buildbot-nix/buildbot_nix/__init__.py
raito 9a67a1f2cc nix/coordinator: enable reporting to Gerrit
Signed-off-by: Raito Bezarius <raito@lix.systems>
2024-03-03 02:58:25 +01:00

749 lines
25 KiB
Python

import json
import multiprocessing
import os
import sys
import uuid
from collections import defaultdict
from collections.abc import Generator
from dataclasses import dataclass
from pathlib import Path
from typing import TYPE_CHECKING, Any
from buildbot.configurators import ConfiguratorBase
from buildbot.plugins import reporters, schedulers, secrets, steps, util, worker
from buildbot.process import buildstep, logobserver, remotecommand
from buildbot.process.project import Project
from buildbot.process.properties import Interpolate, Properties
from buildbot.process.results import ALL_RESULTS, statusToString
from buildbot.steps.trigger import Trigger
from buildbot.util import asyncSleep
from buildbot.www.authz.endpointmatchers import EndpointMatcherBase, Match
from buildbot.www.oauth2 import OAuth2Auth
from buildbot.changes.gerritchangesource import GerritChangeSource
if TYPE_CHECKING:
from buildbot.process.log import Log
from twisted.internet import defer, threads
from twisted.logger import Logger
from twisted.python.failure import Failure
from .github_projects import (
slugify_project_name,
)
SKIPPED_BUILDER_NAME = "skipped-builds"
log = Logger()
class LixSystemsOAuth2(OAuth2Auth):
name = 'Lix'
faIcon = 'fa-login'
resourceEndpoint = "https://identity.lix.systems"
# is passing scope necessary?
authUri = 'https://identity.lix.systems/realms/lix-project/protocol/openid-connect/auth'
tokenUri = 'https://identity.lix.systems/realms/lix-project/protocol/openid-connect/token'
class BuildbotNixError(Exception):
pass
@dataclass
class GerritProject:
# `project` field.
name: str
class BuildTrigger(Trigger):
"""Dynamic trigger that creates a build for every attribute."""
def __init__(
self,
builds_scheduler: str,
skipped_builds_scheduler: str,
jobs: list[dict[str, Any]],
**kwargs: Any,
) -> None:
if "name" not in kwargs:
kwargs["name"] = "trigger"
self.jobs = jobs
self.config = None
self.builds_scheduler = builds_scheduler
self.skipped_builds_scheduler = skipped_builds_scheduler
Trigger.__init__(
self,
waitForFinish=True,
schedulerNames=[builds_scheduler, skipped_builds_scheduler],
haltOnFailure=True,
flunkOnFailure=True,
sourceStamps=[],
alwaysUseLatest=False,
updateSourceStamp=False,
**kwargs,
)
def createTriggerProperties(self, props: Any) -> Any: # noqa: N802
return props
def getSchedulersAndProperties(self) -> list[tuple[str, Properties]]: # noqa: N802
build_props = self.build.getProperties()
source = f"nix-eval-lix"
triggered_schedulers = []
for job in self.jobs:
attr = job.get("attr", "eval-error")
name = attr
name = f"checks.{name}"
error = job.get("error")
props = Properties()
props.setProperty("virtual_builder_name", name, source)
props.setProperty("status_name", f"nix-build .#checks.{attr}", source)
props.setProperty("virtual_builder_tags", "", source)
if error is not None:
props.setProperty("error", error, source)
triggered_schedulers.append((self.skipped_builds_scheduler, props))
continue
if job.get("isCached"):
triggered_schedulers.append((self.skipped_builds_scheduler, props))
continue
drv_path = job.get("drvPath")
system = job.get("system")
out_path = job.get("outputs", {}).get("out")
build_props.setProperty(f"{attr}-out_path", out_path, source)
build_props.setProperty(f"{attr}-drv_path", drv_path, source)
props.setProperty("attr", attr, source)
props.setProperty("system", system, source)
props.setProperty("drv_path", drv_path, source)
props.setProperty("out_path", out_path, source)
# we use this to identify builds when running a retry
props.setProperty("build_uuid", str(uuid.uuid4()), source)
triggered_schedulers.append((self.builds_scheduler, props))
return triggered_schedulers
def getCurrentSummary(self) -> dict[str, str]: # noqa: N802
"""The original build trigger will the generic builder name `nix-build` in this case, which is not helpful"""
if not self.triggeredNames:
return {"step": "running"}
summary = []
if self._result_list:
for status in ALL_RESULTS:
count = self._result_list.count(status)
if count:
summary.append(
f"{self._result_list.count(status)} {statusToString(status, count)}",
)
return {"step": f"({', '.join(summary)})"}
class NixEvalCommand(buildstep.ShellMixin, steps.BuildStep):
"""Parses the output of `nix-eval-jobs` and triggers a `nix-build` build for
every attribute.
