mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-10 09:32:06 +00:00
2233 lines
80 KiB
Python
2233 lines
80 KiB
Python
import argparse
|
|
import concurrent.futures
|
|
import json
|
|
import logging
|
|
import os
|
|
import random
|
|
import re
|
|
import subprocess
|
|
import sys
|
|
import time
|
|
from copy import deepcopy
|
|
from dataclasses import asdict, dataclass
|
|
from enum import Enum
|
|
from pathlib import Path
|
|
from typing import Any, Dict, List, Optional, Sequence, Set, Tuple, Union
|
|
|
|
import docker_images_helper
|
|
import upload_result_helper
|
|
from build_check import get_release_or_pr
|
|
from ci_config import CI_CONFIG, Build, CILabels, CIStages, JobNames
|
|
from ci_utils import GHActions, is_hex, normalize_string
|
|
from clickhouse_helper import (
|
|
CiLogsCredentials,
|
|
ClickHouseHelper,
|
|
InsertException,
|
|
get_instance_id,
|
|
get_instance_type,
|
|
prepare_tests_results_for_clickhouse,
|
|
)
|
|
from commit_status_helper import (
|
|
CommitStatusData,
|
|
RerunHelper,
|
|
format_description,
|
|
get_commit,
|
|
post_commit_status,
|
|
set_status_comment,
|
|
)
|
|
from digest_helper import DockerDigester, JobDigester
|
|
from env_helper import (
|
|
CI,
|
|
GITHUB_JOB_API_URL,
|
|
GITHUB_RUN_URL,
|
|
REPO_COPY,
|
|
REPORT_PATH,
|
|
S3_BUILDS_BUCKET,
|
|
TEMP_PATH,
|
|
)
|
|
from get_robot_token import get_best_robot_token
|
|
from git_helper import GIT_PREFIX, Git
|
|
from git_helper import Runner as GitRunner
|
|
from github_helper import GitHub
|
|
from pr_info import PRInfo
|
|
from report import ERROR, SUCCESS, BuildResult, JobReport
|
|
from s3_helper import S3Helper
|
|
from version_helper import get_version_from_repo
|
|
|
|
# pylint: disable=too-many-lines
|
|
|
|
|
|
@dataclass
|
|
class PendingState:
|
|
updated_at: float
|
|
run_url: str
|
|
|
|
|
|
class CiCache:
|
|
"""
|
|
CI cache is a bunch of records. Record is a file stored under special location on s3.
|
|
The file name has following format
|
|
|
|
<RECORD_TYPE>_[<ATTRIBUTES>]--<JOB_NAME>_<JOB_DIGEST>_<BATCH>_<NUM_BATCHES>.ci
|
|
|
|
RECORD_TYPE:
|
|
SUCCESSFUL - for successfuly finished jobs
|
|
PENDING - for pending jobs
|
|
|
|
ATTRIBUTES:
|
|
release - for jobs being executed on the release branch including master branch (not a PR branch)
|
|
"""
|
|
|
|
_S3_CACHE_PREFIX = "CI_cache_v1"
|
|
_CACHE_BUILD_REPORT_PREFIX = "build_report"
|
|
_RECORD_FILE_EXTENSION = ".ci"
|
|
_LOCAL_CACHE_PATH = Path(TEMP_PATH) / "ci_cache"
|
|
