Files
Alexei FedotovandClaude Opus 5.5 f26e51a98f darwin fast test: enable the job in the PR workflow again
Reverts https://github.com/ClickHouse/ClickHouse/pull/125587. The macOS fleet grew from 10 to 21 runners; 50 parallel copies started without waiting for a runner and left no orphaned processes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-23 22:54:39 +02:00

348 lines
14 KiB
Python

from praktika import Workflow
from ci.defs.defs import (
BASE_BRANCH,
DOCKERS,
SECRETS,
ArtifactConfigs,
JobNames,
)
from ci.defs.job_configs import JobConfigs
from ci.jobs.scripts.workflow_hooks.filter_job import (
PR_SINGLE_ARCH_SKIPPED_BUILDS,
PR_SINGLE_ARCH_SKIPPED_JOBS,
should_skip_job,
)
from ci.jobs.scripts.workflow_hooks.trusted import can_be_tested
# PR sanitizer jobs repeat tests related to the change to find intermittent failures.
# Debug and plain binary jobs run the full suite. Master keeps the full suite in
# every configuration; coverage-based targeting is specific to pull requests.
FUNCTIONAL_TESTS_JOBS = JobConfigs.functional_tests_pr_jobs
ALL_FUNCTIONAL_TESTS = [job.name for job in FUNCTIONAL_TESTS_JOBS]
CORE_BLOCKING_JOB_NAMES = [
job.name
for job in FUNCTIONAL_TESTS_JOBS
if any(
substr in job.name
for substr in (
"_debug, parallel",
"_binary, parallel",
"_binary, sequential",
"_asan_ubsan, db disk, distributed plan, targeted",
)
)
] + [
job.name
for job in JobConfigs.integration_test_targeted_pr_jobs
if "_asan_ubsan, db disk, targeted" in job.name
] + [
job.name
for job in JobConfigs.unittest_jobs
]
STYLE_AND_FAST_TESTS = [
JobNames.STYLE_CHECK,
JobNames.FAST_TEST,
*[j.name for j in JobConfigs.tidy_build_arm_jobs],
]
CODE_REVIEW_BLOCKING_JOBS = [
JobNames.STYLE_CHECK,
JobNames.FAST_TEST,
]
REGULAR_BUILD_NAMES = [job.name for job in JobConfigs.build_jobs]
PLAIN_FUNCTIONAL_TEST_JOB = [
j for j in JobConfigs.functional_tests_jobs if "amd_debug, parallel" in j.name
][0]
# Pull requests run the integration tests only in targeted jobs (the changed tests, the
# tests covering the changed lines and the tests that failed in the PR before, each run
# once), except for the full `arm_binary` run below and, with the `ci-coverage` or
# `ci-llvm-coverage` label, the full LLVM coverage run and the tests excluded from it.
INTEGRATION_TARGETED_JOBS = JobConfigs.integration_test_targeted_pr_jobs
# The LLVM coverage jobs run in pull requests only with the `ci-coverage` or `ci-llvm-coverage`
# label (see `should_skip_job`); by default these `arm_binary` jobs run the same configurations instead.
# Master runs the coverage jobs on every commit.
FUNCTIONAL_COVERAGE_REPLACEMENT_JOBS = (
JobConfigs.functional_tests_arm_binary_coverage_replacement_pr_jobs
)
INTEGRATION_COVERAGE_REPLACEMENT_JOBS = (
JobConfigs.integration_test_arm_binary_coverage_replacement_pr_jobs
)
PLAIN_INTEGRATION_TEST_JOB = [
j for j in INTEGRATION_TARGETED_JOBS if "(amd_tsan, targeted)" in j.name
][0]
workflow = Workflow.Config(
name="PR",
event=Workflow.Event.PULL_REQUEST,
base_branches=[BASE_BRANCH],
engine=Workflow.Engine.GH_ACTIONS,
jobs=[
JobConfigs.style_check,
JobConfigs.code_review.set_run_after(CODE_REVIEW_BLOCKING_JOBS),
JobConfigs.docs_job_mintlify,
JobConfigs.fast_test,
*JobConfigs.darwin_fast_test_jobs,
*JobConfigs.tidy_build_arm_jobs,
*[job.set_run_after(STYLE_AND_FAST_TESTS) for job in JobConfigs.build_jobs],
*[
job.set_run_after(STYLE_AND_FAST_TESTS)
for job in JobConfigs.extra_validation_build_jobs
],
*[
job.set_run_after(REGULAR_BUILD_NAMES)
