def remove_package(install, repository, package_id):
"""Remove a package from the local repository.
Errors if any packages in package_ids are activated in install.
install: pkgpanda.Install
repository: pkgpanda.Repository
package_id: package ID to remove from repository
"""
if package_id in install.get_active():
raise PackageConflict("Refusing to remove active package {0}".format(package_id))
sys.stdout.write("\rRemoving: {0}".format(package_id))
sys.stdout.flush()
try:
# Validate package id, that package is installed.
PackageId(package_id)
repository.remove(package_id)
except ValidationError as ex:
raise ValidationError("Invalid package id {0}".format(package_id)) from ex
except OSError as ex:
raise Exception("Error removing package {0}: {1}".format(package_id, ex)) from ex
else:
sys.stdout.write("\rRemoved: {0}".format(package_id))
finally:
sys.stdout.write("\n")
sys.stdout.flush()
def check_forbidden_services(path, services):
"""Check if package contains systemd services that may break DC/OS
This functions checks the contents of systemd's unit file dirs and
throws the exception if there are reserved services inside.
Args:
path: path where the package contents are
services: list of reserved services to look for
Raises:
ValidationError: Reserved serice names were found inside the package
"""
services_dir_regexp = re.compile(r'dcos.target.wants(?:_.+)?')
forbidden_srv_set = set(services)
pkg_srv_set = set()
for direntry in os.listdir(path):
if not services_dir_regexp.match(direntry):
continue
pkg_srv_set.update(set(os.listdir(os.path.join(path, direntry))))
found_units = forbidden_srv_set.intersection(pkg_srv_set)
if found_units:
msg = "Reverved unit names found: " + ','.join(found_units)
raise ValidationError(msg)
def checkout_to(self, directory):
# fetch into a bare repository so if we're on a host which has a cache we can
# only get the new commits.
fetch_git(self.bare_folder, self.url)
# Warn if the ref_origin is set and gives a different sha1 than the
# current ref.
try:
origin_commit = get_git_sha1(self.bare_folder, self.ref_origin)
except Exception as ex:
raise ValidationError(
"Unable to find sha1 of ref_origin {}: {}".format(
self.ref_origin, ex))
if self.ref != origin_commit:
logger.warning(
"Current ref doesn't match the ref origin. "
"Package ref should probably be updated to pick up "
"new changes to the code:" +
" Current: {}, Origin: {}".format(self.ref, origin_commit))
# Clone into `src/`.
if is_windows:
# Note: Mesos requires autocrlf to be set on Windows otherwise it does not build.
check_call([
"git", "clone", "-q", "--config", "core.autocrlf=true",
self.bare_folder, directory
])
else:
check_call(["git", "clone", "-q", self.bare_folder, directory])
# Checkout from the bare repo in the cache folder at the specific sha1
check_call([
"git", "--git-dir", directory + "/.git", "--work-tree", directory,
"checkout", "-f", "-q", self.ref
])
def validate_name(name):
# [a-zA-Z0-9@._+-]
# May not start with '.' or '-'.
if not re.match(name_regex, name):
raise ValidationError(
"Invalid package name {0}. Must match the regex {1}".format(
name, name_regex))
def symlink_tree(src, dest):
for name in os.listdir(src):
src_path = os.path.join(src, name)
dest_path = os.path.join(dest, name)
# Symlink files and symlinks directly. For directories make a
# real directory and symlink everything inside.
# NOTE: We could relax this and follow symlinks, but then we
# need to be careful about recursive filesystem layouts.
if os.path.isdir(src_path) and not os.path.islink(src_path):
if os.path.exists(dest_path):
# We can only merge a directory into a directory.
# We won't merge into a symlink directory because that could
# result in a package editing inside another package.
if not os.path.isdir(dest_path) and not os.path.islink(
dest_path):
raise ValidationError(
"Can't merge a file `{0}` and directory (or symlink) `{1}` with the same name."
.format(src_path, dest_path))
else:
os.makedirs(dest_path)
# Recurse into the directory symlinking everything so long as the directory isn't
symlink_tree(src_path, dest_path)
else:
try:
os.symlink(src_path, dest_path)
except FileNotFoundError as ex:
raise ConflictingFile(src_path, dest_path, ex) from ex
def validate_group_name(group_name):
if not group_name:
return
if not re.match(linux_group_regex, group_name):
raise ValidationError("Group {} has invalid name, must match the following regex: {}".format(
group_name, linux_group_regex))
def validate_version(version):
# [a-zA-Z0-9@._+:]
# May not contain a '-'.
if not re.match(version_regex, version):
raise ValidationError(
"Invalid package version {0}. Must match the regex {1}".format(
version, version_regex))
def add_package_file(repository, package_filename):
"""Add a package to the repository from a file.
