def deduplicate(log, *subpath, timeout=None): log.info('locking the deduplication db...') with lock.Lock(path.saviour('.duperemove.hashfile-lock'), timeout=timeout): log.info('deduplicating...') run = log.pipe_powered(subprocess.run, stdout=logging.INFO, stderr=logging.WARNING) r = run(['duperemove', '--hashfile', path.saviour('.duperemove.hashfile'), '-hdr', path.saviour('_', *subpath)]) assert r.returncode in (0, 22) # nothing to deduplicate
def wip(normal_path, makedirs=False): norm, wip, lock = normal_path, normal_path + '-WIP', normal_path + '-lock' with util_lock.Lock(lock): if os.path.exists(norm): assert not os.path.exists(wip), f'Both {norm} and {wip} exist' os.rename(norm, wip) if makedirs: os.makedirs(os.path.dirname(wip.rstrip(os.path.sep)), exist_ok=True) yield wip os.rename(wip, norm) os.unlink(lock)
def build(first_step, *args, fingertip_last_step=False, **kwargs): func, tag = step_loader.func_and_autotag(first_step, *args, **kwargs) # Could there already be a cached result? mpath = path.machines(tag) lock_path = path.machines('.' + tag + '-lock') log.info(f'acquiring lock for {tag}...') transient_hint = func.transient if hasattr(func, 'transient') else None if callable(transient_hint): transient_hint = supply_last_step_if_requested(transient_hint, fingertip_last_step) transient_hint = transient_hint(*args, **kwargs) transient = (transient_hint in ('always', True) or transient_hint == 'last' and fingertip_last_step) with lock.Lock(lock_path) if not transient else lock.NoLock(): if not os.path.exists(mpath) or needs_a_rebuild(mpath): log.info(f'building {tag}...') func = supply_last_step_if_requested(func, fingertip_last_step) first = func(*args, **kwargs) if first is None: assert transient, 'first step returned None' return if transient: log.info(f'succesfully built and discarded {tag}') first._finalize() # discard (not fast-dropped though) if transient_hint == 'last' and fingertip_last_step: fname = f'{datetime.datetime.utcnow().isoformat()}.txt' t = path.logs(fname, makedirs=True) with open(t, 'w') as f: f.write(first.log_contents) return t else: log.info(f'succesfully built and saved {tag}') first._finalize(link_as=mpath, name_hint=tag) if fingertip_last_step: return os.path.join(mpath, 'log.txt') m = clone_and_load(mpath) m.log = log.Sublogger('fingertip.<just built>', os.path.join(m.path, 'log.txt')) return m
def _deduplicate(log, db_name, resource_name, timeout=None): log.info(f'locking the deduplication db {db_name}...') hashfilesdir = path.saviour('.duperemove', 'hashfiles') if not os.path.exists(hashfilesdir): os.makedirs(hashfilesdir) os.system(f'chattr +C {hashfilesdir} || true') db_file = path.saviour('.duperemove', 'hashfiles', db_name) db_lock = path.saviour('.duperemove', 'locks', db_name, makedirs=True) with lock.Lock(db_lock, timeout=timeout): log.info(f'deduplicating {resource_name} ({db_name})...') run = log.pipe_powered(subprocess.run, stdout=logging.INFO, stderr=logging.WARNING) r = run([ 'duperemove', '--dedupe-options=nofiemap', '--io-threads=2', '--cpu-threads=2', '--hashfile', db_file, '-hdr', path.saviour('_', resource_name, 'data') ]) assert r.returncode in (0, 22) # nothing to deduplicate
def mirror(config, *what_to_mirror, deduplicate=None): total_failures = [] failures = collections.defaultdict(list) with open(config) as f: config = ruamel.yaml.YAML(typ='safe').load(f) if 'mirror' in config and not config['mirror']: log.warning('mirroring is disabled in config') return hows, whats = config['how'], config['what'] if not what_to_mirror: what_to_mirror = whats.keys() else: what_to_mirror = ([ k for k in whats.keys() if any( fnmatch.fnmatch(k, req) for req in what_to_mirror) ] + [k for k in what_to_mirror if '=' in k]) if not what_to_mirror: log.error('nothing to mirror') return for resource in what_to_mirror: log.debug(f'processing {resource}...') if '=' not in resource: # example: alpine-3.13=alpine/v3.13/main/x86 resource_name, tail = resource, '' s = whats[resource_name] else: # example: alpine-3.13=alpine/v3.13/main/x86 resource_name, s = resource.split('=', 1) # FIXME UGLY: config overrides are stronger that = (more syntax?) # TODO: whats shouldn't be a dict, I think, just a list of strings if resource_name in whats: s = whats[resource_name] if