for job in JobConfigs.release_build_jobs_with_examples
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.special_build_jobs
],
# Gated like the regular builds rather than like the special ones: it is the only job
# that compiles the standalone parser at all, and it needs no build artifact, so there
# is nothing to gain by deferring it behind the functional tests.
*[
job.set_run_after(STYLE_AND_FAST_TESTS)
for job in JobConfigs.wasm_parser_build_jobs
],
*[
job.set_run_after(STYLE_AND_FAST_TESTS)
for job in JobConfigs.build_llvm_coverage_job
],
# TODO: stabilize new jobs and remove set_allow_failure
JobConfigs.lightweight_functional_tests_job,
*[j.set_allow_failure() for j in JobConfigs.stateless_tests_targeted_pr_jobs],
*[
job.set_run_after(
CORE_BLOCKING_JOB_NAMES if job.name not in CORE_BLOCKING_JOB_NAMES else []
)
for job in INTEGRATION_TARGETED_JOBS
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.integration_test_contrib_tsan_pr_jobs
],
JobConfigs.ast_fuzzer_targeted_pr_jobs[0].set_allow_failure(),
JobConfigs.ast_fuzzer_targeted_pr_jobs[1].set_allow_failure(),
*JobConfigs.stateless_tests_flaky_pr_jobs,
# The merge queue's non-sanitizer flaky check also runs here, so a test
# that is only too slow (or only flaky) without a sanitizer is reported
# in the PR rather than first bouncing it from the merge queue. Same job
# config as in `ci/workflows/merge_queue.py`.
*JobConfigs.stateless_tests_flaky_mq_jobs,
*JobConfigs.integration_test_asan_flaky_pr_jobs,
# Per-arch Bugfix Validation Checks (functional + integration tests on
# both amd64 and aarch64). Each per-arch variant has
# `allow_failure=True` so an individual FAIL doesn't block PR merge -
# the aggregate decision (validate iff at least one arch passed) lives
# in the `new_tests_check.py` workflow post-hook below.
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.bugfix_validation_ft_pr_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.bugfix_validation_it_jobs
],
# Unit-test (gtest) bugfix validation: a single AMD-only job (allow_failure)
# that builds a merge-base "before" binary and runs the touched suite against it.
# It is not part of the per-arch FT/IT aggregation; instead new_tests_check.py
# blocks the unit case iff this job reported a definitive FAIL (failed to
# reproduce) — a reproduction or an inconclusive ERROR does not block.
# Like the sibling FT/IT jobs, it is deferred behind the core blocking jobs.
JobConfigs.bugfix_validation_ut_job.set_run_after(CORE_BLOCKING_JOB_NAMES),
*[
j.set_run_after(
CORE_BLOCKING_JOB_NAMES if j.name not in CORE_BLOCKING_JOB_NAMES else []
)
for j in FUNCTIONAL_TESTS_JOBS
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.functional_tests_jobs_azure
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in FUNCTIONAL_COVERAGE_REPLACEMENT_JOBS
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in INTEGRATION_COVERAGE_REPLACEMENT_JOBS
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.functional_test_llvm_coverage_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.functional_test_excluded_from_llvm_job
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.integration_test_llvm_coverage_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.integration_test_excluded_from_llvm_job
],
*JobConfigs.unittest_jobs,
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.unittest_llvm_coverage_job
],
JobConfigs.docker_server.set_run_after(CORE_BLOCKING_JOB_NAMES),
JobConfigs.docker_keeper.set_run_after(CORE_BLOCKING_JOB_NAMES),
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.install_check_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.compatibility_test_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.stress_test_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.upgrade_test_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.ast_fuzzer_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.buzz_fuzzer_jobs
],
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.performance_comparison_with_master_head_jobs
],
JobConfigs.parser_memory_check_job,
JobConfigs.storage_memory_check_job,
# ClickBench runs on PRs only when files in its digest change
# (see `clickbench_jobs.digest_config`), so the cost is bounded.
*[
job.set_run_after(CORE_BLOCKING_JOB_NAMES)
