zuul/zuul/merger/merger.py
James E. Blair cdaa86d78b Stop logging refs set
We currently emit a debug log line for each ref that we set when
preparing a repo for a job.  This might be an enormous number of
very boring log lines for large repos.  Since we now store this
information in a file accessible to the job (which can be saved
with the job logs), this information is now accessible for debug
purposes that way.

Instead, just log how many refs we intend to set (so that if there
are a lot, we know what the executor/merger is doing) and log any
missing objects.

Change-Id: Iebc52cbc4aa4fb778b371f90343e53ece6fa3699
2024-06-10 08:00:52 -07:00

1459 lines
61 KiB
Python

# Copyright 2012 Hewlett-Packard Development Company, L.P.
# Copyright 2013-2014 OpenStack Foundation
# Copyright 2021-2023 Acme Gating, LLC
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
from contextlib import contextmanager
from urllib.parse import urlsplit, urlunsplit, urlparse
import hashlib
import logging
import math
import sys
import os
import re
import shutil
import time
from collections import OrderedDict
from concurrent.futures.process import BrokenProcessPool
import git
import gitdb
import paramiko
from zuul.lib import strings
import zuul.model
from zuul.lib.logutil import get_annotated_logger
NULL_REF = '0000000000000000000000000000000000000000'
def redact_url(url):
parsed = urlsplit(url)
if parsed.password is None:
return url
# items[1] is the netloc containing credentials and hostname
items = list(parsed)
items[1] = re.sub('.*@', '******@', items[1])
return urlunsplit(items)
@contextmanager
def timeout_handler(path):
try:
yield
except git.exc.GitCommandError as e:
if e.status == -9:
# Timeout. The repo could be in a bad state, so delete it.
if os.path.exists(path):
shutil.rmtree(path)
raise
@contextmanager
def nullcontext():
yield
class Repo(object):
commit_re = re.compile(r'^commit ([0-9a-f]{40})$')
diff_re = re.compile(r'^@@ -\d+,\d \+(\d+),\d @@$')
retry_attempts = 3
retry_interval = 30
def __init__(self, remote, local, email, username, speed_limit, speed_time,
sshkey=None, cache_path=None, logger=None, git_timeout=300,
zuul_event_id=None, retry_timeout=None, skip_refs=False):
if logger is None:
self.log = logging.getLogger("zuul.Repo")
else:
self.log = logger
log = get_annotated_logger(self.log, zuul_event_id)
self.skip_refs = skip_refs
self.env = {
'GIT_HTTP_LOW_SPEED_LIMIT': speed_limit,
'GIT_HTTP_LOW_SPEED_TIME': speed_time,
}
self.git_timeout = git_timeout
if retry_timeout:
self.retry_attempts = math.ceil(
retry_timeout / self.retry_interval)
self.sshkey = sshkey
if sshkey:
self.env['GIT_SSH_COMMAND'] = 'ssh -i %s' % (sshkey,)
self.remote_url = remote
self.local_path = local
self.email = email
self.username = username
self.cache_path = cache_path
self._initialized = False
try:
self._setup_known_hosts()
except Exception:
log.exception("Unable to set up known_hosts for %s",
redact_url(remote))
try:
self._ensure_cloned(zuul_event_id)
self._git_set_remote_url(
git.Repo(self.local_path), self.remote_url)
except Exception:
log.exception("Unable to initialize repo for %s",
redact_url(remote))
def __repr__(self):
return "<Repo {} {}>".format(hex(id(self)), self.local_path)
def _setup_known_hosts(self):
url = urlparse(self.remote_url)
if 'ssh' not in url.scheme:
return
port = url.port or 22
username = url.username or self.username
path = os.path.expanduser('~/.ssh')
os.makedirs(path, exist_ok=True)
path = os.path.expanduser('~/.ssh/known_hosts')
if not os.path.exists(path):
with open(path, 'w'):
pass
client = paramiko.SSHClient()
client.load_system_host_keys()
client.load_host_keys(path)
client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
try:
client.connect(url.hostname,
username=username,
port=port,
key_filename=self.sshkey)
finally:
# If we don't close on exceptions to connect we can leak the
# connection and DoS Gerrit.
client.close()
def _ensure_cloned(self, zuul_event_id, build=None):
log = get_annotated_logger(self.log, zuul_event_id, build=build)
repo_is_cloned = os.path.exists(os.path.join(self.local_path, '.git'))
if self._initialized and repo_is_cloned:
try:
# validate that the repo isn't corrupt
git.Repo(self.local_path)
return
except Exception:
# the repo is corrupt, delete the local path
shutil.rmtree(self.local_path)
repo_is_cloned = False
self._initialized = False
# If the repo does not exist, clone the repo.
rewrite_url = False
if not repo_is_cloned:
if self.cache_path:
clone_url = self.cache_path
rewrite_url = True
else:
clone_url = self.remote_url
log.debug("Cloning from %s to %s",
redact_url(clone_url), self.local_path)
self._git_clone(clone_url, zuul_event_id, build=build)
repo = git.Repo(self.local_path)
repo.git.update_environment(**self.env)
# Create local branches corresponding to all the remote
# branches. Skip this when cloning the workspace repo since
# we will restore the refs there.
if not repo_is_cloned and not self.skip_refs:
origin = repo.remotes.origin
for ref in origin.refs:
if ref.remote_head == 'HEAD':
continue
repo.create_head('refs/heads/' + ref.remote_head,
ref.commit,
force=True)
with repo.config_writer() as config_writer:
if self.email:
config_writer.set_value('user', 'email', self.email)
if self.username:
config_writer.set_value('user', 'name', self.username)