repository: pkgpanda.Repository
package_filename: location of the package file
"""
filename_suffix = '.tar.xz'
# Extract Package Id (Filename must be path/{pkg-id}.tar.xz).
name = os.path.basename(package_filename)
if not name.endswith(filename_suffix):
raise ValidationError(
"ERROR: Can only add package tarballs which have names like "
"{{pkg-id}}{}".format(filename_suffix))
pkg_id = name[:-len(filename_suffix)]
# Validate the package id
PackageId(pkg_id)
def fetch(_, target):
extract_tarball(package_filename, target)
repository.add(fetch, pkg_id)
def swap_active_package(install, repository, package_id, systemd, block_systemd):
"""Replace an active package with a package_id with the same name.
swap(install, repository, 'foo--version') will replace the active 'foo'
package with 'foo--version'.
install: pkgpanda.Install
repository: pkgpanda.Repository
package_id: package ID to activate
systemd: start/stop systemd services
block_systemd: if systemd, block waiting for systemd services to come up
"""
active = install.get_active()
# TODO(cmaloney): I guarantee there is a better way to write this and
# I've written the same logic before...
packages_by_name = dict()
for id_str in active:
pkg_id = PackageId(id_str)
packages_by_name[pkg_id.name] = pkg_id
new_id = PackageId(package_id)
if new_id.name not in packages_by_name:
raise ValidationError("No package with name {} currently active to swap with.".format(new_id.name))
packages_by_name[new_id.name] = new_id
new_active = list(map(str, packages_by_name.values()))
# Activate with the new package name
activate_packages(install, repository, new_active, systemd, block_systemd)
def validate_compatible(packages, roles):
# Every package name appears only once.
names = set()
ids = set()
tuples = set()
for package in packages:
if package.name in names:
raise ValidationError(
"Repeated name {0} in set of packages {1}".format(
package.name, ' '.join(map(lambda x: str(x.id),
packages))))
names.add(package.name)
ids.add(str(package.id))
tuples.add((package.name, package.variant))
# All requires are met.
# NOTE: Requires are given just to make it harder to accidentally
# break a cluster.
# Environment variables in packages, mapping from variable to package.
environment = dict()
for package in packages:
# Check that all requirements of the package are met.
# Requirements can be specified on a package name or full version string.
for requirement in package.requires:
name, variant = expand_require(requirement)
if name not in names:
raise ValidationError(
("Package {} variant {} requires {} variant {} but that " +
"is not in the set of packages {}").format(
package.id, package.variant, name, variant,
', '.join(str(x.id) for x in packages)))
# No repeated/conflicting environment variables with other packages as
# well as magic system enviornment variables.
for k, v in package.environment.items():
if k in reserved_env_vars:
raise ValidationError(
"{0} are reserved enviornment vars and cannot be specified in packages. Present in package {1}"
.format(", ".join(reserved_env_vars), package))
if k in environment:
raise ValidationError(
"Repeated environment variable {0}. In both packages {1} and {2}."
.format(k, v, package))
environment[k] = package
def add_user(self, username, group):
UserManagement.validate_username(username)
if not self._manage_users:
return
# Check if the user already exists and exit.
try:
UserManagement.validate_user_group(username, group)
self._users.add(username)
return
except KeyError as ex:
# Doesn't exist, fall through
pass
# If we're not allowed to manage users, error
if not self._add_users:
raise ValidationError(
"User {} doesn't exist but is required by a DC/OS Component, and "
"automatic user addition is disabled".format(username))
# Add the user:
add_user_cmd = [
'useradd',
'--system',
'--home-dir',
'/opt/mesosphere',
'--shell',
'/sbin/nologin',
'-c',
'DCOS System User',
]
if group is not None:
UserManagement.validate_group(group)
add_user_cmd += ['-g', group]
add_user_cmd += [username]
try:
check_output(add_user_cmd)
self._users.add(username)
except CalledProcessError as ex:
raise ValidationError(
"User {} doesn't exist and couldn't be created because of: {}".
format(username, ex.output))
def __init__(self, src_info, cache_dir):
super().__init__(src_info)
assert self.kind == 'git'
if src_info.keys() != {'kind', 'git', 'ref', 'ref_origin'}:
raise ValidationError(
"git source must have keys 'git' (the repo to fetch), 'ref' (the sha-1 to "
"checkout), and 'ref_origin' (the branch/tag ref was derived from)")
if not is_sha(src_info['ref']):
raise ValidationError("ref must be a sha1. Got: {}".format(src_info['ref']))
self.url = src_info['git']
self.ref = src_info['ref']
self.ref_origin = src_info['ref_origin']
self.bare_folder = cache_dir + "/cache.git".format()
def get_git_sha1(bare_folder, ref):
try:
return check_output(
["git", "--git-dir", bare_folder, "rev-parse",
ref + "^{commit}"]).decode('ascii').strip()
except CalledProcessError as ex:
raise ValidationError("Unable to find ref '{}' in '{}': {}".format(
ref, bare_folder, ex)) from ex
def validate_group(group):