for job in JobConfigs.clickbench_jobs
],
JobConfigs.llvm_coverage_job,
# Waits for the integration jobs of this workflow, which upload the compliance results.
# The full `arm_binary` run (or, with `ci-coverage`, the LLVM coverage run) is the one that
# always runs the compliance suite.
JobConfigs.promql_compliance_job.set_run_after(
INTEGRATION_TARGETED_JOBS
+ INTEGRATION_COVERAGE_REPLACEMENT_JOBS
+ JobConfigs.integration_test_llvm_coverage_jobs
+ JobConfigs.integration_test_excluded_from_llvm_job,
reset=True,
),
# TODO: stabilize and remove set_allow_failure
JobConfigs.build_profile_diff_job.set_allow_failure(),
JobConfigs.sqllogic_test_master_job.set_run_after(CORE_BLOCKING_JOB_NAMES),
JobConfigs.sqlstorm_test_job.set_run_after(CORE_BLOCKING_JOB_NAMES),
JobConfigs.docs_examples_job.set_run_after(CORE_BLOCKING_JOB_NAMES),
# Keeper stress (PR): 3 no-fault scenarios (prod-mix, read-multi, write-multi),
# default backend only, 15 min each. Runs when src/Coordination or stress test files change.
JobConfigs.keeper_stress_job.set_name("Keeper Stress Tests (PR)").set_timeout(
3 * 3600
),
*JobConfigs.toolchain_build_jobs,
],
artifacts=[
*ArtifactConfigs.unittests_binaries,
*ArtifactConfigs.clickhouse_binaries,
*ArtifactConfigs.clickhouse_darwin_plain_binaries,
*ArtifactConfigs.clickhouse_debians,
*ArtifactConfigs.clickhouse_rpms,
*ArtifactConfigs.clickhouse_tgzs,
ArtifactConfigs.clickhouse_wasm,
ArtifactConfigs.wasm_parser,
ArtifactConfigs.fuzzers,
ArtifactConfigs.fuzzers_corpus,
ArtifactConfigs.clickhouse_examples,
*ArtifactConfigs.llvm_profdata_file,
ArtifactConfigs.llvm_coverage_info_file,
ArtifactConfigs.toolchain_pgo_bolt_amd,
ArtifactConfigs.toolchain_pgo_bolt_arm,
],
dockers=DOCKERS,
enable_dockers_manifest_merge=True,
secrets=SECRETS,
enable_job_filtering_by_changes=True,
enable_cache=True,
enable_report=True,
enable_cidb=True,
enable_merge_ready_status=True,
enable_gh_summary_comment=True,
enable_commit_status_on_failure=False,
enable_open_issues_check=True,
enable_slack_feed=True,
pre_hooks=[
can_be_tested,
"python3 ./ci/jobs/scripts/workflow_hooks/ci_links.py",
"python3 ./ci/jobs/scripts/workflow_hooks/store_data.py",
"python3 ./ci/jobs/scripts/workflow_hooks/pr_labels_and_category.py",
"python3 ./ci/jobs/scripts/workflow_hooks/version_log.py",
"python3 ./ci/jobs/scripts/workflow_hooks/team_notifications.py",
],
workflow_filter_hooks=[should_skip_job],
post_hooks=[
"python3 ./ci/jobs/scripts/workflow_hooks/pr_body_check.py",
"python3 ./ci/jobs/scripts/workflow_hooks/feature_docs.py",
"python3 ./ci/jobs/scripts/workflow_hooks/new_tests_check.py",
"python3 ./ci/jobs/scripts/workflow_hooks/can_be_merged.py",
"python3 ./ci/jobs/scripts/workflow_hooks/check_report_messages.py",
],
job_aliases={
# plain integration test job, no dist plan; runs `--test` as a regular job locally
"integration": PLAIN_INTEGRATION_TEST_JOB.name,
"fast": "Fast test",
"functional": PLAIN_FUNCTIONAL_TEST_JOB.name,
"build_debug": "Build (amd_debug)",
"build": "Build (amd_binary)",
},
runs_on_label_prefix="pr-",
ai_orchestrator=Workflow.OrchestratorAI.Config(
enabled=False,
provider="bedrock",
model="global.anthropic.claude-sonnet-5",
),
)
# `should_skip_job` skips these builds together with the stress tests that use them, so
# no other job may require their binaries: it would find no artifact to download.
_SINGLE_ARCH_SKIPPED_ARTIFACTS = {
artifact
for job in workflow.jobs
if job.name in PR_SINGLE_ARCH_SKIPPED_BUILDS
for artifact in job.provides
}
assert len(_SINGLE_ARCH_SKIPPED_ARTIFACTS) == len(PR_SINGLE_ARCH_SKIPPED_BUILDS), (
"every build in PR_SINGLE_ARCH_SKIPPED_BUILDS must be in the PR workflow"
)
_SINGLE_ARCH_SKIPPED_ARTIFACT_USERS = [
job.name
for job in workflow.jobs
if job.name not in PR_SINGLE_ARCH_SKIPPED_JOBS
and _SINGLE_ARCH_SKIPPED_ARTIFACTS & set(job.requires or [])
]
assert not _SINGLE_ARCH_SKIPPED_ARTIFACT_USERS, (
f"jobs {_SINGLE_ARCH_SKIPPED_ARTIFACT_USERS} require binaries of "
f"{PR_SINGLE_ARCH_SKIPPED_BUILDS}, which pull requests skip: remove the build "
"from PR_SINGLE_ARCH_SKIPPED_BUILDS in ci/jobs/scripts/workflow_hooks/filter_job.py"
)
WORKFLOWS = [
workflow,
]