# By default automatic garbage collection in git runs
# asynchronously in the background. This can lead to broken repos
# caused by a race in the following scenario:
# 1. git fetch (eventually triggers async gc)
# 2. zuul deletes all refs as part of reset
# 3. git gc looks for unreachable objects
# 4. zuul re-creates all refs as part of reset
# 5. git gc deletes unreachable objects it found
# Result is a repo with refs pointing to not existing objects.
# To prevent this race autoDetach can be disabled so git fetch
# returns after the gc finished.
config_writer.set_value('gc', 'autoDetach', 'false')
# Lower the threshold of how many loose objects can trigger
# automatic garbage collection. With the default value of 6700
# we observed that with some repos automatic garbage collection
# simply refused to do its job because it refuses to prune if the
# number of unreachable objects it needs to prune exceeds a certain
# threshold. Thus lower the threshold to trigger automatic garbage
# collection more often.
config_writer.set_value('gc', 'auto', '512')
# By default garbage collection keeps unreachable objects for two
# weeks. However we don't need to carry around any unreachable
# objects so just prune them all when gc kicks in.
config_writer.set_value('gc', 'pruneExpire', 'now')
# By default git keeps a reflog of each branch for 90 days. Objects
# that are reachable from a reflog entry are not considered
# unrechable and thus won't be pruned for 90 days. This can blow up
# the repo significantly over time. Since the reflog is only really
# useful for humans working with repos we can just drop all the
# reflog when gc kicks in.
config_writer.set_value('gc', 'reflogExpire', 'now')
config_writer.write()
if rewrite_url:
self._git_set_remote_url(repo, self.remote_url)
self._initialized = True
def isInitialized(self):
return self._initialized
def _git_clone(self, url, zuul_event_id, build=None):
log = get_annotated_logger(self.log, zuul_event_id, build=build)
mygit = git.cmd.Git(os.getcwd())
mygit.update_environment(**self.env)
kwargs = dict(kill_after_timeout=self.git_timeout)
if self.skip_refs:
kwargs.update(dict(
no_checkout=True,
))
for attempt in range(1, self.retry_attempts + 1):
try:
with timeout_handler(self.local_path):
mygit.clone(git.cmd.Git.polish_url(url), self.local_path,
**kwargs)
break
except Exception:
if attempt < self.retry_attempts:
time.sleep(self.retry_interval)
log.warning("Retry %s: Clone %s", attempt, self.local_path)
else:
raise
def _git_fetch(self, repo, remote, zuul_event_id, ref=None, **kwargs):
log = get_annotated_logger(self.log, zuul_event_id)
for attempt in range(1, self.retry_attempts + 1):
try:
with timeout_handler(self.local_path):
ref_to_fetch = ref
if ref_to_fetch:
ref_to_fetch += ':refs/zuul/fetch'
repo.git.fetch(remote, ref_to_fetch,
kill_after_timeout=self.git_timeout, f=True,
**kwargs)
break
except Exception as e:
if attempt < self.retry_attempts:
if 'fatal: bad config' in e.stderr.lower():
# This error can be introduced by a merge conflict
# or someone committing faulty configuration
# in the .gitmodules which was left by the last
# merge operation. In this case reset and clean
# the repo and try again immediately.
reset_ref = 'HEAD'
try:
if not repo.is_dirty():
reset_ref = "{}^".format(repo.git.log(
'--diff-filter=A',
'-n', '1',
'--pretty=format:%H',
'--', '.gitmodules'))
repo.head.reset(reset_ref, working_tree=True)
repo.git.clean('-x', '-f', '-d')
except Exception:
# If we get here there probably isn't
# a valid commit we can easily find so
# delete the repo to make sure it doesn't
# get stuck in a broken state.
shutil.rmtree(self.local_path)
elif 'fatal: not a git repository' in e.stderr.lower():
# If we get here the git.Repo object was happy with its
# lightweight way of checking if this is a valid git
# repository. However if e.g. the .git/HEAD file is
# empty git operations fail. So there is something
# fundamentally broken with the repo and we need to
# delete it before advancing to _ensure_cloned.
shutil.rmtree(self.local_path)
elif 'error: object file' in e.stderr.lower():
# If we get here the git.Repo object was happy with its
# lightweight way of checking if this is a valid git
# repository. However if git complains about corrupt
# object files the repository is essentially broken and
# needs to be cloned cleanly.
shutil.rmtree(self.local_path)
elif 'bad object' in e.stderr.lower():
# If we get here the git.Repo object was happy with its
# lightweight way of checking if this is a valid git
# repository. However if git reports a bad object, then
# we may have written the repo refs with bad data.
shutil.rmtree(self.local_path)
else:
time.sleep(self.retry_interval)
log.exception("Retry %s: Fetch %s %s %s" % (
attempt, self.local_path, remote, ref))
self._ensure_cloned(zuul_event_id)
else:
raise
def _git_set_remote_url(self, repo, url):
with repo.remotes.origin.config_writer as config_writer:
config_writer.set('url', url)
@staticmethod
def _createRepoObject(path, env):
repo = git.Repo(path)
repo.git.update_environment(**env)
return repo
def createRepoObject(self, zuul_event_id, build=None):
self._ensure_cloned(zuul_event_id, build=build)
return self._createRepoObject(self.local_path, self.env)
@staticmethod
def _cleanup_leaked_ref_dirs(local_path, log, messages):
for root, dirs, files in os.walk(
os.path.join(local_path, '.git/refs'), topdown=False):
if not os.listdir(root) and not root.endswith('.git/refs'):
if log:
log.debug("Cleaning empty ref dir %s", root)
else:
messages.append("Cleaning empty ref dir %s" % root)
os.rmdir(root)
@staticmethod
def _cleanup_leaked_rebase_dirs(local_path, log, messages):
for rebase_dir in [".git/rebase-merge", ".git/rebase-apply"]:
leaked_dir = os.path.join(local_path, rebase_dir)
if not os.path.exists(leaked_dir):
continue
if log:
log.debug("Cleaning leaked %s dir", leaked_dir)
else:
messages.append(
f"Cleaning leaked {leaked_dir} dir")
try:
shutil.rmtree(leaked_dir)
except Exception as exc:
msg = f"Failed to remove leaked {leaked_dir} dir:"
if log:
log.exception(msg)
else:
messages.append(f"{msg}\n{exc}")
@staticmethod
def refNameToZuulRef(ref_name: str) -> str:
return "refs/zuul/{}".format(
hashlib.sha1(ref_name.encode("utf-8")).hexdigest()
)
@staticmethod
def _reset(local_path, env, log=None):
messages = []
repo = Repo._createRepoObject(local_path, env)