s is None: s = resource_name if '/' in s: how_name, suffix = s.split('/', 1) suffix = '/' + suffix else: how_name, suffix = s, '' try: how = hows[how_name] except KeyError: log.error(f'missing how section on {how_name}') raise SystemExit() url = how['url'] + suffix method = how['method'] sources = (how['sources'] if 'sources' in how else [how['url']]) sources = [s + suffix for s in sources] extra_args = { k: v for k, v in how.items() if k not in ('url', 'sources', 'method', 'validate', 'deduplicate') } if f'method_{method}' not in globals(): log.error(f'unsupported method {method}') raise SystemExit() meth = globals()[f'method_{method}'] symlink = path.saviour(url.rstrip('/')) # usually symlink points to data, but while we're working on it, # it temporarily points to a consistent snapshot of it named `snap` data = os.path.realpath(path.saviour('_', resource_name, 'data')) snap = os.path.realpath(path.saviour('_', resource_name, 'snap')) temp = os.path.realpath(path.saviour('_', resource_name, 'temp')) lockfile = path.saviour('_', resource_name) + '-lock' assert data.startswith(os.path.realpath(path.SAVIOUR)) assert snap.startswith(os.path.realpath(path.SAVIOUR)) assert temp.startswith(os.path.realpath(path.SAVIOUR)) sublog = log.Sublogger(f'{method} {resource_name}') sublog.info('locking...') with lock.Lock(lockfile): os.makedirs(os.path.dirname(snap), exist_ok=True) if os.path.exists(temp): sublog.info('removing stale temp...') _remove(temp) if os.path.exists(symlink): # it's already published if os.path.exists(data) and not os.path.exists(snap): # `data` is present and is the best we have to publish sublog.info('snapshotting...') reflink.always(data, temp, preserve=True) os.rename(temp, snap) if os.path.exists(snap): # link to a consistent snapshot while we work on `data` _symlink(snap, symlink) for source in sources: sublog.info(f'trying {source}...') try: meth(sublog, source, snap, data, **extra_args) assert os.path.exists(data) if 'validate' in how: sublog.info(f'validating with {how["validate"]}...') validator = globals()[f'validate_{how["validate"]}'] validator(sublog, source, data) sublog.info('validated') break except Exception as _: traceback.print_exc() failures[resource_name].append(source) fingertip.util.log.warning(f'failed to mirror {source}') if len(failures[resource_name]) == len(sources): sublog.error(f'failed to mirror ' f'from all {len(sources)} sources') total_failures.append(resource_name) continue _symlink(data, symlink) if os.path.exists(snap): os.rename(snap, temp) # move it out the way asap sublog.info('removing now obsolete snapshot...') _remove(temp) how_deduplicate = how.get('deduplicate', True) db_name = how_deduplicate if how_deduplicate is not True else how_name if how_deduplicate and deduplicate is not False: try: _deduplicate(sublog, db_name, resource_name, timeout=1) except lock.LockTimeout: log.warning(f'skipped deduplication of {resource_name}, ' f'db {db_name} was locked') if total_failures: fingertip.util.log.error(f'failed: {", ".join(total_failures)}') raise FailureToMirrorError(", ".join(total_failures)) log.info('saviour has completed mirroring')
def __init__(self, url, *path_components, enough_to_have=None): if not path_components: path_components = [url.replace('/', '::')] self.url = url cache_path = path.downloads('git', *path_components, makedirs=True) self.path = temp.disappearing_dir(os.path.dirname(cache_path), path_components[-1]) lock_working_copy_path = self.path + '-lock' lock_cache_path = cache_path + '-lock' lock.Lock.__init__(self, lock_working_copy_path) sources = saviour_sources() self.self_destruct = False with lock.Lock(lock_cache_path), lock.Lock(lock_working_copy_path): cache_is_enough = False if os.path.exists(cache_path): try: cr = git.Repo(cache_path) cache_is_enough = (enough_to_have and _has_rev(cr, enough_to_have)) except git.GitError as e: log.error(f'something wrong with git cache {cache_path}') log.error(str(e)) _remove(self.path) for i, (source, cache) in enumerate(sources): last_source = i == len(sources) - 1 if cache and cache_is_enough: log.info(f'not re-fetching {url} from {source} ' f'because {enough_to_have} ' 'is already present in cache') git.Repo.clone_from(cache_path, self.path, mirror=True) break if source == 'local': surl = path.saviour(url).replace('//', '/') # workaround if not os.path.exists(surl) and not last_source: continue log.info(f'cloning {url} from local saviour mirror') git.Repo.clone_from(surl, self.path, mirror=True) break elif source == 'direct': surl = url else: surl = source + '/' + url surl = 'http://' + surl if '://' not in source else surl log.info(f'cloning {url} from {source} ' f'cache_exists={os.path.exists(cache_path)}...') try: # TODO: bare clone # no harm in referencing cache, even w/o cached+ git.Repo.clone_from(surl, self.path, mirror=True, dissociate=True, reference_if_able=cache_path) except git.GitError: log.warning(f'could not clone {url} from {source}') if last_source: raise continue break _remove(cache_path) reflink.auto(self.path, cache_path) git.Repo.__init__(self, self.path) self.remotes[0].set_url(url) self.self_destruct = True
def _cache_aware_apply(self, step, tag, func, args, kwargs, last_step): assert self._state == 'loaded' transient_hint = func.transient if hasattr(func, 'transient') else None if callable(transient_hint): transient_hint = supply_last_step_if_requested( transient_hint, last_step) transient_hint = transient_hint(self, *args, **kwargs) return_as_transient = self._transient exec_as_transient = (transient_hint in ('always', True) or transient_hint == 'last' and last_step) log.debug(f'transient: {transient_hint}') log.debug(f'exec_as_transient: {exec_as_transient}') log.debug(f'return_as_transient: {return_as_transient}') self._transient = exec_as_transient # Could there already be a cached result? log.debug(f'PATH {self.path} {tag}') new_mpath = os.path.join(self._parent_path, tag) lock_path = os.path.join(self._parent_path, '.' + tag + '-lock') do_lock = not self._transient if do_lock: log.info(f'acquiring lock for {tag}...') prev_log_name = self.log.name self.log.finalize() with lock.Lock(lock_path) if do_lock else lock.NoLock(): if (os.path.exists(new_mpath) and not needs_a_rebuild(new_mpath) and not exec_as_transient): # sweet, scratch this instance, fast-forward to cached result log.info(f'reusing {step} @ {new_mpath}') self._finalize() clone_from_path = new_mpath else: # loaded, not spun up, step not cached: perform step, cache log.info(f'applying (and, possibly, caching) {tag}') self.log = log.Sublogger('plugins.' + tag.split(':', 1)[0], os.path.join(self.path, 'log.txt')) func = supply_last_step_if_requested(func, last_step) m = func(self, *args, **kwargs) if m: if m._transient and transient_hint == 'last' and last_step: assert m._state == 'dropped' # transient-when-last step returned m # just in case it's not the last, but it was. # m is dropped already, only log contents is preserved. fname = f'{datetime.datetime.utcnow().isoformat()}.txt' t = path.logs(fname, makedirs=True) with open(t, 'w') as f: f.write(m.log_contents) return t assert not m._transient, 'transient step returned a value' m._finalize(link_as=new_mpath, name_hint=tag) clone_from_path = new_mpath log.info(f'successfully applied and saved {tag}') else: # transient step, either had hints or just returned None clone_from_path = self._parent_path log.info(f'successfully applied and dropped {tag}') if last_step: return os.path.join(clone_from_path, 'log.txt') m = clone_and_load(clone_from_path) m.log = log.Sublogger(prev_log_name, os.path.join(m.path, 'log.txt')) m._transient = return_as_transient return m
def mirror(config, *what_to_mirror): total_failures = [] failures = collections.defaultdict(list) with open(config) as f: config = ruamel.yaml.YAML(typ='safe').load(f) hows, whats = config['how'], config['what'] if not what_to_mirror: what_to_mirror = whats.keys() else: what_to_mirror = [k for k in whats.keys() if any((fnmatch.fnmatch(k, req) for req in what_to_mirror))] for resource_name in what_to_mirror or whats.keys(): s = whats[resource_name] log.debug(f'processing {resource_name}...') if s is None: how, suffix = resource_name, '' elif '/' in s: how, suffix = s.split('/', 1) suffix = '/' + suffix else: how, suffix = s, '' try: how = hows[how] except KeyError: log.error(f'missing how section on {how}') raise SystemExit() url = how['url'] + suffix method = how['method'] sources = (how['sources'] if 'sources' in how else [how['url']]) sources = [s + suffix for