# Empty group is allowed.
if not group:
return
UserManagement.validate_group_name(group)
try:
grp.getgrnam(group)
except KeyError:
raise ValidationError("Group {} does not exist on the system".format(group))
def parse(id: str):
parts = id.split('--')
if len(parts) != 2:
raise ValidationError(
"Invalid package id {0}. Package ids may only ".format(id) +
"contain one '--' which seperates the name and version")
PackageId.validate_name(parts[0])
PackageId.validate_version(parts[1])
return parts[0], parts[1]
def extract_archive(archive, dst_dir):
archive_type = _identify_archive_type(archive)
if archive_type == 'tar':
check_call(["tar", "-xf", archive, "--strip-components=1", "-C", dst_dir])
elif archive_type == 'zip':
check_call(["unzip", "-x", archive, "-d", dst_dir])
# unzip binary does not support '--strip-components=1',
_strip_first_path_component(dst_dir)
else:
raise ValidationError("Unsupported archive: {}".format(os.path.basename(archive)))
def _get_package_list(package_list_id: str, repository_url: str) -> List[str]:
package_list_url = repository_url + '/package_lists/{}.package_list.json'.format(package_list_id)
with tempfile.NamedTemporaryFile() as f:
download(f.name, package_list_url, os.getcwd(), rm_on_error=False)
package_list = load_json(f.name)
if not isinstance(package_list, list):
raise ValidationError('{} should contain a JSON list of packages. Got a {}'.format(
package_list_url, type(package_list)
))
return package_list
def __init__(self, src_info, cache_dir, working_directory):
super().__init__(src_info)
assert self.kind == 'git_local'
if src_info.keys() > {'kind', 'rel_path'}:
raise ValidationError(
"Only kind, rel_path can be specified for git_local")
if os.path.isabs(src_info['rel_path']):
raise ValidationError(
"rel_path must be a relative path to the current directory "
"when used with git_local. Using a relative path means others "
"that clone the repository will have things just work rather "
"than a path.")
self.src_repo_path = os.path.normpath(working_directory + '/' +
src_info['rel_path']).rstrip('/')
# Make sure there are no local changes, we can't `git clone` local changes.
try:
git_status = check_output([
'git', '-C', self.src_repo_path, 'status', '--porcelain',
'-uno', '-z'
]).decode()
if len(git_status):
raise ValidationError(
"No local changse are allowed in the git_local_work base repository. "
"Use `git -C {0} status` to see local changes. "
"All local changes must be committed or stashed before the "
"package can be built. One workflow (temporary commit): `git -C {0} "
"commit -am TMP` to commit everything, build the package, "
"`git -C {0} reset --soft HEAD^` to get back to where you were.\n\n"
"Found changes: {1}".format(self.src_repo_path,
git_status))
except CalledProcessError:
raise ValidationError(
"Unable to check status of git_local_work checkout {}. Is the "
"rel_path correct?".format(src_info['rel_path']))
self.commit = get_git_sha1(self.src_repo_path + "/.git", "HEAD")
def expand_require(require):
name = None
variant = None
if isinstance(require, str):
name = require
elif isinstance(require, dict):
if 'name' not in require or 'variant' not in require:
raise ValidationError(
"When specifying a dependency in requires by dictionary to " +
"depend on a variant both the name of the package and the " +
"variant name must always be specified")
name = require['name']
variant = require['variant']
if PackageId.is_id(name):
raise ValidationError(
"ERROR: Specifying a dependency on '" + name + "', an exact" +
"package id isn't allowed. Dependencies may be specified by" +
"package name alone or package name + variant (to change the" +
"package variant).")
return (name, variant)
def __init__(self, name, src_info, package_dir):
super().__init__(name, src_info, package_dir)
assert self.kind in {'url', 'url_extract'}
if src_info.keys() != {'kind', 'sha1', 'url'}:
raise ValidationError(
"url and url_extract sources must have exactly 'sha1' (sha1 of the artifact"
" which will be downloaded), and 'url' (url to download artifact) as options")
self.url = src_info['url']
self.extract = (self.kind == 'url_extract')
self.cache_filename = self._get_filename(package_dir + "/cache")
self.sha = src_info['sha1']
def validate_user_group(username, group_name):
user = pwd.getpwnam(username)
if not group_name:
return
group = grp.getgrnam(group_name)
if user.pw_gid != group.gr_gid:
# check if the user is the right group, but the group is not primary.
if username in group.gr_mem:
return
raise ValidationError(
"User {} exists with current UID {}, however he should be assigned to group {} with {} UID, please "
"check `buildinfo.json`".format(username, user.pw_gid, group_name, group.gr_gid))
def extract_archive(archive, dst_dir):
archive_type = _identify_archive_type(archive)
if archive_type == 'tar':
if is_windows:
check_call(["bsdtar", "-xf", archive, "-C", dst_dir])
else:
check_call(["tar", "-xf", archive, "--strip-components=1", "-C", dst_dir])
elif archive_type == 'zip':
if is_windows:
check_call(["powershell.exe", "-command", "expand-archive", "-path", archive, "-destinationpath", dst_dir])
else:
check_call(["unzip", "-x", archive, "-d", dst_dir])
# unzip binary does not support '--strip-components=1',
_strip_first_path_component(dst_dir)
else:
raise ValidationError("Unsupported archive: {}".format(os.path.basename(archive)))
def _check_components_sanity(path):
"""Check if archive is sane
Check if there is only one top level component (directory) in the extracted
archive's directory.