_ATTRIBUTE_RELEASE = "release"
|
|
# divider symbol 1
|
|
_DIV1 = "--"
|
|
# divider symbol 2
|
|
_DIV2 = "_"
|
|
assert _DIV1 != _DIV2
|
|
|
|
class RecordType(Enum):
|
|
SUCCESSFUL = "successful"
|
|
PENDING = "pending"
|
|
FAILED = "failed"
|
|
|
|
@dataclass
|
|
class Record:
|
|
record_type: "CiCache.RecordType"
|
|
job_name: str
|
|
job_digest: str
|
|
batch: int
|
|
num_batches: int
|
|
release_branch: bool
|
|
file: str = ""
|
|
|
|
def to_str_key(self):
|
|
"""other fields must not be included in the hash str"""
|
|
return "_".join(
|
|
[self.job_name, self.job_digest, str(self.batch), str(self.num_batches)]
|
|
)
|
|
|
|
class JobType(Enum):
|
|
DOCS = "DOCS"
|
|
SRCS = "SRCS"
|
|
|
|
@classmethod
|
|
def is_docs_job(cls, job_name: str) -> bool:
|
|
return job_name == JobNames.DOCS_CHECK
|
|
|
|
@classmethod
|
|
def is_srcs_job(cls, job_name: str) -> bool:
|
|
return not cls.is_docs_job(job_name)
|
|
|
|
@classmethod
|
|
def get_type_by_name(cls, job_name: str) -> "CiCache.JobType":
|
|
res = cls.SRCS
|
|
if cls.is_docs_job(job_name):
|
|
res = cls.DOCS
|
|
elif cls.is_srcs_job(job_name):
|
|
res = cls.SRCS
|
|
else:
|
|
assert False
|
|
return res
|
|
|
|
def __init__(
|
|
self,
|
|
s3: S3Helper,
|
|
job_digests: Dict[str, str],
|
|
):
|
|
self.s3 = s3
|
|
self.job_digests = job_digests
|
|
self.cache_s3_paths = {
|
|
job_type: f"{self._S3_CACHE_PREFIX}/{job_type.value}-{self._get_digest_for_job_type(self.job_digests, job_type)}/"
|
|
for job_type in self.JobType
|
|
}
|
|
self.s3_record_prefixes = {
|
|
record_type: record_type.value for record_type in self.RecordType
|
|
}
|
|
self.records: Dict["CiCache.RecordType", Dict[str, "CiCache.Record"]] = {
|
|
record_type: {} for record_type in self.RecordType
|
|
}
|
|
|
|
self.cache_updated = False
|
|
self.cache_data_fetched = True
|
|
if not self._LOCAL_CACHE_PATH.exists():
|
|
self._LOCAL_CACHE_PATH.mkdir(parents=True, exist_ok=True)
|
|
|
|
def _get_digest_for_job_type(
|
|
self, job_digests: Dict[str, str], job_type: JobType
|
|
) -> str:
|
|
if job_type == self.JobType.DOCS:
|
|
res = job_digests[JobNames.DOCS_CHECK]
|
|
elif job_type == self.JobType.SRCS:
|
|
# any build type job has the same digest - pick up Build.PACKAGE_RELEASE or Build.PACKAGE_ASAN as a failover
|
|
# Build.PACKAGE_RELEASE may not exist in the list if we have reduced CI pipeline
|
|
if Build.PACKAGE_RELEASE in job_digests:
|
|
res = job_digests[Build.PACKAGE_RELEASE]
|
|
elif Build.PACKAGE_ASAN in job_digests:
