Args: root_dir: Path to repo's root directory. Returns: Repo instance. Raises: KeyError: Raises an exception. """ if no_scm and subdir: raise InvalidArgumentError( "Cannot initialize repo with `--no-scm` and `--subdir`") root_dir = os.path.realpath(root_dir) dvc_dir = os.path.join(root_dir, Repo.DVC_DIR) try: scm = SCM(root_dir, search_parent_directories=subdir, no_scm=no_scm) except SCMError: raise InitError( "{repo} is not tracked by any supported SCM tool (e.g. Git). " "Use `--no-scm` if you don't want to use any SCM or " "`--subdir` if initializing inside a subdirectory of a parent SCM " "repository.".format(repo=root_dir)) if os.path.isdir(dvc_dir): if not force: raise InitError("'{repo}' exists. Use `-f` to force.".format( repo=relpath(dvc_dir))) remove(dvc_dir) os.mkdir(dvc_dir) config = Config.init(dvc_dir) if no_scm: with config.edit() as conf: conf["core"]["no_scm"] = True dvcignore = init_dvcignore(root_dir) proj = Repo(root_dir) scm.add( [config.files["repo"], dvcignore, proj.plot_templates.templates_dir]) if scm.ignore_file: scm.add([os.path.join(dvc_dir, scm.ignore_file)]) logger.info("\nYou can now commit the changes to git.\n") _welcome_message() return proj
def scm(self): from dvc.scm import SCM no_scm = self.config["core"].get("no_scm", False) return self._scm if self._scm else SCM(self.root_dir, no_scm=no_scm)
class Project(object): DVC_DIR = '.dvc' def __init__(self, root_dir): self.root_dir = os.path.abspath(os.path.realpath(root_dir)) self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR) self.config = Config(self.dvc_dir) self.scm = SCM(self.root_dir) self.lock = Lock(self.dvc_dir) self.link_state = LinkState(self.root_dir, self.dvc_dir) self.logger = Logger(self.config._config[Config.SECTION_CORE].get( Config.SECTION_CORE_LOGLEVEL, None)) self.cache = Cache(self) self.cloud = DataCloud(cache=self.cache, config=self.config._config) self.updater = Updater(self.dvc_dir) self._ignore() self.updater.check() @staticmethod def init(root_dir=os.curdir, no_scm=False): """ Initiate dvc project in directory. Args: root_dir: Path to project's root directory. Returns: Project instance. Raises: KeyError: Raises an exception. """ root_dir = os.path.abspath(root_dir) dvc_dir = os.path.join(root_dir, Project.DVC_DIR) scm = SCM(root_dir) if type(scm) == Base and not no_scm: msg = "{} is not tracked by any supported scm tool(e.g. git).".format( root_dir) raise InitError(msg) os.mkdir(dvc_dir) config = Config.init(dvc_dir) proj = Project(root_dir) scm.add([config.config_file]) if scm.ignore_file(): scm.add([os.path.join(dvc_dir, scm.ignore_file())]) return proj def _ignore(self): l = [ self.link_state.state_file, self.link_state._lock_file.lock_file, self.lock.lock_file, self.config.config_local_file, self.updater.updater_file ] if self.cache.local.cache_dir.startswith(self.root_dir): l += [self.cache.local.cache_dir] self.scm.ignore_list(l) def install(self): self.scm.install() def to_dvc_path(self, path): return os.path.relpath(path, self.root_dir) def add(self, fname): out = os.path.basename(os.path.normpath(fname)) stage_fname = out + Stage.STAGE_FILE_SUFFIX cwd = os.path.dirname(os.path.abspath(fname)) stage = Stage.loads(project=self, cmd=None, deps=[], outs=[out], fname=stage_fname, cwd=cwd) stage.save() stage.dump() return stage def remove(self, target): if not Stage.is_stage_file(target): raise StageNotFoundError(target) stage = Stage.load(self, target) for out in stage.outs: out.remove() return stage def run(self, cmd=None, deps=[], outs=[], outs_no_cache=[], fname=Stage.STAGE_FILE, cwd=os.curdir, no_exec=False): stage = Stage.loads(project=self, fname=fname, cmd=cmd, cwd=cwd, outs=outs, outs_no_cache=outs_no_cache, deps=deps) if not no_exec: stage.run() stage.dump() return stage def _reproduce_stage(self, stages, node, force): stage = stages[node].reproduce(force=force) if not stage: return [] stage.dump() return [stage] def reproduce(self, target, recursive=True, force=False): stages = nx.get_node_attributes(self.graph(), 'stage') node = os.path.relpath(os.path.abspath(target), self.root_dir) if node not in stages: raise StageNotFoundError(target) if recursive: return self._reproduce_stages(stages, node, force) return self._reproduce_stage(stages, node, force) def _reproduce_stages(self, stages, node, force): result = [] for n in nx.dfs_postorder_nodes(self.graph(), node): try: result += self._reproduce_stage(stages, n, force) except