Args:
path: path to the extracted archive's directory
Raises:
Raise an exception if there is anything else than a single directory
"""
dir_contents = os.listdir(path)
if len(dir_contents) != 1 or not os.path.isdir(os.path.join(path, dir_contents[0])):
raise ValidationError("Extracted archive has more than one top level"
"component, unable to strip it.")
def get_src_fetcher(src_info, cache_dir, working_directory):
try:
kind = src_info['kind']
if kind not in pkgpanda.build.src_fetchers.all_fetchers:
raise ValidationError(
"No known way to catch src with kind '{}'. Known kinds: {}".
format(kind, pkgpanda.src_fetchers.all_fetchers.keys()))
args = {'src_info': src_info, 'cache_dir': cache_dir}
if src_info['kind'] in ['git_local', 'url', 'url_extract']:
args['working_directory'] = working_directory
return pkgpanda.build.src_fetchers.all_fetchers[kind](**args)
except ValidationError as ex:
raise BuildError(
"Validation error when fetching sources for package: {}".format(
ex))
def __init__(self, src_info, cache_dir, working_directory):
super().__init__(src_info)
assert self.kind in {'url', 'url_extract'}
if ('kind' not in src_info) or ('sha1'
not in src_info) or ('url'
not in src_info):
raise ValidationError(
"url and url_extract sources must have exactly 'sha1' (sha1 of the artifact"
" which will be downloaded), and 'url' (url to download artifact) as options"
)
self.url = src_info['url']
self.extract = (self.kind == 'url_extract')
self.cache_dir = cache_dir
self.cache_filename = self._get_filename(cache_dir)
self.working_directory = working_directory
self.sha = src_info['sha1']
def checkout_to(self, directory):
# Download file to cache if it isn't already there
if not os.path.exists(self.cache_filename):
print("Downloading source tarball {}".format(self.url))
download_atomic(self.cache_filename, self.url, self.working_directory)
# Validate the sha1 of the source is given and matches the sha1
file_sha = sha1(self.cache_filename)
if self.sha != file_sha:
corrupt_filename = self.cache_filename + '.corrupt'
check_call(['mv', self.cache_filename, corrupt_filename])
raise ValidationError(
"Provided sha1 didn't match sha1 of downloaded file, corrupt download saved as {}. "
"Provided: {}, Download file's sha1: {}, Url: {}".format(
corrupt_filename, self.sha, file_sha, self.url))
if self.extract:
extract_archive(self.cache_filename, directory)
else:
# Copy the file(s) into src/
# TODO(cmaloney): Hardlink to save space?
shutil.copyfile(self.cache_filename, self._get_filename(directory))
def fetcher(id, target):
if repository_url is None:
raise ValidationError("ERROR: Non-local package {} but no repository url given.".format(id))
return requests_fetcher(repository_url, id, target, os.getcwd())
def _do_bootstrap(install, repository):
# These files should be set by the environment which initially builds
# the host (cloud-init).
repository_url = if_exists(load_string, install.get_config_filename("setup-flags/repository-url"))
def fetcher(id, target):
if repository_url is None:
raise ValidationError("ERROR: Non-local package {} but no repository url given.".format(id))
return requests_fetcher(repository_url, id, target, os.getcwd())
setup_pkg_dir = install.get_config_filename("setup-packages")
if os.path.exists(setup_pkg_dir):
raise ValidationError(
"setup-packages is no longer supported. It's functionality has been replaced with late "
"binding packages. Found setup packages dir: {}".format(setup_pkg_dir))
setup_packages_to_activate = []
# If the host has late config values, build the late config package from them.
late_config = if_exists(load_yaml, install.get_config_filename("setup-flags/late-config.yaml"))
if late_config:
pkg_id_str = late_config['late_bound_package_id']
late_values = late_config['bound_values']
print("Binding late config to late package {}".format(pkg_id_str))
print("Bound values: {}".format(late_values))
if not PackageId.is_id(pkg_id_str):
raise ValidationError("Invalid late package id: {}".format(pkg_id_str))
pkg_id = PackageId(pkg_id_str)
if pkg_id.version != "setup":
raise ValidationError("Late package must have the version setup. Bad package: {}".format(pkg_id_str))
# Collect the late config package.
with tempfile.NamedTemporaryFile() as f:
download(
f.name,
repository_url + '/packages/{0}/{1}.dcos_config'.format(pkg_id.name, pkg_id_str),
os.getcwd(),
rm_on_error=False,
)
late_package = load_yaml(f.name)
# Resolve the late package using the bound late config values.
final_late_package = resolve_late_package(late_package, late_values)