"""
def __init__(self, supported_systems: list[str], **kwargs: Any) -> None:
kwargs = self.setupShellMixin(kwargs)
super().__init__(**kwargs)
self.observer = logobserver.BufferLogObserver()
self.addLogObserver("stdio", self.observer)
self.supported_systems = supported_systems
@defer.inlineCallbacks
def run(self) -> Generator[Any, object, Any]:
# run nix-eval-jobs --flake .#checks to generate the dict of stages
cmd: remotecommand.RemoteCommand = yield self.makeRemoteShellCommand()
yield self.runCommand(cmd)
# if the command passes extract the list of stages
result = cmd.results()
if result == util.SUCCESS:
# create a ShellCommand for each stage and add them to the build
jobs = []
for line in self.observer.getStdout().split("\n"):
if line != "":
try:
job = json.loads(line)
except json.JSONDecodeError as e:
msg = f"Failed to parse line: {line}"
raise BuildbotNixError(msg) from e
jobs.append(job)
build_props = self.build.getProperties()
filtered_jobs = []
for job in jobs:
system = job.get("system")
if not system or system in self.supported_systems: # report eval errors
filtered_jobs.append(job)
self.build.addStepsAfterCurrentStep(
[
BuildTrigger(
builds_scheduler=f"lix-nix-build",
skipped_builds_scheduler=f"lix-nix-skipped-build",
name="build flake",
jobs=filtered_jobs,
),
],
)
return result
# FIXME this leaks memory... but probably not enough that we care
class RetryCounter:
def __init__(self, retries: int) -> None:
self.builds: dict[uuid.UUID, int] = defaultdict(lambda: retries)
def retry_build(self, build_id: uuid.UUID) -> int:
retries = self.builds[build_id]
if retries > 1:
self.builds[build_id] = retries - 1
return retries
return 0
# For now we limit this to two. Often this allows us to make the error log
# shorter because we won't see the logs for all previous succeeded builds
RETRY_COUNTER = RetryCounter(retries=2)
class EvalErrorStep(steps.BuildStep):
"""Shows the error message of a failed evaluation."""
@defer.inlineCallbacks
def run(self) -> Generator[Any, object, Any]:
error = self.getProperty("error")
attr = self.getProperty("attr")
# show eval error
error_log: Log = yield self.addLog("nix_error")
error_log.addStderr(f"{attr} failed to evaluate:\n{error}")
return util.FAILURE
class NixBuildCommand(buildstep.ShellMixin, steps.BuildStep):
"""Builds a nix derivation."""
def __init__(self, **kwargs: Any) -> None:
kwargs = self.setupShellMixin(kwargs)
super().__init__(**kwargs)
@defer.inlineCallbacks
def run(self) -> Generator[Any, object, Any]:
# run `nix build`
cmd: remotecommand.RemoteCommand = yield self.makeRemoteShellCommand()
yield self.runCommand(cmd)
res = cmd.results()
if res == util.FAILURE:
retries = RETRY_COUNTER.retry_build(self.getProperty("build_uuid"))
if retries > 0:
return util.RETRY
return res
class UpdateBuildOutput(steps.BuildStep):
"""Updates store paths in a public www directory.
This is useful to prefetch updates without having to evaluate
on the target machine.
"""
def __init__(self, path: Path, **kwargs: Any) -> None:
super().__init__(**kwargs)
self.path = path
def run(self) -> Generator[Any, object, Any]:
props = self.build.getProperties()
if props.getProperty("branch") != props.getProperty(
"github.repository.default_branch",
):
return util.SKIPPED
attr = Path(props.getProperty("attr")).name
out_path = props.getProperty("out_path")
# XXX don't hardcode this
self.path.mkdir(parents=True, exist_ok=True)
(self.path / attr).write_text(out_path)