|
|
# failover, if failover does not work - fix it!
|
|
res = job_digests[Build.PACKAGE_ASAN]
|
|
else:
|
|
assert False, "BUG, no build job in digest' list"
|
|
else:
|
|
assert False, "BUG, New JobType? - please update func"
|
|
return res
|
|
|
|
def _get_record_file_name(
|
|
self,
|
|
record_type: RecordType,
|
|
job_name: str,
|
|
batch: int,
|
|
num_batches: int,
|
|
release_branch: bool,
|
|
) -> str:
|
|
prefix = self.s3_record_prefixes[record_type]
|
|
prefix_extended = (
|
|
self._DIV2.join([prefix, self._ATTRIBUTE_RELEASE])
|
|
if release_branch
|
|
else prefix
|
|
)
|
|
assert self._DIV1 not in job_name, f"Invalid job name {job_name}"
|
|
job_name = self._DIV2.join(
|
|
[job_name, self.job_digests[job_name], str(batch), str(num_batches)]
|
|
)
|
|
file_name = self._DIV1.join([prefix_extended, job_name])
|
|
file_name += self._RECORD_FILE_EXTENSION
|
|
return file_name
|
|
|
|
def _get_record_s3_path(self, job_name: str) -> str:
|
|
return self.cache_s3_paths[self.JobType.get_type_by_name(job_name)]
|
|
|
|
def _parse_record_file_name(
|
|
self, record_type: RecordType, file_name: str
|
|
) -> Optional["CiCache.Record"]:
|
|
# validate filename
|
|
if (
|
|
not file_name.endswith(self._RECORD_FILE_EXTENSION)
|
|
or not len(file_name.split(self._DIV1)) == 2
|
|
):
|
|
print("ERROR: wrong file name format")
|
|
return None
|
|
|
|
file_name = file_name.removesuffix(self._RECORD_FILE_EXTENSION)
|
|
release_branch = False
|
|
|
|
prefix_extended, job_suffix = file_name.split(self._DIV1)
|
|
record_type_and_attribute = prefix_extended.split(self._DIV2)
|
|
|
|
# validate filename prefix
|
|
failure = False
|
|
if not 0 < len(record_type_and_attribute) <= 2:
|
|
print("ERROR: wrong file name prefix")
|
|
failure = True
|
|
if (
|
|
len(record_type_and_attribute) > 1
|
|
and record_type_and_attribute[1] != self._ATTRIBUTE_RELEASE
|
|
):
|
|
print("ERROR: wrong record attribute")
|
|
failure = True
|
|
if record_type_and_attribute[0] != self.s3_record_prefixes[record_type]:
|
|
print("ERROR: wrong record type")
|
|
failure = True
|
|
if failure:
|
|
return None
|
|
|
|
if (
|
|
len(record_type_and_attribute) > 1
|
|
and record_type_and_attribute[1] == self._ATTRIBUTE_RELEASE
|
|
):
|
|
release_branch = True
|
|
|
|
job_properties = job_suffix.split(self._DIV2)
|
|
job_name, job_digest, batch, num_batches = (
|
|
self._DIV2.join(job_properties[:-3]),
|
|
job_properties[-3],
|
|
int(job_properties[-2]),
|
|
int(job_properties[-1]),
|
|
)
|
|
|
|
if not is_hex(job_digest):
|
|
print("ERROR: wrong record job digest")
|
|
return None
|
|
|
|
record = self.Record(
|
|
record_type,
|
|
job_name,
|
|
job_digest,
|
|
batch,
|
|
num_batches,
|
|
release_branch,
|
|
file="",
|
|
)
|
|
return record
|
|
|
|
def print_status(self):
|
|
for record_type in self.RecordType:
|
|
GHActions.print_in_group(
|
|
f"Cache records: [{record_type}]", list(self.records[record_type])
|
|
)
|
|
return self
|
|
|
|
def update(self):
|
|
"""
|
|
Pulls cache records from s3. Only records name w/o content.