# Render the package onto the filesystem and add it to the package
# repository.
with tempfile.NamedTemporaryFile() as f:
do_gen_package(final_late_package, f.name)
repository.add(lambda _, target: extract_tarball(f.name, target), pkg_id_str)
setup_packages_to_activate.append(pkg_id_str)
# If active.json is set on the host, use that as the set of packages to
# activate. Otherwise just use the set of currently active packages (those
# active in the bootstrap tarball)
to_activate = None
active_path = install.get_config_filename("setup-flags/active.json")
if os.path.exists(active_path):
print("Loaded active packages from", active_path)
to_activate = load_json(active_path)
# Ensure all packages are local
print("Ensuring all packages in active set {} are local".format(",".join(to_activate)))
for package in to_activate:
repository.add(fetcher, package)
else:
print("Calculated active packages from bootstrap tarball")
to_activate = list(install.get_active())
package_list_filename = install.get_config_filename("setup-flags/cluster-package-list")
print("Checking for cluster packages in:", package_list_filename)
package_list_id = if_exists(load_string, package_list_filename)
if package_list_id:
print("Cluster package list:", package_list_id)
cluster_packages = _get_package_list(package_list_id, repository_url)
print("Loading cluster-packages: {}".format(cluster_packages))
for package_id_str in cluster_packages:
# Validate the package ids
pkg_id = PackageId(package_id_str)
# Fetch the packages if not local
if not repository.has_package(package_id_str):
repository.add(fetcher, package_id_str)
# Add the package to the set to activate
setup_packages_to_activate.append(package_id_str)
else:
print("No cluster-packages specified")
# Calculate the full set of final packages (Explicit activations + setup packages).
# De-duplicate using a set.
to_activate = list(set(to_activate + setup_packages_to_activate))
print("Activating packages")
install.activate(repository.load_packages(to_activate))
def _do_bootstrap(install, repository):
# These files should be set by the environment which initially builds
# the host (cloud-init).
repository_url = if_exists(
load_string, install.get_config_filename("setup-flags/repository-url"))
# TODO(cmaloney): If there is 1+ master, grab the active config from a master.
# If the config can't be grabbed from any of them, fail.
def fetcher(id, target):
if repository_url is None:
raise ValidationError(
"ERROR: Non-local package {} but no repository url given.".
format(repository_url))
return requests_fetcher(repository_url, id, target, os.getcwd())
# Copy host/cluster-specific packages written to the filesystem manually
# from the setup-packages folder into the repository. Do not overwrite or
# merge existing packages, hard fail instead.
setup_packages_to_activate = []
setup_pkg_dir = install.get_config_filename("setup-packages")
copy_fetcher = partial(_copy_fetcher, setup_pkg_dir)
if os.path.exists(setup_pkg_dir):
for pkg_id_str in os.listdir(setup_pkg_dir):
print("Installing setup package: {}".format(pkg_id_str))
if not PackageId.is_id(pkg_id_str):
raise ValidationError(
"Invalid package id in setup package: {}".format(
pkg_id_str))
pkg_id = PackageId(pkg_id_str)
if pkg_id.version != "setup":
raise ValidationError(
"Setup packages (those in `{0}`) must have the version setup. "
"Bad package: {1}".format(setup_pkg_dir, pkg_id_str))
# Make sure there is no existing package
if repository.has_package(pkg_id_str):
print("WARNING: Ignoring already installed package {}".format(
pkg_id_str))
repository.add(copy_fetcher, pkg_id_str)
setup_packages_to_activate.append(pkg_id_str)
# If active.json is set on the host, use that as the set of packages to
# activate. Otherwise just use the set of currently active packages (those
# active in the bootstrap tarball)
to_activate = None
active_path = install.get_config_filename("setup-flags/active.json")
if os.path.exists(active_path):
print("Loaded active packages from", active_path)
to_activate = load_json(active_path)
# Ensure all packages are local
print("Ensuring all packages in active set {} are local".format(
",".join(to_activate)))
for package in to_activate:
repository.add(fetcher, package)
else:
print("Calculated active packages from bootstrap tarball")
to_activate = list(install.get_active())
# Fetch and activate all requested additional packages to accompany the bootstrap packages.
cluster_packages_filename = install.get_config_filename(
"setup-flags/cluster-packages.json")
cluster_packages = if_exists(load_json, cluster_packages_filename)
print("Checking for cluster packages in:", cluster_packages_filename)
if cluster_packages:
if not isinstance(cluster_packages, list):
print(
'ERROR: {} should contain a JSON list of packages. Got a {}'
.format(cluster_packages_filename, type(cluster_packages)))
print("Loading cluster-packages: {}".format(cluster_packages))
for package_id_str in cluster_packages:
# Validate the package ids
pkg_id = PackageId(package_id_str)
# Fetch the packages if not local
if not repository.has_package(package_id_str):
repository.add(fetcher, package_id_str)
# Add the package to the set to activate
setup_packages_to_activate.append(package_id_str)
else:
print("No cluster-packages specified")