origin_refs = {}
head_refs = {}
zuul_refs = {}
# Get all of the local and remote refs in the repo at once.
for commit, ref in Repo._getRefs(repo):
if ref.startswith('refs/remotes/origin/'):
origin_refs[ref[20:]] = commit
if ref.startswith('refs/heads/'):
head_refs[ref[11:]] = commit
if ref.startswith('refs/zuul/'):
zuul_refs[ref] = commit
# Detach HEAD so we can work with references without interfering
# with any active branch. Any remote ref will do as long as it can
# be dereferenced to an existing commit.
for ref, commit in origin_refs.items():
try:
repo.head.reference = commit
break
except Exception:
if log:
log.debug("Unable to detach HEAD to %s", ref)
else:
messages.append("Unable to detach HEAD to %s" % ref)
else:
# There are no remote refs; proceed with the assumption we
# don't have a checkout yet.
if log:
log.debug("Couldn't detach HEAD to any existing commit")
else:
messages.append("Couldn't detach HEAD to any existing commit")
# Delete local heads that no longer exist on the remote end
zuul_refs_to_keep = [
"refs/zuul/fetch", # ref to last FETCH_HEAD
]
for ref in head_refs.keys():
if ref not in origin_refs:
if log:
log.debug("Delete stale local ref %s", ref)
else:
messages.append("Delete stale local ref %s" % ref)
repo.delete_head(ref, force=True)
else:
zuul_refs_to_keep.append(Repo.refNameToZuulRef(ref))
# Delete local zuul refs when the related branch no longer exists
for ref in zuul_refs.keys():
if ref in zuul_refs_to_keep:
continue
if log:
log.debug("Delete stale Zuul ref %s", ref)
else:
messages.append("Delete stale Zuul ref {}".format(ref))
Repo._deleteRef(ref, repo)
Repo._cleanup_leaked_rebase_dirs(local_path, log, messages)
# Note: Before git 2.13 deleting a a ref foo/bar leaves an empty
# directory foo behind that will block creating the reference foo
# in the future. As a workaround we must clean up empty directories
# in .git/refs.
if repo.git.version_info[:2] < (2, 13):
Repo._cleanup_leaked_ref_dirs(local_path, log, messages)
# Update our local heads to match the remote
for ref, commit in origin_refs.items():
if ref == 'HEAD':
continue
repo.create_head('refs/heads/' + ref,
commit,
force=True)
return messages
def reset(self, zuul_event_id=None, build=None, process_worker=None):
log = get_annotated_logger(self.log, zuul_event_id, build=build)
log.debug("Resetting repository %s", self.local_path)
self.createRepoObject(zuul_event_id, build=build)
try:
if process_worker is None:
self._reset(self.local_path, self.env, log)
else:
job = process_worker.submit(
Repo._reset, self.local_path, self.env)
messages = job.result()
for message in messages:
log.debug(message)
except Exception:
shutil.rmtree(self.local_path)
raise
def getBranchHead(self, branch, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
branch_head = repo.heads[branch]
return branch_head.commit
def setBranchHead(self, branch, head_ref, repo=None, zuul_event_id=None):
if repo is None:
repo = self.createRepoObject(zuul_event_id)
repo.create_head(branch, head_ref, force=True)
def hasBranch(self, branch, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
origin = repo.remotes.origin
return branch in origin.refs
def getBranches(self, zuul_event_id=None):
# TODO(jeblair): deprecate with override-branch; replaced by
# getRefs().
repo = self.createRepoObject(zuul_event_id)
return [x.name for x in repo.heads]
def getRef(self, refname, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
if refname not in repo.refs:
return None
ref = repo.refs[refname]
return ref
def getRefs(self, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
return Repo._getRefs(repo)
@staticmethod
def _getRefs(repo):
refs = repo.git.for_each_ref(
'--format=%(objectname) %(refname)'
)
for ref in refs.splitlines():
parts = ref.split(" ")
if len(parts) == 2:
commit, ref = parts
yield commit, ref
def setRef(self, path, hexsha, repo=None, zuul_event_id=None):
ref_log = get_annotated_logger(
logging.getLogger("zuul.Repo.Ref"), zuul_event_id)
ref_log.debug("Create reference %s at %s in %s",
path, hexsha, self.local_path)
if repo is None:
repo = self.createRepoObject(zuul_event_id)
self._setRef(path, hexsha, repo)
@staticmethod
def _setRef(path, hexsha, repo):
binsha = gitdb.util.to_bin_sha(hexsha)
obj = git.objects.Object.new_from_sha(repo, binsha)
git.refs.Reference.create(repo, path, obj, force=True)
return 'Created reference %s at %s in %s' % (
path, hexsha, repo.git_dir)
def setRefs(self, refs, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
ref_log = get_annotated_logger(
logging.getLogger("zuul.Repo.Ref"), zuul_event_id)
self._setRefs(repo, refs, log=ref_log)
@staticmethod
def setRefsAsync(local_path, refs):
repo = git.Repo(local_path)
messages = Repo._setRefs(repo, refs)
return messages
@staticmethod
def _setRefs(repo, refs, log=None):
messages = []
# Rewrite packed refs with our content. In practice, this
# should take care of almost every ref in the repo, except
# maybe HEAD and master.
refs_path = f"{repo.git_dir}/packed-refs"
encoding = sys.getfilesystemencoding()
with open(refs_path, 'wb') as f:
f.write(b'# pack-refs with: peeled fully-peeled sorted \n')
sorted_paths = sorted(refs.keys())
msg = f"Setting {len(sorted_paths)} refs in {repo.git_dir}"
if log:
log.debug(msg)
else:
messages.append(msg)
for path in sorted_paths:
hexsha = refs[path]
try:
# Attempt a lookup for the side effect of
# verifying the object exists.
binsha = gitdb.util.to_bin_sha(hexsha)
oinfo = repo.odb.info(binsha)
f.write(f'{hexsha} {path}\n'.encode(encoding))
if oinfo.type == b'tag':
# We are an annotated or signed tag which
# refers to another commit. We must handle this
# special case when packing refs
tagobj = git.Object.new_from_sha(repo, binsha)
tagsha = tagobj.object.hexsha
f.write(f'^{tagsha}\n'.encode(encoding))
except ValueError:
# If the object does not exist, skip setting it.
msg = (
f"Unable to resolve reference {path} at {hexsha}"
f" in {repo.git_dir}"
)
if log:
log.warning(msg)
else:
messages.append(msg)
# Delete all the loose refs
for dname in ('remotes', 'tags', 'heads'):
path = f"{repo.git_dir}/refs/{dname}"
if os.path.exists(path):
shutil.rmtree(path)
return messages
def setRemoteRef(self, branch, rev, zuul_event_id=None):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.createRepoObject(zuul_event_id)
try:
log.debug("Updating remote reference origin/%s to %s", branch, rev)
repo.remotes.origin.refs[branch].commit = rev
except IndexError:
log.warning("No remote ref found for branch %s, creating", branch)
Repo._setRef(f"refs/remotes/origin/{branch}", str(rev), repo)
def deleteRef(self, path, repo=None, zuul_event_id=None):
ref_log = get_annotated_logger(
logging.getLogger("zuul.Repo.Ref"), zuul_event_id)
if repo is None:
repo = self.createRepoObject(zuul_event_id)
ref_log.debug("Delete reference %s", path)
Repo._deleteRef(path, repo)
@staticmethod
def _deleteRef(path, repo):
git.refs.SymbolicReference.delete(repo, path)
return "Deleted reference %s" % path
def checkout(self, ref, zuul_event_id=None):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.createRepoObject(zuul_event_id)
# NOTE(pabelanger): We need to check for detached repo head, otherwise
# gitpython will raise an exception if we access the reference.
if not repo.head.is_detached and repo.head.reference == ref:
log.debug("Repo is already at %s" % ref)
else:
log.debug("Checking out %s" % ref)
try:
self._checkout(repo, ref)
except Exception:
lock_path = f"{self.local_path}/.git/index.lock"
if os.path.isfile(lock_path):
log.warning("Deleting stale index.lock file: %s",
lock_path)
os.unlink(lock_path)
# Retry the checkout
self._checkout(repo, ref)
else:
raise
return repo.head.commit
def _checkout(self, repo, ref):
# Perform a hard reset to the correct ref before checking out so
# that we clean up anything that might be left over from a merge
# while still only preparing the working copy once.
repo.head.reference = ref
repo.head.reset(working_tree=True)
repo.git.clean('-x', '-f', '-d')
repo.git.checkout(ref)
@staticmethod
def _getTimestampEnv(timestamp):
if timestamp:
return {
'GIT_COMMITTER_DATE': str(int(timestamp)) + '+0000',
'GIT_AUTHOR_DATE': str(int(timestamp)) + '+0000',
}
return {}
def merge(self, ref, strategy=None, zuul_event_id=None, timestamp=None):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.createRepoObject(zuul_event_id)
args = []
if strategy:
args += ['-s', strategy]
args.append('FETCH_HEAD')
self.fetch(ref, zuul_event_id=zuul_event_id)
log.debug("Merging %s with args %s", ref, args)
with repo.git.custom_environment(**self._getTimestampEnv(timestamp)):
# Use a custom message to avoid introducing
# merger/executor path details
repo.git.merge(message=f"Merge '{ref}'", *args)
return repo.head.commit
def squashMerge(self, item, zuul_event_id=None, timestamp=None):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.createRepoObject(zuul_event_id)
args = ['--squash', 'FETCH_HEAD']
ref = item['ref']
self.fetch(ref, zuul_event_id=zuul_event_id)
log.debug("Squash-Merging %s with args %s", ref, args)
with repo.git.custom_environment(**self._getTimestampEnv(timestamp)):
repo.git.merge(*args)
# Use a custom message to avoid introducing
# merger/executor path details
repo.git.commit(
message='Merge change %s,%s' % (
item['number'], item['patchset']),
allow_empty=True)
return repo.head.commit
def rebaseMerge(self, item, base, zuul_event_id=None, timestamp=None):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.createRepoObject(zuul_event_id)
args = [base]
ref = item['ref']
self.fetch(ref, zuul_event_id=zuul_event_id)
log.debug("Rebasing %s with args %s", ref, args)
repo.git.checkout('FETCH_HEAD')
with repo.git.custom_environment(**self._getTimestampEnv(timestamp)):
try:
repo.git.rebase(*args)
except Exception:
repo.git.rebase(abort=True)
raise
return repo.head.commit
def cherryPick(self, ref, zuul_event_id=None, timestamp=None):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.createRepoObject(zuul_event_id)
self.fetch(ref, zuul_event_id=zuul_event_id)
fetch_head = repo.commit("FETCH_HEAD")
if len(fetch_head.parents) > 1:
args = ["-s", "resolve", "FETCH_HEAD"]
log.debug("Merging %s with args %s instead of cherry-picking",
ref, args)
with repo.git.custom_environment(
**self._getTimestampEnv(timestamp)):
# Use a custom message to avoid introducing
# merger/executor path details
repo.git.merge(message=f"Merge '{ref}'", *args)
else:
log.debug("Cherry-picking %s", ref)
# Git doesn't have an option to ignore commits that are already
# applied to the working tree when cherry-picking, so pass the
# --keep-redundant-commits option, which will cause it to make an
# empty commit
with repo.git.custom_environment(
**self._getTimestampEnv(timestamp)):
repo.git.cherry_pick("FETCH_HEAD", keep_redundant_commits=True)
# If the newly applied commit is empty, it means either:
# 1) The commit being cherry-picked was empty, in which the empty
# commit should be kept
# 2) The commit being cherry-picked was already applied to the
# tree, in which case the empty commit should be backed out
head = repo.commit("HEAD")
parent = head.parents[0]
if not any(head.diff(parent)) and \
any(fetch_head.diff(fetch_head.parents[0])):
log.debug("%s was already applied. Removing it", ref)
self._checkout(repo, parent)