|
|
"""
|
|
for record_type in self.RecordType:
|
|
prefix = self.s3_record_prefixes[record_type]
|
|
cache_list = self.records[record_type]
|
|
for job_type in self.JobType:
|
|
path = self.cache_s3_paths[job_type]
|
|
records = self.s3.list_prefix(f"{path}{prefix}", S3_BUILDS_BUCKET)
|
|
records = [record.split("/")[-1] for record in records]
|
|
for file in records:
|
|
record = self._parse_record_file_name(
|
|
record_type=record_type, file_name=file
|
|
)
|
|
if not record:
|
|
print(f"ERROR: failed to parse cache record [{file}]")
|
|
continue
|
|
if (
|
|
record.job_name not in self.job_digests
|
|
or self.job_digests[record.job_name] != record.job_digest
|
|
):
|
|
# skip records we are not interested in
|
|
continue
|
|
|
|
if record.to_str_key() not in cache_list:
|
|
cache_list[record.to_str_key()] = record
|
|
self.cache_data_fetched = False
|
|
elif (
|
|
not cache_list[record.to_str_key()].release_branch
|
|
and record.release_branch
|
|
):
|
|
# replace a non-release record with a release one
|
|
cache_list[record.to_str_key()] = record
|
|
self.cache_data_fetched = False
|
|
|
|
self.cache_updated = True
|
|
return self
|
|
|
|
def fetch_records_data(self):
|
|
"""
|
|
Pulls CommitStatusData for all cached jobs from s3
|
|
"""
|
|
if not self.cache_updated:
|
|
self.update()
|
|
|
|
if self.cache_data_fetched:
|
|
# there are no records without fetched data - no need to fetch
|
|
return self
|
|
|
|
# clean up
|
|
for file in self._LOCAL_CACHE_PATH.glob("*.ci"):
|
|
file.unlink()
|
|
|
|
# download all record files
|
|
for job_type in self.JobType:
|
|
path = self.cache_s3_paths[job_type]
|
|
for record_type in self.RecordType:
|
|
prefix = self.s3_record_prefixes[record_type]
|
|
_ = self.s3.download_files(
|
|
bucket=S3_BUILDS_BUCKET,
|
|
s3_path=f"{path}{prefix}",
|
|
file_suffix=self._RECORD_FILE_EXTENSION,
|
|
local_directory=self._LOCAL_CACHE_PATH,
|
|
)
|
|
|
|
# validate we have files for all records and save file names meanwhile
|
|
for record_type in self.RecordType:
|
|
record_list = self.records[record_type]
|
|
for _, record in record_list.items():
|
|
record_file_name = self._get_record_file_name(
|
|
record_type,
|
|
record.job_name,
|
|
record.batch,
|
|
record.num_batches,
|
|
record.release_branch,
|
|
)
|
|
assert (
|
|
self._LOCAL_CACHE_PATH / record_file_name
|
|
).is_file(), f"BUG. Record file must be present: {self._LOCAL_CACHE_PATH / record_file_name}"
|
|
record.file = record_file_name
|
|
|
|
self.cache_data_fetched = True
|
|
return self
|
|
|
|
def exist(
|
|
self,
|
|
record_type: "CiCache.RecordType",
|
|
job: str,
|
|
batch: int,
|
|
num_batches: int,
|
|
release_branch: bool,
|
|
) -> bool:
|
|
if not self.cache_updated:
|
|
self.update()
|
|
record_key = self.Record(
|
|
record_type,
|
|
job,
|
|
self.job_digests[job],
|
|
batch,
|
|
num_batches,
|
|
release_branch,
|
|
).to_str_key()
|
|
res = record_key in self.records[record_type]
|
|
if release_branch:
|
|
return res and self.records[record_type][record_key].release_branch
|
|
else:
|
|
return res
|
|
|
|
def push(
|
|
self,
|
|
record_type: "CiCache.RecordType",
|
|
job: str,
|
|
batches: Union[int, Sequence[int]],
|
|
num_batches: int,
|
|
status: Union[CommitStatusData, PendingState],
|