return util.SUCCESS
# The builtin retry mechanism doesn't seem to work for github,
# since github is sometimes not delivering the pull request ref fast enough.
class GitWithRetry(steps.Git):
@defer.inlineCallbacks
def run_vc(
self,
branch: str,
revision: str,
patch: str,
) -> Generator[Any, object, Any]:
retry_counter = 0
while True:
try:
res = yield super().run_vc(branch, revision, patch)
except Exception as e: # noqa: BLE001
retry_counter += 1
if retry_counter == 3:
msg = "Failed to clone"
raise BuildbotNixError(msg) from e
log: Log = yield self.addLog("log")
yield log.addStderr(f"Retrying git clone (error: {e})\n")
yield asyncSleep(2 << retry_counter) # 2, 4, 8
else:
return res
def nix_eval_config(
project: GerritProject,
gerrit_private_key: str,
worker_names: list[str],
supported_systems: list[str],
eval_lock: util.MasterLock,
worker_count: int,
max_memory_size: int,
) -> util.BuilderConfig:
"""Uses nix-eval-jobs to evaluate hydraJobs from flake.nix in parallel.
For each evaluated attribute a new build pipeline is started.
"""
factory = util.BuildFactory()
# check out the source
factory.addStep(
steps.Gerrit(
repourl="ssh://buildbot@gerrit.lix.systems:2022/lix",
mode="full",
retry=[60, 60],
timeout=3600,
sshPrivateKey=gerrit_private_key
),
)
drv_gcroots_dir = util.Interpolate(
"/nix/var/nix/gcroots/per-user/buildbot-worker/%(prop:project)s/drvs/",
)
factory.addStep(
NixEvalCommand(
env={},
name="evaluate flake",
supported_systems=supported_systems,
command=[
"nix-eval-jobs",
"--workers",
str(worker_count),
"--max-memory-size",
str(max_memory_size),
"--option",
"accept-flake-config",
"true",
"--gc-roots-dir",
drv_gcroots_dir,
"--force-recurse",
"--check-cache-status",
"--flake",
".#checks",
],
haltOnFailure=True,
locks=[eval_lock.access("exclusive")],
),
)
factory.addStep(
steps.ShellCommand(
name="Cleanup drv paths",
command=[
"rm",
"-rf",
drv_gcroots_dir,
],
),
)
return util.BuilderConfig(
name=f"{project.name}/nix-eval",
workernames=worker_names,
project=project.name,
factory=factory,
properties=dict(status_name="nix-eval"),
)
@dataclass
class CachixConfig:
name: str
signing_key_secret_name: str | None = None
auth_token_secret_name: str | None = None
def cachix_env(self) -> dict[str, str]:
env = {}
if self.signing_key_secret_name is not None:
env["CACHIX_SIGNING_KEY"] = util.Secret(self.signing_key_secret_name)
if self.auth_token_secret_name is not None:
env["CACHIX_AUTH_TOKEN"] = util.Secret(self.auth_token_secret_name)
return env
def nix_build_config(
project: GerritProject,
worker_names: list[str],
cachix: CachixConfig | None = None,
outputs_path: Path | None = None,
) -> util.BuilderConfig:
"""Builds one nix flake attribute."""
factory = util.BuildFactory()