# Calculate the full set of final packages (Explicit activations + setup packages).
# De-duplicate using a set.
to_activate = list(set(to_activate + setup_packages_to_activate))
print("Activating packages")
install.activate(repository.load_packages(to_activate))
def build(package_store, name, variant, clean_after_build, recursive=False):
assert isinstance(package_store, PackageStore)
print("Building package {} variant {}".format(
name, pkgpanda.util.variant_str(variant)))
tmpdir = tempfile.TemporaryDirectory(prefix="pkgpanda_repo")
repository = Repository(tmpdir.name)
package_dir = package_store.get_package_folder(name)
def src_abs(name):
return package_dir + '/' + name
def cache_abs(filename):
return package_store.get_package_cache_folder(name) + '/' + filename
# Build pkginfo over time, translating fields from buildinfo.
pkginfo = {}
# Build up the docker command arguments over time, translating fields as needed.
cmd = DockerCmd()
assert (name, variant) in package_store.packages, \
"Programming error: name, variant should have been validated to be valid before calling build()."
buildinfo = copy.deepcopy(package_store.get_buildinfo(name, variant))
if 'name' in buildinfo:
raise BuildError(
"'name' is not allowed in buildinfo.json, it is implicitly the name of the "
"folder containing the buildinfo.json")
# Convert single_source -> sources
try:
sources = expand_single_source_alias(name, buildinfo)
except ValidationError as ex:
raise BuildError(
"Invalid buildinfo.json for package: {}".format(ex)) from ex
# Save the final sources back into buildinfo so it gets written into
# buildinfo.json. This also means buildinfo.json is always expanded form.
buildinfo['sources'] = sources
# Construct the source fetchers, gather the checkout ids from them
checkout_ids = dict()
fetchers = dict()
try:
for src_name, src_info in sorted(sources.items()):
# TODO(cmaloney): Switch to a unified top level cache directory shared by all packages
cache_dir = package_store.get_package_cache_folder(
name) + '/' + src_name
check_call(['mkdir', '-p', cache_dir])
fetcher = get_src_fetcher(src_info, cache_dir, package_dir)
fetchers[src_name] = fetcher
checkout_ids[src_name] = fetcher.get_id()
except ValidationError as ex:
raise BuildError(
"Validation error when fetching sources for package: {}".format(
ex))
for src_name, checkout_id in checkout_ids.items():
# NOTE: single_source buildinfo was expanded above so the src_name is
# always correct here.
# Make sure we never accidentally overwrite something which might be
# important. Fields should match if specified (And that should be
# tested at some point). For now disallowing identical saves hassle.
assert_no_duplicate_keys(checkout_id, buildinfo['sources'][src_name])
buildinfo['sources'][src_name].update(checkout_id)
# Add the sha1 of the buildinfo.json + build file to the build ids
build_ids = {"sources": checkout_ids}
build_ids['build'] = pkgpanda.util.sha1(src_abs(buildinfo['build_script']))
build_ids['pkgpanda_version'] = pkgpanda.build.constants.version
build_ids['variant'] = '' if variant is None else variant
extra_dir = src_abs("extra")
# Add the "extra" folder inside the package as an additional source if it
# exists
if os.path.exists(extra_dir):
extra_id = hash_folder(extra_dir)
build_ids['extra_source'] = extra_id
buildinfo['extra_source'] = extra_id
# Figure out the docker name.
docker_name = buildinfo['docker']
cmd.container = docker_name
# Add the id of the docker build environment to the build_ids.
try:
docker_id = get_docker_id(docker_name)
except CalledProcessError:
# docker pull the container and try again
check_call(['docker', 'pull', docker_name])
docker_id = get_docker_id(docker_name)
build_ids['docker'] = docker_id
# TODO(cmaloney): The environment variables should be generated during build
# not live in buildinfo.json.
build_ids['environment'] = buildinfo['environment']
# Packages need directories inside the fake install root (otherwise docker
# will try making the directories on a readonly filesystem), so build the
# install root now, and make the package directories in it as we go.
install_dir = tempfile.mkdtemp(prefix="pkgpanda-")
active_packages = list()
active_package_ids = set()
active_package_variants = dict()
auto_deps = set()
# Verify all requires are in the repository.
if 'requires' in buildinfo:
# Final package has the same requires as the build.
pkginfo['requires'] = buildinfo['requires']
# TODO(cmaloney): Pull generating the full set of requires a function.
to_check = copy.deepcopy(buildinfo['requires'])
if type(to_check) != list:
raise BuildError(
"`requires` in buildinfo.json must be an array of dependencies."