return repo.head.commit
def fetch(self, ref, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
# NOTE: The following is currently not applicable, but if we
# switch back to fetch methods from GitPython, we need to
# consider it:
# The git.remote.fetch method may read in git progress info and
# interpret it improperly causing an AssertionError. Because the
# data was fetched properly subsequent fetches don't seem to fail.
# So try again if an AssertionError is caught.
self._git_fetch(repo, 'origin', zuul_event_id, ref=ref)
def revParse(self, ref, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
return repo.git.rev_parse(ref)
def fetchFrom(self, repository, ref, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
self._git_fetch(repo, repository, zuul_event_id, ref=ref)
def push(self, local, remote, zuul_event_id=None):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.createRepoObject(zuul_event_id)
log.debug("Pushing %s:%s to %s", local, remote, self.remote_url)
repo.remotes.origin.push('%s:%s' % (local, remote))
def update(self, zuul_event_id=None, build=None):
log = get_annotated_logger(self.log, zuul_event_id, build=build)
repo = self.createRepoObject(zuul_event_id, build=build)
log.debug("Updating repository %s" % self.local_path)
if repo.git.version_info[:2] < (1, 9):
# Before 1.9, 'git fetch --tags' did not include the
# behavior covered by 'git --fetch', so we run both
# commands in that case. Starting with 1.9, 'git fetch
# --tags' is all that is necessary. See
# https://github.com/git/git/blob/master/Documentation/RelNotes/1.9.0.txt#L18-L20
self._git_fetch(repo, 'origin', zuul_event_id)
self._git_fetch(repo, 'origin', zuul_event_id, tags=True,
prune=True, prune_tags=True)
def isUpdateNeeded(self, project_repo_state, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
refs = [x.path for x in repo.refs]
for ref, rev in project_repo_state.items():
# Check that each ref exists and that each commit exists
if ref not in refs:
return True
try:
repo.commit(rev)
except Exception:
# GitPython throws an error if a revision does not
# exist
return True
return False
def getFiles(self, files, dirs=[], branch=None, commit=None,
zuul_event_id=None):
log = get_annotated_logger(self.log, zuul_event_id)
ret = {}
repo = self.createRepoObject(zuul_event_id)
if branch:
head = repo.heads[branch].commit
else:
head = repo.commit(commit)
log.debug("Getting files for %s at %s", self.local_path, head.hexsha)
tree = head.tree
for fn in files:
if fn in tree:
if tree[fn].type != 'blob':
log.warning(
"%s: object %s is not a blob", self.local_path, fn)
ret[fn] = tree[fn].data_stream.read().decode('utf8')
else:
ret[fn] = None
if dirs:
for dn in dirs:
try:
sub_tree = tree[dn]
except KeyError:
continue
if sub_tree.type != "tree":
continue
# Some people like to keep playbooks, etc. grouped
# under their zuul config dirs; record the leading
# directories of any .zuul.ignore files and prune them
# from the config read.
to_ignore = []
for blob in sub_tree.traverse():
if blob.path.endswith(".zuul.ignore"):
to_ignore.append(os.path.split(blob.path)[0])
def _ignored(blob):
for prefix in to_ignore:
if blob.path.startswith(prefix):
return True
return False
for blob in sub_tree.traverse():
if not _ignored(blob) and blob.path.endswith(".yaml"):
ret[blob.path] = blob.data_stream.read().decode(
'utf-8')
return ret
def getFilesChanges(self, branch, tosha=None, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
self.fetch(branch, zuul_event_id=zuul_event_id)
head = repo.commit(
self.revParse('FETCH_HEAD', zuul_event_id=zuul_event_id))
files = set()
if tosha:
# When "tosha" is the target branch, the result of diff() correctly
# excluds the files whose changes are reverted between the commits.
# But it may also include the files that are not changed in the
# referenced commit(s). This can happen, e.g. if the target branch
# has diverged from the feature branch.
# The idea is to use this result to filter out the files whose
# changes are reverted between the commits.
diff_files = set()
head_commit = repo.commit(head.hexsha)
diff_index = head_commit.diff(tosha)
diff_files.update((item.a_path for item in diff_index))
commit_diff = "{}..{}".format(tosha, head.hexsha)
for cmt in repo.iter_commits(commit_diff, no_merges=True):
files.update(f for f in cmt.stats.files.keys()
if f in diff_files)
else:
files.update(head.stats.files.keys())
return list(files)
def deleteRemote(self, remote, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
repo.delete_remote(repo.remotes[remote])
def setRemoteUrl(self, url, zuul_event_id=None):
if self.remote_url == url:
return
log = get_annotated_logger(self.log, zuul_event_id)
log.debug("Set remote url to %s", redact_url(url))
try:
# Update the remote URL as it is used for the clone if the
# repo doesn't exist.
self.remote_url = url
self._git_set_remote_url(
self.createRepoObject(zuul_event_id), self.remote_url)
except Exception:
# Clear out the stored remote URL so we will always set
# the Git URL after a failed attempt. This prevents us from
# using outdated credentials that might still be stored in
# the Git config as part of the URL.
self.remote_url = None
raise
def mapLine(self, commit, filename, lineno, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
# Trace the specified line back to the specified commit and
# return the line number in that commit.
cur_commit = None
out = repo.git.log(L='%s,%s:%s' % (lineno, lineno, filename))
for l in out.split('\n'):
if cur_commit is None:
m = self.commit_re.match(l)
if m:
if m.group(1) == commit:
cur_commit = commit
continue
m = self.diff_re.match(l)
if m:
return int(m.group(1))
return None
def contains(self, hexsha, zuul_event_id=None):
repo = self.createRepoObject(zuul_event_id)
log = get_annotated_logger(self.log, zuul_event_id)
try:
branches = repo.git.branch(contains=hexsha, color='never')
except git.GitCommandError as e:
if e.status == 129:
log.debug("Found commit %s in no branches", hexsha)
return []
branches = [x.replace('*', '').strip() for x in branches.split('\n')]
branches = [x for x in branches if x != '(no branch)']
log.debug("Found commit %s in branches: %s", hexsha, branches)
return branches
class MergerTree:
"""
A tree structure for quickly determining if a repo collides with
another in the same merger workspace.