factory.addStep(
NixBuildCommand(
env={},
name="Build flake attr",
command=[
"nix",
"build",
"-L",
"--option",
"keep-going",
"true",
"--option",
# stop stuck builds after 20 minutes
"--max-silent-time",
str(60 * 20),
"--accept-flake-config",
"--out-link",
util.Interpolate("result-%(prop:attr)s"),
util.Interpolate("%(prop:drv_path)s^*"),
],
# 3 hours, defaults to 20 minutes
# We increase this over the default since the build output might end up in a different `nix build`.
timeout=60 * 60 * 3,
haltOnFailure=True,
),
)
if cachix:
factory.addStep(
steps.ShellCommand(
name="Upload cachix",
env=cachix.cachix_env(),
command=[
"cachix",
"push",
cachix.name,
util.Interpolate("result-%(prop:attr)s"),
],
),
)
factory.addStep(
steps.ShellCommand(
name="Register gcroot",
command=[
"nix-store",
"--add-root",
# FIXME: cleanup old build attributes
util.Interpolate(
"/nix/var/nix/gcroots/per-user/buildbot-worker/%(prop:project)s/%(prop:attr)s",
),
"-r",
util.Property("out_path"),
],
doStepIf=lambda s: s.getProperty("branch")
== s.getProperty("github.repository.default_branch"),
),
)
factory.addStep(
steps.ShellCommand(
name="Delete temporary gcroots",
command=["rm", "-f", util.Interpolate("result-%(prop:attr)s")],
),
)
if outputs_path is not None:
factory.addStep(
UpdateBuildOutput(
name="Update build output",
path=outputs_path,
),
)
return util.BuilderConfig(
name=f"{project.name}/nix-build",
project=project.name,
workernames=worker_names,
collapseRequests=False,
env={},
factory=factory,
)
def nix_skipped_build_config(
project: GerritProject,
worker_names: list[str],
) -> util.BuilderConfig:
"""Dummy builder that is triggered when a build is skipped."""
factory = util.BuildFactory()
factory.addStep(
EvalErrorStep(
name="Nix evaluation",
doStepIf=lambda s: s.getProperty("error"),
hideStepIf=lambda _, s: not s.getProperty("error"),
),
)
# This is just a dummy step showing the cached build
factory.addStep(
steps.BuildStep(
name="Nix build (cached)",
doStepIf=lambda _: False,
hideStepIf=lambda _, s: s.getProperty("error"),
),
)
return util.BuilderConfig(
name=f"{project.name}/nix-skipped-build",
project=project.name,
workernames=worker_names,
collapseRequests=False,
env={},
factory=factory,
)
def read_secret_file(secret_name: str) -> str:
directory = os.environ.get("CREDENTIALS_DIRECTORY")
if directory is None:
print("directory not set", file=sys.stderr)
sys.exit(1)
return Path(directory).joinpath(secret_name).read_text().rstrip()
def config_for_project(
config: dict[str, Any],
project: GerritProject,
worker_names: list[str],
nix_supported_systems: list[str],
nix_eval_worker_count: int,
nix_eval_max_memory_size: int,
eval_lock: util.MasterLock,
cachix: CachixConfig | None = None,
outputs_path: Path | None = None,
) -> Project:
config["projects"].append(Project(project.name))
config["schedulers"].extend(
[
# build everything pertaining to a project
# TODO(raito): will this catch also post-merge? we don't really care about that… do we?
schedulers.SingleBranchScheduler(
name=f"{project.name}-changes",
change_filter=util.ChangeFilter(
project=project.name,
),
builderNames=[f"{project.name}/nix-eval"],
),
# this is triggered from `nix-eval`
schedulers.Triggerable(
name=f"{project.name}-nix-build",
builderNames=[f"{project.name}/nix-build"],
),
# this is triggered from `nix-eval` when the build is skipped
schedulers.Triggerable(
name=f"{project.name}-nix-skipped-build",
builderNames=[f"{project.name}/nix-skipped-build"],
),
# allow to manually trigger a nix-build
schedulers.ForceScheduler(
name=f"{project.name}-force",
builderNames=[f"{project.name}/nix-eval"],
properties=[
util.StringParameter(
name="project",
label="Name of the Gerrit repository.",
default=project.name,
),
],
),
],
)
gerrit_private_key = None
with open('/var/lib/buildbot/master/id_gerrit', 'r') as f:
gerrit_private_key = f.read()
if gerrit_private_key is None:
raise RuntimeError('No gerrit private key to fetch the repositories')
config["builders"].extend(
[
# Since all workers run on the same machine, we only assign one of them to do the evaluation.
# This should prevent exessive memory usage.
nix_eval_config(
project,
gerrit_private_key,
worker_names,
supported_systems=nix_supported_systems,
worker_count=nix_eval_worker_count,
max_memory_size=nix_eval_max_memory_size,
eval_lock=eval_lock,
),
nix_build_config(
project,
worker_names,
cachix=cachix,
outputs_path=outputs_path,
),
nix_skipped_build_config(project, [SKIPPED_BUILDER_NAME]),
],
)
class PeriodicWithStartup(schedulers.Periodic):
def __init__(self, *args: Any, run_on_startup: bool = False, **kwargs: Any) -> None:
super().__init__(*args, **kwargs)
self.run_on_startup = run_on_startup
@defer.inlineCallbacks
def activate(self) -> Generator[Any, object, Any]:
if self.run_on_startup:
yield self.setState("last_build", None)
yield super().activate()
def gerritReviewCB(builderName, build, result, master, arg):
if result == util.RETRY:
return dict()
message = "Buildbot finished compiling your patchset\n"
message += "on configuration: %s\n" % builderName
message += "The result is: %s\n" % util.Results[result].upper()
if arg:
message += "\nFor more details visit:\n"
message += build['url'] + "\n"
if result == util.SUCCESS:
verified = 1
else:
verified = -1
return dict(message=message, labels={'Verified': verified})
def gerritStartCB(builderName, build, arg):
message = "Buildbot started compiling your patchset\n"
message += "on configuration: %s\n" % builderName
message += "See your build here: %s" % build['url']
return dict(message=message)
def gerritSummaryCB(buildInfoList, results, status, arg):
success = False
failure = False
msgs = []
for buildInfo in buildInfoList:
msg = "Builder %(name)s %(resultText)s (%(text)s)" % buildInfo
link = buildInfo.get('url', None)
if link:
msg += " - " + link
else:
msg += "."