|
release_branch: bool = False,
|
|
) -> None:
|
|
"""
|
|
Pushes a cache record (CommitStatusData)
|
|
@release_branch adds "release" attribute to a record
|
|
"""
|
|
if isinstance(batches, int):
|
|
batches = [batches]
|
|
for batch in batches:
|
|
record_file = self._LOCAL_CACHE_PATH / self._get_record_file_name(
|
|
record_type, job, batch, num_batches, release_branch
|
|
)
|
|
record_s3_path = self._get_record_s3_path(job)
|
|
if record_type == self.RecordType.SUCCESSFUL:
|
|
assert isinstance(status, CommitStatusData)
|
|
status.dump_to_file(record_file)
|
|
elif record_type == self.RecordType.FAILED:
|
|
assert isinstance(status, CommitStatusData)
|
|
status.dump_to_file(record_file)
|
|
elif record_type == self.RecordType.PENDING:
|
|
assert isinstance(status, PendingState)
|
|
with open(record_file, "w", encoding="utf-8") as json_file:
|
|
json.dump(asdict(status), json_file)
|
|
else:
|
|
assert False
|
|
|
|
_ = self.s3.upload_file(
|
|
bucket=S3_BUILDS_BUCKET,
|
|
file_path=record_file,
|
|
s3_path=record_s3_path + record_file.name,
|
|
)
|
|
record = self.Record(
|
|
record_type,
|
|
job,
|
|
self.job_digests[job],
|
|
batch,
|
|
num_batches,
|
|
release_branch,
|
|
file=record_file.name,
|
|
)
|
|
if (
|
|
record.release_branch
|
|
or record.to_str_key() not in self.records[record_type]
|
|
):
|
|
self.records[record_type][record.to_str_key()] = record
|
|
|
|
def get(
|
|
self, record_type: "CiCache.RecordType", job: str, batch: int, num_batches: int
|
|
) -> Optional[Union[CommitStatusData, PendingState]]:
|
|
"""
|
|
Gets a cache record data for a job, or None if a cache miss
|
|
"""
|
|
|
|
if not self.cache_data_fetched:
|
|
self.fetch_records_data()
|
|
|
|
record_key = self.Record(
|
|
record_type,
|
|
job,
|
|
self.job_digests[job],
|
|
batch,
|
|
num_batches,
|
|
release_branch=False,
|
|
).to_str_key()
|
|
|
|
if record_key not in self.records[record_type]:
|
|
return None
|
|
|
|
record_file_name = self.records[record_type][record_key].file
|
|
|
|
res = CommitStatusData.load_from_file(
|
|
self._LOCAL_CACHE_PATH / record_file_name
|
|
) # type: CommitStatusData
|
|
|
|
return res
|
|
|
|
def delete(
|
|
self,
|
|
record_type: "CiCache.RecordType",
|
|
job: str,
|
|
batch: int,
|
|
num_batches: int,
|
|
release_branch: bool,
|
|
) -> None:
|
|
"""
|
|
deletes record from the cache
|
|
"""
|
|
raise NotImplementedError("Let's try make cache push-and-read-only")
|
|
# assert (
|
|
# record_type == self.RecordType.PENDING
|
|
# ), "FIXME: delete is supported for pending records only"
|
|
# record_file_name = self._get_record_file_name(
|
|
# self.RecordType.PENDING,
|
|
# job,
|
|
# batch,
|
|
# num_batches,
|
|
# release_branch=release_branch,
|
|
# )
|
|
# record_s3_path = self._get_record_s3_path(job)
|
|
# self.s3.delete_file_from_s3(S3_BUILDS_BUCKET, record_s3_path + record_file_name)
|
|
|
|
# record_key = self.Record(
|
|
# record_type,
|
|
# job,
|
|
# self.job_digests[job],
|
|
# batch,
|
|
# num_batches,
|
|
# release_branch=False,
|
|
# ).to_str_key()
|
|
|
|
# if record_key in self.records[record_type]:
|
|
# del self.records[record_type][record_key]
|
|
|
|
def is_successful(
|
|
self, job: str, batch: int, num_batches: int, release_branch: bool
|
|
) -> bool:
|
|
"""
|
|
checks if a given job have already been done successfuly
|
|
"""
|
|
return self.exist(
|
|
self.RecordType.SUCCESSFUL, job, batch, num_batches, release_branch
|
|
)
|
|
|
|
def is_failed(
|
|
self, job: str, batch: int, num_batches: int, release_branch: bool
|
|
) -> bool:
|
|
"""
|
|
checks if a given job have already been done with failure
|
|
"""
|
|
return self.exist(
|
|
self.RecordType.FAILED, job, batch, num_batches, release_branch
|
|
)
|
|
|
|
def is_pending(
|
|
self, job: str, batch: int, num_batches: int, release_branch: bool
|
|
) -> bool:
|
|
"""
|
|
check pending record in the cache for a given job
|
|
@release_branch - checks that "release" attribute is set for a record
|
|
"""
|
|
if self.is_successful(
|
|
job, batch, num_batches, release_branch
|
|
) or self.is_failed(job, batch, num_batches, release_branch):
|
|
return False
|
|
|
|
return self.exist(
|
|
self.RecordType.PENDING, job, batch, num_batches, release_branch
|
|
)
|
|
|
|
def push_successful(
|
|
self,
|
|
job: str,
|
|
batch: int,
|
|
num_batches: int,
|
|
job_status: CommitStatusData,
|
|
release_branch: bool = False,
|
|
) -> None:
|
|
"""
|
|
Pushes a cache record (CommitStatusData)
|
|
@release_branch adds "release" attribute to a record
|
|
"""
|
|
self.push(
|
|
self.RecordType.SUCCESSFUL,
|
|
job,
|
|
[batch],
|
|
num_batches,
|
|
job_status,
|
|
release_branch,
|
|
)
|
|
|
|
def push_failed(
|
|
self,
|
|
job: str,
|
|
batch: int,
|
|
num_batches: int,
|
|
job_status: CommitStatusData,
|
|
release_branch: bool = False,
|
|
) -> None:
|
|
"""
|
|
Pushes a cache record of type Failed (CommitStatusData)
|
|
@release_branch adds "release" attribute to a record
|
|
"""
|
|
self.push(
|
|
self.RecordType.FAILED,
|
|
job,
|
|
[batch],
|
|
num_batches,
|
|
job_status,
|
|
release_branch,
|
|
)
|
|
|
|
def push_pending(
|
|
self, job: str, batches: List[int], num_batches: int, release_branch: bool
|
|
) -> None:
|
|
"""
|
|
pushes pending record for a job to the cache
|
|
"""
|
|
pending_state = PendingState(time.time(), run_url=GITHUB_RUN_URL)
|
|
self.push(
|
|
self.RecordType.PENDING,
|
|
job,
|
|
batches,
|
|
num_batches,
|
|
pending_state,
|
|
release_branch,
|
|
)
|
|
|
|
def get_successful(
|
|
self, job: str, batch: int, num_batches: int
|
|
) -> Optional[CommitStatusData]:
|
|
"""
|
|
Gets a cache record (CommitStatusData) for a job, or None if a cache miss
|
|
"""
|
|
res = self.get(self.RecordType.SUCCESSFUL, job, batch, num_batches)
|
|
assert res is None or isinstance(res, CommitStatusData)
|
|
return res
|
|
|
|
def delete_pending(
|
|
self, job: str, batch: int, num_batches: int, release_branch: bool
|
|
) -> None:
|
|
"""
|
|
deletes pending record from the cache
|
|
"""
|
|
self.delete(self.RecordType.PENDING, job, batch, num_batches, release_branch)
|
|
|
|
def download_build_reports(self, file_prefix: str = "") -> List[str]:
|
|
"""
|
|
not ideal class for this method,
|
|
but let it be as we store build reports in CI cache directory on s3
|
|
and CiCache knows where exactly
|
|
|
|
@file_prefix allows to filter out reports by git head_ref
|
|
"""
|
|
report_path = Path(REPORT_PATH)
|
|
report_path.mkdir(exist_ok=True, parents=True)
|
|
path = (
|
|
self._get_record_s3_path(Build.PACKAGE_RELEASE)