s in sources] extra_args = {k: v for k, v in how.items() if k not in ('url', 'sources', 'method')} if f'method_{method}' not in globals(): log.error(f'unsupported method {method}') raise SystemExit() meth = globals()[f'method_{method}'] symlink = path.saviour(url.rstrip('/')) # usually symlink points to data, but while we're working on it, # it temporarily points to a consistent snapshot of it named `snap` data = path.saviour('_', resource_name, 'data') snap = path.saviour('_', resource_name, 'snap') temp = path.saviour('_', resource_name, 'temp') lockfile = path.saviour('_', resource_name) + '-lock' assert data.startswith(path.SAVIOUR) assert snap.startswith(path.SAVIOUR) assert temp.startswith(path.SAVIOUR) sublog = log.Sublogger(f'{method} {resource_name}') sublog.info('locking...') with lock.Lock(lockfile): os.makedirs(os.path.dirname(snap), exist_ok=True) if os.path.exists(temp): sublog.info('removing stale temp...') _remove(temp) if os.path.exists(symlink): # it's already published if os.path.exists(data) and not os.path.exists(snap): # `data` is present and is the best we have to publish sublog.info('snapshotting...') reflink.always(data, temp, preserve=True) os.rename(temp, snap) if os.path.exists(snap): # link to a consistent snapshot while we work on `data` _symlink(snap, symlink) for source in sources: sublog.info(f'trying {source}...') try: meth(sublog, source, snap, data, **extra_args) assert os.path.exists(data) break except Exception as _: traceback.print_exc() failures[resource_name].append(source) fingertip.util.log.warning(f'failed to mirror {source}') if len(failures[resource_name]) == len(sources): sublog.error(f'failed to mirror ' f'from all {len(sources)} sources') total_failures.append(resource_name) continue _symlink(data, symlink) if os.path.exists(snap): os.rename(snap, temp) # move it out the way asap sublog.info('removing now obsolete snapshot...') _remove(temp) try: deduplicate(sublog, resource_name, timeout=1) except lock.LockTimeout: log.warning('skipped deduplication, db was locked') if total_failures: fingertip.util.log.error(f'failed: {", ".join(total_failures)}') raise SystemExit() log.info('saviour has completed mirroring')
def __init__(self, url, *path_components, enough_to_have=None): assert path_components self.url = url cache_path = path.downloads('git', *path_components, makedirs=True) cache_exists = os.path.exists(cache_path) self.path = temp.disappearing_dir(os.path.dirname(cache_path), path_components[-1]) lock_working_copy_path = self.path + '-lock' lock_cache_path = cache_path + '-lock' lock.Lock.__init__(self, lock_working_copy_path) update_not_needed = None sources = saviour_sources() self.self_destruct = False with lock.Lock(lock_cache_path), lock.Lock(lock_working_copy_path): _remove(self.path) for i, (source, cache) in enumerate(sources): last_source = i == len(sources) - 1 if cache and cache_exists and update_not_needed is None: cr = git.Repo(cache_path) update_not_needed = enough_to_have and ( enough_to_have in (t.name for t in cr.tags) or enough_to_have in (h.name for h in cr.heads) or enough_to_have in (c.hexsha for c in cr.iter_commits()) # that's not all revspecs, but best-effort is fine ) if update_not_needed: log.info(f'not re-fetching {url} from {source} ' f'because {enough_to_have} ' 'is already present in cache') git.Repo.clone_from(cache_path, self.path, mirror=True) break if source == 'local': surl = path.saviour(url).replace('//', '/') # workaround if not os.path.exists(surl) and not last_source: continue log.info(f'cloning {url} from local saviour mirror') git.Repo.clone_from(surl, self.path, mirror=True) break elif source == 'direct': surl = url else: surl = source + '/' + url surl = 'http://' + surl if '://' not in source else surl log.info(f'cloning {url} from {source} ' f'cache_exists={cache_exists}...') try: # TODO: bare clone # no harm in referencing cache, even w/o cached+ git.Repo.clone_from(surl, self.path, mirror=True, dissociate=True, reference_if_able=cache_path) except git.GitError: log.warning(f'could not clone {url} from {source}') if last_source: raise continue break _remove(cache_path) reflink.auto(self.path, cache_path) git.Repo.__init__(self, self.path) self.remotes[0].set_url(url) self.self_destruct = True