msgs.append(msg)
if buildInfo['result'] == util.SUCCESS:
success = True
else:
failure = True
if success and not failure:
verified = 1
else:
verified = -1
return dict(message='\n\n'.join(msgs),
labels={
'Verified': verified
})
class GerritNixConfigurator(ConfiguratorBase):
"""Janitor is a configurator which create a Janitor Builder with all needed Janitor steps"""
def __init__(
self,
# Shape of this file: [ { "name": "<worker-name>", "pass": "<worker-password>", "cores": "<cpu-cores>" } ]
gerrit_server: str,
gerrit_user: str,
gerrit_port: int,
gerrit_sshkey_path: str,
url: str,
nix_supported_systems: list[str],
nix_eval_worker_count: int | None,
nix_eval_max_memory_size: int,
nix_workers_secret_name: str = "buildbot-nix-workers", # noqa: S107
cachix: CachixConfig | None = None,
outputs_path: str | None = None,
) -> None:
super().__init__()
self.gerrit_server = gerrit_server
self.gerrit_user = gerrit_user
self.gerrit_port = gerrit_port
self.nix_workers_secret_name = nix_workers_secret_name
self.nix_eval_max_memory_size = nix_eval_max_memory_size
self.nix_eval_worker_count = nix_eval_worker_count
self.nix_supported_systems = nix_supported_systems
self.gerrit_change_source = GerritChangeSource(gerrit_server, gerrit_user, gerritport=gerrit_port, identity_file=gerrit_sshkey_path)
self.url = url
self.cachix = cachix
if outputs_path is None:
self.outputs_path = None
else:
self.outputs_path = Path(outputs_path)
def configure(self, config: dict[str, Any]) -> None:
worker_config = json.loads(read_secret_file(self.nix_workers_secret_name))
worker_names = []
config.setdefault("projects", [])
config.setdefault("secretsProviders", [])
config.setdefault("www", {})
for item in worker_config:
cores = item.get("cores", 0)
for i in range(cores):
worker_name = f"{item['name']}-{i:03}"
config["workers"].append(worker.Worker(worker_name, item["pass"]))
worker_names.append(worker_name)
eval_lock = util.MasterLock("nix-eval")
# Configure the Lix project.
config_for_project(
config,
GerritProject(name="lix"),
worker_names,
self.nix_supported_systems,
self.nix_eval_worker_count or multiprocessing.cpu_count(),
self.nix_eval_max_memory_size,
eval_lock,
self.cachix,
self.outputs_path,
)
config["change_source"] = self.gerrit_change_source
config["workers"].append(worker.LocalWorker(SKIPPED_BUILDER_NAME))
config["services"].append(
reporters.GerritStatusPush(self.gerrit_server, self.gerrit_user,
port=2022,
identity_file='/var/lib/buildbot/master/id_gerrit',
reviewCB=gerritReviewCB,
reviewArg=self.url,
startCB=gerritStartCB,
startArg=self.url,
summaryCB=gerritSummaryCB,
summaryArg=self.url)
)
systemd_secrets = secrets.SecretInAFile(
dirname=os.environ["CREDENTIALS_DIRECTORY"],
)
config["secretsProviders"].append(systemd_secrets)
config["www"].setdefault("plugins", {})
config["www"]["plugins"].update(dict(base_react={}))
if "auth" not in config["www"]:
config["www"]["auth"] = LixSystemsOAuth2('buildbot', read_secret_file('buildbot-oauth2-secret'), autologin=True)