Each node is a dictionary represents a path element. The keys are
child path elements and their values are either another dictionary
for another node, or None if the child node is a git repo.
"""
def __init__(self):
self.root = {}
def add(self, path):
parts = path.split('/')
root = self.root
for i, part in enumerate(parts[:-1]):
root = root.setdefault(part, {})
if root is None:
other = '/'.join(parts[:i])
raise Exception(f"Repo {path} collides with {other}")
last = parts[-1]
if last in root:
raise Exception(f"Repo {path} collides with "
"an existing repo with a longer path")
root[last] = None
class Merger(object):
def __init__(self, working_root, connections, zk_client, email,
username, speed_limit, speed_time, cache_root=None,
logger=None, execution_context=False, git_timeout=300,
scheme=None, cache_scheme=None):
self.logger = logger
if logger is None:
self.log = logging.getLogger("zuul.Merger")
else:
self.log = logger
self.repos = {}
self.working_root = working_root
os.makedirs(working_root, exist_ok=True)
self.connections = connections
self.zk_client = zk_client
self.email = email
self.username = username
self.speed_limit = speed_limit
self.speed_time = speed_time
self.git_timeout = git_timeout
self.cache_root = cache_root
self.scheme = scheme or zuul.model.SCHEME_UNIQUE
self.cache_scheme = cache_scheme or zuul.model.SCHEME_UNIQUE
# Flag to determine if the merger is used for preparing repositories
# for job execution. This flag can be used to enable executor specific
# behavior e.g. to keep the 'origin' remote intact.
self.execution_context = execution_context
# A tree of repos in our working root for detecting collisions
self.repo_roots = MergerTree()
# If we're not in an execution context, then check to see if
# our working root needs a "schema" upgrade.
if not execution_context:
self._upgradeRootScheme()
def _upgradeRootScheme(self):
# If the scheme for the root directory has changed, delete all
# the repos so they can be re-cloned.
try:
with open(os.path.join(self.working_root,
'.zuul_merger_scheme')) as f:
scheme = f.read().strip()
except FileNotFoundError:
# The previous default was golang
scheme = zuul.model.SCHEME_GOLANG
if scheme == self.scheme:
return
if os.listdir(self.working_root):
self.log.info(f"Existing merger scheme {scheme} does not match "
f"requested scheme {self.scheme}, deleting merger "
"root (repositories will be re-cloned).")
shutil.rmtree(self.working_root)
os.makedirs(self.working_root)
with open(os.path.join(self.working_root,
'.zuul_merger_scheme'), 'w') as f:
f.write(self.scheme)
def _addProject(self, hostname, connection_name, project_name, url, sshkey,
zuul_event_id, process_worker=None, retry_timeout=None):
repo = None
key = '/'.join([hostname, project_name])
try:
path = os.path.join(self.working_root,
strings.workspace_project_path(
hostname, project_name, self.scheme))
self.repo_roots.add(path)
if self.cache_root:
cache_path = os.path.join(
self.cache_root,
strings.workspace_project_path(
hostname, project_name, self.cache_scheme))
else:
cache_path = None
repo = Repo(
url, path, self.email, self.username, self.speed_limit,
self.speed_time, sshkey=sshkey, cache_path=cache_path,
logger=self.logger, git_timeout=self.git_timeout,
zuul_event_id=zuul_event_id, retry_timeout=retry_timeout,
skip_refs=self.execution_context)
self.repos[key] = repo
except Exception:
log = get_annotated_logger(self.log, zuul_event_id)
log.exception("Unable to add project %s/%s",
hostname, project_name)
return repo
def getRepo(self, connection_name, project_name,
zuul_event_id=None, process_worker=None,
keep_remote_url=False):
source = self.connections.getSource(connection_name)
project = source.getProject(project_name)
hostname = project.canonical_hostname
url = source.getGitUrl(project)
retry_timeout = source.getRetryTimeout(project)
key = '/'.join([hostname, project_name])
if key in self.repos:
repo = self.repos[key]
if not keep_remote_url:
repo.setRemoteUrl(url)
return repo
sshkey = self.connections.connections.get(connection_name).\
connection_config.get('sshkey')
if not url:
raise Exception("Unable to set up repo for project %s/%s"
" without a url" %
(connection_name, project_name,))
return self._addProject(hostname, connection_name, project_name, url,
sshkey, zuul_event_id,
process_worker=process_worker,
retry_timeout=retry_timeout)
def updateRepo(self, connection_name, project_name, repo_state=None,
zuul_event_id=None, build=None, process_worker=None):
"""Fetch from origin if needed
If repo_state is None, then this will always git fetch.
If repo_state is provided, then this may no-op if
the shas specified by repo_state are already present.