Exception as ex: raise ReproductionError(stages[n].relpath, ex) return result def checkout(self, target=None): if target: if not Stage.is_stage_file(target): raise StageNotFoundError(target) stages = [Stage.load(self, target)] else: self.link_state.remove_all() stages = self.stages() for stage in stages: stage.checkout() def _used_cache(self, target=None): cache_set = set() if target: stages = [Stage.load(self, target)] else: stages = self.stages() for stage in stages: for out in stage.outs: if out.path_info['scheme'] != 'local': continue if not out.use_cache or not out.cache: continue cache_set |= set([out.cache]) if self.cache.local.is_dir_cache(out.cache) and os.path.isfile( out.cache): dir_cache = self.cache.local.dir_cache(out.cache) cache_set |= set(dir_cache.values()) return list(cache_set) def gc(self): clist = self._used_cache() for cache in self.cache.local.all(): if cache in clist: continue os.unlink(cache) self.logger.info(u'\'{}\' was removed'.format( self.to_dvc_path(cache))) def push(self, target=None, jobs=1, remote=None): return self.cloud.push(self._used_cache(target), jobs, remote=remote) def fetch(self, target=None, jobs=1, remote=None): return self.cloud.pull(self._used_cache(target), jobs, remote=remote) def pull(self, target=None, jobs=1, remote=None): ret = self.fetch(target, jobs, remote=remote) self.checkout() return ret def _local_status(self, target=None): status = {} if target: stages = [Stage.load(self, target)] else: stages = self.stages() for stage in stages: status.update(stage.status()) return status def _cloud_status(self, target=None, jobs=1, remote=None): status = {} for target, ret in self.cloud.status(self._used_cache(target), jobs, remote=remote): if ret == cloud.STATUS_UNKNOWN or ret == cloud.STATUS_OK: continue prefix_map = { cloud.STATUS_DELETED: 'deleted', cloud.STATUS_MODIFIED: 'modified', cloud.STATUS_NEW: 'new', } path = os.path.relpath(target, self.cache.local.cache_dir) status[path] = prefix_map[ret] return status def status(self, target=None, jobs=1, cloud=False, remote=None): if cloud: return self._cloud_status(target, jobs, remote=remote) return self._local_status(target) def _read_metric_json(self, fd, json_path): parser = parse(json_path) return [x.value for x in parser.find(json.load(fd))] def _do_read_metric_tsv(self, reader, row, col): if col != None and row != None: return [reader[row][col]] elif col != None: return [r[col] for r in reader] elif row != None: return reader[row] return None def _read_metric_htsv(self, fd, htsv_path): col, row = htsv_path.split(',') row = int(row) reader = list(csv.DictReader(fd, delimiter='\t')) return self._do_read_metric_tsv(reader, row, col) def _read_metric_tsv(self, fd, tsv_path): col, row = tsv_path.split(',') row = int(row) col = int(col) reader = list(csv.reader(fd, delimiter='\t')) return self._do_read_metric_tsv(reader, row, col) def _read_metric(self, path, json_path=None, tsv_path=None, htsv_path=None): ret = None try: with open(path, 'r') as fd: if json_path: ret = self._read_metric_json(fd, json_path) elif tsv_path: ret = self._read_metric_tsv(fd, tsv_path) elif htsv_path: ret = self._read_metric_htsv(fd, htsv_path) else: ret = fd.read() except Exception as exc: self.logger.error('Unable to read metric in \'{}\''.format(path), exc) return ret def metrics(self, path, json_path=None, tsv_path=None, htsv_path=None): res = {} saved = self.scm.active_branch() for branch in self.scm.list_branches(): self.scm.checkout(branch) self.checkout() res[branch] = self._read_metric(path, json_path=json_path, tsv_path=tsv_path, htsv_path=htsv_path) self.scm.checkout(saved) self.checkout() return res def graph(self): G = nx.DiGraph() stages = self.stages() outs_map = {} for stage in stages: for o in stage.outs: outs_map[o.path] = stage for stage in stages: node = os.path.relpath(stage.path, self.root_dir) G.add_node(node, stage=stage) for dep in stage.deps: dep_stage = outs_map.get(dep.path, None) if not dep_stage: continue dep_node = os.path.relpath(dep_stage.path, self.root_dir) G.add_node(dep_node, stage=dep_stage) G.add_edge(node, dep_node) return G def stages(self): stages = [] for root, dirs, files in os.walk(self.root_dir): for fname in files: path = os.path.join(root, fname) if not Stage.is_stage_file(path): continue stages.append(Stage.load(self, path)) return stages def outs(self): outs = [] for stage in self.stages(): outs += stage.outs return outs