)
while to_check:
requires_info = to_check.pop(0)
requires_name, requires_variant = expand_require(requires_info)
if requires_name in active_package_variants:
# TODO(cmaloney): If one package depends on the <default>
# variant of a package and 1+ others depends on a non-<default>
# variant then update the dependency to the non-default variant
# rather than erroring.
if requires_variant != active_package_variants[requires_name]:
# TODO(cmaloney): Make this contain the chains of
# dependencies which contain the conflicting packages.
# a -> b -> c -> d {foo}
# e {bar} -> d {baz}
raise BuildError(
"Dependncy on multiple variants of the same package {}. "
"variants: {} {}".format(
requires_name, requires_variant,
active_package_variants[requires_name]))
# The variant has package {requires_name, variant} already is a
# dependency, don't process it again / move on to the next.
continue
active_package_variants[requires_name] = requires_variant
# Figure out the last build of the dependency, add that as the
# fully expanded dependency.
requires_last_build = package_store.get_last_build_filename(
requires_name, requires_variant)
if not os.path.exists(requires_last_build):
if recursive:
# Build the dependency
build(package_store, requires_name, requires_variant,
clean_after_build, recursive)
else:
raise BuildError(
"No last build file found for dependency {} variant {}. Rebuild "
"the dependency".format(requires_name,
requires_variant))
try:
pkg_id_str = load_string(requires_last_build)
auto_deps.add(pkg_id_str)
pkg_buildinfo = package_store.get_buildinfo(
requires_name, requires_variant)
pkg_requires = pkg_buildinfo['requires']
pkg_path = repository.package_path(pkg_id_str)
pkg_tar = pkg_id_str + '.tar.xz'
if not os.path.exists(
package_store.get_package_cache_folder(requires_name) +
'/' + pkg_tar):
raise BuildError(
"The build tarball {} refered to by the last_build file of the "
"dependency {} variant {} doesn't exist. Rebuild the dependency."
.format(pkg_tar, requires_name, requires_variant))
active_package_ids.add(pkg_id_str)
# Mount the package into the docker container.
cmd.volumes[
pkg_path] = "/opt/mesosphere/packages/{}:ro".format(
pkg_id_str)
os.makedirs(
os.path.join(install_dir,
"packages/{}".format(pkg_id_str)))
# Add the dependencies of the package to the set which will be
# activated.
# TODO(cmaloney): All these 'transitive' dependencies shouldn't
# be available to the package being built, only what depends on
# them directly.
to_check += pkg_requires
except ValidationError as ex:
raise BuildError(
"validating package needed as dependency {0}: {1}".format(
requires_name, ex)) from ex
except PackageError as ex:
raise BuildError(
"loading package needed as dependency {0}: {1}".format(
requires_name, ex)) from ex
# Add requires to the package id, calculate the final package id.
# NOTE: active_packages isn't fully constructed here since we lazily load
# packages not already in the repository.
build_ids['requires'] = list(active_package_ids)
version_base = hash_checkout(build_ids)
version = None
if "version_extra" in buildinfo:
version = "{0}-{1}".format(buildinfo["version_extra"], version_base)
else:
version = version_base
pkg_id = PackageId.from_parts(name, version)
# Save the build_ids. Useful for verify exactly what went into the
# package build hash.
buildinfo['build_ids'] = build_ids
buildinfo['package_version'] = version
# Save the package name and variant. The variant is used when installing
# packages to validate dependencies.
buildinfo['name'] = name
buildinfo['variant'] = variant
# If the package is already built, don't do anything.
pkg_path = package_store.get_package_cache_folder(
name) + '/{}.tar.xz'.format(pkg_id)
# Done if it exists locally
if exists(pkg_path):
print("Package up to date. Not re-building.")
# TODO(cmaloney): Updating / filling last_build should be moved out of
# the build function.
write_string(package_store.get_last_build_filename(name, variant),
str(pkg_id))
return pkg_path
# Try downloading.
dl_path = package_store.try_fetch_by_id(pkg_id)
if dl_path:
print(
"Package up to date. Not re-building. Downloaded from repository-url."
)
# TODO(cmaloney): Updating / filling last_build should be moved out of
# the build function.
write_string(package_store.get_last_build_filename(name, variant),
str(pkg_id))
print(dl_path, pkg_path)
assert dl_path == pkg_path
return pkg_path
# Fall out and do the build since it couldn't be downloaded
print("Unable to download from cache. Proceeding to build")
print("Building package {} with buildinfo: {}".format(
pkg_id, json.dumps(buildinfo, indent=2, sort_keys=True)))
# Clean out src, result so later steps can use them freely for building.
def clean():
# Run a docker container to remove src/ and result/
cmd = DockerCmd()
cmd.volumes = {
package_store.get_package_cache_folder(name): "/pkg/:rw",
}
cmd.container = "ubuntu:14.04.4"
cmd.run(["rm", "-rf", "/pkg/src", "/pkg/result"])
clean()
# Only fresh builds are allowed which don't overlap existing artifacts.
result_dir = cache_abs("result")
if exists(result_dir):
raise BuildError(
"result folder must not exist. It will be made when the package is "
"built. {}".format(result_dir))
# 'mkpanda add' all implicit dependencies since we actually need to build.
for dep in auto_deps:
print("Auto-adding dependency: {}".format(dep))
# NOTE: Not using the name pkg_id because that overrides the outer one.
id_obj = PackageId(dep)
add_package_file(repository, package_store.get_package_path(id_obj))
package = repository.load(dep)
active_packages.append(package)
# Checkout all the sources int their respective 'src/' folders.
try:
src_dir = cache_abs('src')
if os.path.exists(src_dir):
raise ValidationError(
"'src' directory already exists, did you have a previous build? "
+
"Currently all builds must be from scratch. Support should be "
+ "added for re-using a src directory when possible. src={}".