"""
log = get_annotated_logger(self.log, zuul_event_id, build=build)
repo = self.getRepo(connection_name, project_name,
zuul_event_id=zuul_event_id)
if repo_state:
projects = repo_state.get(connection_name, {})
project_repo_state = projects.get(project_name, None)
else:
project_repo_state = None
try:
# Check if we need an update if we got a repo_state and
# our project appears in it (otherwise we always update).
if project_repo_state and not repo.isUpdateNeeded(
project_repo_state, zuul_event_id=zuul_event_id):
log.info("Skipping updating local repository %s/%s",
connection_name, project_name)
else:
log.info("Updating local repository %s/%s",
connection_name, project_name)
repo.update(zuul_event_id=zuul_event_id, build=build)
except Exception:
log.exception("Unable to update %s/%s",
connection_name, project_name)
raise
def checkoutBranch(self, connection_name, project_name, branch,
repo_state=None, zuul_event_id=None,
process_worker=None):
"""Check out a branch
Call Merger.updateRepo() first. This does not reset the repo,
and is expected to be called only after a fresh clone.
"""
log = get_annotated_logger(self.log, zuul_event_id)
log.info("Checking out %s/%s branch %s",
connection_name, project_name, branch)
repo = self.getRepo(connection_name, project_name,
process_worker=process_worker,
zuul_event_id=zuul_event_id)
# We don't need to reset because this is only called by the
# executor after a clone.
if repo_state:
self._restoreRepoState(connection_name, project_name, repo,
repo_state, zuul_event_id,
process_worker=process_worker)
return repo.checkout(branch, zuul_event_id=zuul_event_id)
def _saveRepoState(self, connection_name, project_name, repo,
repo_state, recent, branches):
projects = repo_state.setdefault(connection_name, {})
project = projects.setdefault(project_name, {})
for commit, ref in repo.getRefs():
if ref.startswith('refs/zuul/'):
continue
if ref.startswith('refs/remotes/'):
continue
if ref.startswith('refs/heads/'):
branch = ref[len('refs/heads/'):]
if branches is not None and branch not in branches:
continue
key = (connection_name, project_name, branch)
if key not in recent:
recent[key] = commit
project[ref] = commit
def _alterRepoState(self, connection_name, project_name,
repo_state, path, hexsha):
projects = repo_state.setdefault(connection_name, {})
project = projects.setdefault(project_name, {})
if hexsha == NULL_REF:
if path in project:
del project[path]
else:
project[path] = hexsha
def _restoreRepoState(self, connection_name, project_name, repo,
repo_state, zuul_event_id,
process_worker=None):
log = get_annotated_logger(self.log, zuul_event_id)
projects = repo_state.get(connection_name, {})
project = projects.get(project_name, {})
if not project:
# We don't have a state for this project.
return
log.debug("Restore repo state for project %s/%s",
connection_name, project_name)
if process_worker is None:
repo.setRefs(project, zuul_event_id=zuul_event_id)
else:
job = process_worker.submit(
Repo.setRefsAsync, repo.local_path, project)
messages = job.result()
ref_log = get_annotated_logger(
logging.getLogger("zuul.Repo.Ref"), zuul_event_id)
for message in messages:
ref_log.debug(message)
def _mergeChange(self, item, base, zuul_event_id):
log = get_annotated_logger(self.log, zuul_event_id)
repo = self.getRepo(item['connection'], item['project'],
zuul_event_id=zuul_event_id)
try:
repo.checkout(base, zuul_event_id=zuul_event_id)
except Exception:
log.exception("Unable to checkout %s", base)
return None, None
timestamp = item.get('configured_time')
try:
mode = item['merge_mode']
if mode == zuul.model.MERGER_MERGE:
commit = repo.merge(item['ref'], zuul_event_id=zuul_event_id,
timestamp=timestamp)
elif mode == zuul.model.MERGER_MERGE_RESOLVE:
commit = repo.merge(item['ref'], 'resolve',
zuul_event_id=zuul_event_id,
timestamp=timestamp)
elif mode == zuul.model.MERGER_MERGE_RECURSIVE:
commit = repo.merge(item['ref'], 'recursive',
zuul_event_id=zuul_event_id,
timestamp=timestamp)
elif mode == zuul.model.MERGER_MERGE_ORT:
commit = repo.merge(item['ref'], 'ort',
zuul_event_id=zuul_event_id,
timestamp=timestamp)
elif mode == zuul.model.MERGER_CHERRY_PICK:
commit = repo.cherryPick(item['ref'],
zuul_event_id=zuul_event_id,
timestamp=timestamp)
elif mode == zuul.model.MERGER_SQUASH_MERGE:
commit = repo.squashMerge(
item, zuul_event_id=zuul_event_id,
timestamp=timestamp)
elif mode == zuul.model.MERGER_REBASE:
commit = repo.rebaseMerge(
item, base, zuul_event_id=zuul_event_id,
timestamp=timestamp)
else:
raise Exception("Unsupported merge mode: %s" % mode)
except git.GitCommandError:
# Log git exceptions at debug level because they are
# usually benign merge conflicts
log.debug("Unable to merge %s", item, exc_info=True)
return None, None
except Exception:
log.exception("Exception while merging a change:")
return None, None
orig_commit = repo.revParse('FETCH_HEAD')
return orig_commit, commit
def _mergeItem(self, item, recent, repo_state, zuul_event_id,
branches=None, process_worker=None):
log = get_annotated_logger(self.log, zuul_event_id)
log.debug("Processing ref %s for project %s/%s / %s uuid %s" %
(item['ref'], item['connection'],
item['project'], item['branch'],
item['buildset_uuid']))
repo = self.getRepo(item['connection'], item['project'])
key = (item['connection'], item['project'], item['branch'])
# We need to merge the change
# Get the most recent commit for this project-branch
base = recent.get(key)
if not base:
# There is none, so use the branch tip
# we need to reset here in order to call getBranchHead
log.debug("No base commit found for %s" % (key,))
try:
repo.reset(zuul_event_id=zuul_event_id,
process_worker=process_worker)
except BrokenProcessPool:
raise
except Exception:
log.exception("Unable to reset repo %s" % repo)
return None, None
self._restoreRepoState(item['connection'], item['project'], repo,
repo_state, zuul_event_id,
process_worker=process_worker)
base = repo.getBranchHead(item['branch'])
# Save the repo state so that later mergers can repeat
# this process.
self._saveRepoState(item['connection'], item['project'], repo,
repo_state, recent, branches)
else:
log.debug("Found base commit %s for %s" % (base, key,))
if self.execution_context:
# Set origin branch to the rev of the current (speculative) base.
# This allows tools to determine the commits that are part of a
# change by looking at origin/master..master.
repo.setRemoteRef(item['branch'], base,
zuul_event_id=zuul_event_id)
# Merge the change
orig_commit, commit = self._mergeChange(item, base, zuul_event_id)
if not commit:
return None, None
# Store this commit as the most recent for this project-branch
recent[key] = commit.hexsha
# Make sure to have a local ref that points to the most recent
# (intermediate) speculative state of a branch, so commits are not
# garbage collected. The branch name is hashed to not cause any
# problems with empty directories in case of branch names containing
# slashes. In order to prevent issues with Git garbage collection
# between merger and executor jobs, we create refs in "refs/zuul"
# instead of updating local branch heads.
repo.setRef(Repo.refNameToZuulRef(item["branch"]),
commit.hexsha, zuul_event_id=zuul_event_id)
return orig_commit, commit
def mergeChanges(self, items, files=None, dirs=None, repo_state=None,
repo_locks=None, branches=None, zuul_event_id=None,
process_worker=None, errors=None):
"""Merge changes
Call Merger.updateRepo() first.
"""
# _mergeItem calls reset as necessary.
log = get_annotated_logger(self.log, zuul_event_id)
# connection+project+branch -> commit
recent = {}
commit = None
# tuple(connection, project, branch) -> dict(config state)
read_files = OrderedDict()
# connection -> project -> ref -> commit
if repo_state is None:
repo_state = {}
for item in items:
# If we're in the executor context we have the repo_locks object
# and perform per repo locking.
if repo_locks is not None:
lock = repo_locks.getRepoLock(
item['connection'], item['project'])
else:
lock = nullcontext()
err_msg = (
f"Error merging {item['connection']}/{item['project']} "
f"for {item['number']},{item['patchset']}"
)
with lock:
log.debug("Merging for change %s,%s" %
(item["number"], item["patchset"]))
try:
orig_commit, commit = self._mergeItem(
item, recent, repo_state, zuul_event_id,
branches=branches,
process_worker=process_worker)
except BrokenProcessPool:
raise
except Exception:
self.log.exception("Error merging item %s", item)
if errors is not None:
errors.append(err_msg)
if not commit:
if errors is not None:
errors.append(err_msg)
return None
if files or dirs:
repo = self.getRepo(item['connection'], item['project'])
repo_files = repo.getFiles(files, dirs, commit=commit)
key = item['connection'], item['project'], item['branch']
read_files[key] = dict(
connection=item['connection'],
project=item['project'],
branch=item['branch'],
files=repo_files)
return (
commit.hexsha, list(read_files.values()), repo_state, recent,
orig_commit
)
def setRepoState(self, connection_name, project_name, repo_state,
zuul_event_id=None, process_worker=None):
"""Set the repo state
Call Merger.updateRepo() first.
"""
repo = self.getRepo(connection_name, project_name,
zuul_event_id=zuul_event_id)
self._restoreRepoState(connection_name, project_name, repo,
repo_state, zuul_event_id)
def getRepoState(self, items, repo_locks, branches=None):
"""Gets the repo state for items.
This will perform repo updates as needed, so there is no need
to call Merger.updateRepo() first.
"""
# Generally this will be called in any non-change pipeline. We
# will return the repo state for each item, but manipulated with
# any information in the item (eg, if it creates a ref, that
# will be in the repo state regardless of the actual state).
seen = set()
recent = {}
repo_state = {}
for item in items:
# If we're in the executor context we need to lock the repo.
# If not repo_locks will give us a fake lock.
lock = repo_locks.getRepoLock(item['connection'], item['project'])
with lock:
repo = self.getRepo(item['connection'], item['project'])
key = (item['connection'], item['project'])
if key not in seen:
try:
repo.update()
repo.reset()
seen.add(key)
except Exception:
self.log.exception("Unable to reset repo %s" % repo)
return (False, {}, [])
self._saveRepoState(item['connection'], item['project'],
repo, repo_state, recent, branches)
if item.get('newrev'):
# This is a ref update rather than a branch tip, so make
# sure our returned state includes this change.
self._alterRepoState(
item['connection'], item['project'], repo_state,
item['ref'], item['newrev'])
item = items[-1]
# A list of branch names the last item appears in.
item_in_branches = []
if item.get('newrev'):
lock = repo_locks.getRepoLock(item['connection'], item['project'])
with lock:
repo = self.getRepo(item['connection'], item['project'])
item_in_branches = repo.contains(item['newrev'])
return (True, repo_state, item_in_branches)
def getFiles(self, connection_name, project_name, branch, files, dirs=[]):
"""Get file contents on branch.
Call Merger.updateRepo() first to make sure the repo is up to
date.
"""
# We don't update the repo so that it can happen outside the
# lock.
repo = self.getRepo(connection_name, project_name)
# TODO: why is reset required here?
repo.reset()
# This does not fetch, update, or reset, it operates on the
# working state.
return repo.getFiles(files, dirs, branch=branch)
def getFilesChanges(self, connection_name, project_name, branch,
tosha=None, zuul_event_id=None):
"""Get a list of files changed in one or more commits
Gets files changed between tosha and branch (or just the
commit on branch if tosha is not specified).
Call Merger.updateRepo() first to make sure the repo is up to
date.
"""
# Note, the arguments to this method should be reworked. We
# fetch branch, and therefore it is typically actually a
# change ref. tosha is typically the branch name.
repo = self.getRepo(connection_name, project_name,
zuul_event_id=zuul_event_id)
# This performs a fetch, and therefore update/reset are not
# required.
return repo.getFilesChanges(branch, tosha, zuul_event_id=zuul_event_id)