format(src_dir))
os.mkdir(src_dir)
for src_name, fetcher in sorted(fetchers.items()):
root = cache_abs('src/' + src_name)
os.mkdir(root)
fetcher.checkout_to(root)
except ValidationError as ex:
raise BuildError(
"Validation error when fetching sources for package: {}".format(
ex))
# Copy over environment settings
pkginfo['environment'] = buildinfo['environment']
# Whether pkgpanda should on the host make sure a `/var/lib` state directory is available
pkginfo['state_directory'] = buildinfo.get('state_directory', False)
if pkginfo['state_directory'] not in [True, False]:
raise BuildError(
"state_directory in buildinfo.json must be a boolean `true` or `false`"
)
username = buildinfo.get('username')
if not (username is None or isinstance(username, str)):
raise BuildError(
"username in buildinfo.json must be either not set (no user for this"
" package), or a user name string")
if username:
try:
pkgpanda.UserManagement.validate_username(username)
except ValidationError as ex:
raise BuildError(
"username in buildinfo.json didn't meet the validation rules. {}"
.format(ex))
pkginfo['username'] = username
# Activate the packages so that we have a proper path, environment
# variables.
# TODO(cmaloney): RAII type thing for temproary directory so if we
# don't get all the way through things will be cleaned up?
install = Install(root=install_dir,
config_dir=None,
rooted_systemd=True,
manage_systemd=False,
block_systemd=True,
fake_path=True,
manage_users=False,
manage_state_dir=False)
install.activate(active_packages)
# Rewrite all the symlinks inside the active path because we will
# be mounting the folder into a docker container, and the absolute
# paths to the packages will change.
# TODO(cmaloney): This isn't very clean, it would be much nicer to
# just run pkgpanda inside the package.
rewrite_symlinks(install_dir, repository.path, "/opt/mesosphere/packages/")
print("Building package in docker")
# TODO(cmaloney): Run as a specific non-root user, make it possible
# for non-root to cleanup afterwards.
# Run the build, prepping the environment as necessary.
mkdir(cache_abs("result"))
# Copy the build info to the resulting tarball
write_json(cache_abs("src/buildinfo.full.json"), buildinfo)
write_json(cache_abs("result/buildinfo.full.json"), buildinfo)
write_json(cache_abs("result/pkginfo.json"), pkginfo)
# Make the folder for the package we are building. If docker does it, it
# gets auto-created with root permissions and we can't actually delete it.
os.makedirs(os.path.join(install_dir, "packages", str(pkg_id)))
# TOOD(cmaloney): Disallow writing to well known files and directories?
# Source we checked out
cmd.volumes.update({
# TODO(cmaloney): src should be read only...
cache_abs("src"):
"/pkg/src:rw",
# The build script
src_abs(buildinfo['build_script']):
"/pkg/build:ro",
# Getting the result out
cache_abs("result"):
"/opt/mesosphere/packages/{}:rw".format(pkg_id),
install_dir:
"/opt/mesosphere:ro"
})
if os.path.exists(extra_dir):
cmd.volumes[extra_dir] = "/pkg/extra:ro"
cmd.environment = {
"PKG_VERSION": version,
"PKG_NAME": name,
"PKG_ID": pkg_id,
"PKG_PATH": "/opt/mesosphere/packages/{}".format(pkg_id),
"PKG_VARIANT": variant if variant is not None else "<default>"
}
try:
# TODO(cmaloney): Run a wrapper which sources
# /opt/mesosphere/environment then runs a build. Also should fix
# ownership of /opt/mesosphere/packages/{pkg_id} post build.
cmd.run([
"/bin/bash", "-o", "nounset", "-o", "pipefail", "-o", "errexit",
"/pkg/build"
])
except CalledProcessError as ex:
raise BuildError("docker exited non-zero: {}\nCommand: {}".format(
ex.returncode, ' '.join(ex.cmd)))
# Clean up the temporary install dir used for dependencies.
# TODO(cmaloney): Move to an RAII wrapper.
check_call(['rm', '-rf', install_dir])
print("Building package tarball")
# Check for forbidden services before packaging the tarball:
try:
check_forbidden_services(cache_abs("result"), RESERVED_UNIT_NAMES)
except ValidationError as ex:
raise BuildError("Package validation failed: {}".format(ex))
# TODO(cmaloney): Updating / filling last_build should be moved out of
# the build function.
write_string(package_store.get_last_build_filename(name, variant),
str(pkg_id))
# Bundle the artifacts into the pkgpanda package
tmp_name = pkg_path + "-tmp.tar.xz"
make_tar(tmp_name, cache_abs("result"))
os.rename(tmp_name, pkg_path)
print("Package built.")
if clean_after_build:
clean()
return pkg_path