|
|
+ self._CACHE_BUILD_REPORT_PREFIX
|
|
)
|
|
if file_prefix:
|
|
path += "_" + file_prefix
|
|
reports_files = self.s3.download_files(
|
|
bucket=S3_BUILDS_BUCKET,
|
|
s3_path=path,
|
|
file_suffix=".json",
|
|
local_directory=report_path,
|
|
)
|
|
return reports_files
|
|
|
|
def upload_build_report(self, build_result: BuildResult) -> str:
|
|
result_json_path = build_result.write_json(Path(TEMP_PATH))
|
|
s3_path = (
|
|
self._get_record_s3_path(Build.PACKAGE_RELEASE) + result_json_path.name
|
|
)
|
|
return self.s3.upload_file(
|
|
bucket=S3_BUILDS_BUCKET, file_path=result_json_path, s3_path=s3_path
|
|
)
|
|
|
|
def await_jobs(
|
|
self, jobs_with_params: Dict[str, Dict[str, Any]], is_release_branch: bool
|
|
) -> Dict[str, List[int]]:
|
|
"""
|
|
await pending jobs to be finished
|
|
@jobs_with_params - jobs to await. {JOB_NAME: {"batches": [BATCHES...], "num_batches": NUM_BATCHES}}
|
|
returns successfully finished jobs: {JOB_NAME: [BATCHES...]}
|
|
"""
|
|
if not jobs_with_params:
|
|
return {}
|
|
poll_interval_sec = 300
|
|
# TIMEOUT * MAX_ROUNDS_TO_WAIT must be less than 6h (GH job timeout) with a room for rest RunConfig work
|
|
TIMEOUT = 3000 # 50 min
|
|
MAX_ROUNDS_TO_WAIT = 6
|
|
MAX_JOB_NUM_TO_WAIT = 3
|
|
await_finished: Dict[str, List[int]] = {}
|
|
round_cnt = 0
|
|
while (
|
|
len(jobs_with_params) > MAX_JOB_NUM_TO_WAIT
|
|
and round_cnt < MAX_ROUNDS_TO_WAIT
|
|
):
|
|
round_cnt += 1
|
|
GHActions.print_in_group(
|
|
f"Wait pending jobs, round [{round_cnt}/{MAX_ROUNDS_TO_WAIT}]:",
|
|
list(jobs_with_params),
|
|
)
|
|
# this is initial approach to wait pending jobs:
|
|
# start waiting for the next TIMEOUT seconds if there are more than X(=4) jobs to wait
|
|
# wait TIMEOUT seconds in rounds. Y(=5) is the max number of rounds
|
|
expired_sec = 0
|
|
start_at = int(time.time())
|
|
while expired_sec < TIMEOUT and jobs_with_params:
|
|
time.sleep(poll_interval_sec)
|
|
self.update()
|
|
jobs_with_params_copy = deepcopy(jobs_with_params)
|
|
for job_name in jobs_with_params:
|
|
num_batches = jobs_with_params[job_name]["num_batches"]
|
|
job_config = CI_CONFIG.get_job_config(job_name)
|
|
for batch in jobs_with_params[job_name]["batches"]:
|
|
if self.is_pending(
|
|
job_name,
|
|
batch,
|
|
num_batches,
|
|
release_branch=is_release_branch
|
|
and job_config.required_on_release_branch,
|
|
):
|
|
continue
|
|
print(
|
|
f"Job [{job_name}_[{batch}/{num_batches}]] is not pending anymore"
|
|
)
|
|
|
|
# some_job_ready = True
|
|
jobs_with_params_copy[job_name]["batches"].remove(batch)
|
|
if not jobs_with_params_copy[job_name]["batches"]:
|
|
del jobs_with_params_copy[job_name]
|
|
|
|
if not self.is_successful(
|
|
job_name,
|
|
batch,
|
|
num_batches,
|
|
release_branch=is_release_branch
|
|
and job_config.required_on_release_branch,
|
|
):
|
|
print(
|
|
f"NOTE: Job [{job_name}:{batch}] finished but no success - remove from awaiting list, do not add to ready"
|
|
)
|
|
continue
|
|
if job_name in await_finished:
|
|
await_finished[job_name].append(batch)
|
|
else:
|
|
await_finished[job_name] = [batch]
|
|
jobs_with_params = jobs_with_params_copy
|
|
expired_sec = int(time.time()) - start_at
|
|
print(
|
|
f"...awaiting continues... seconds left [{TIMEOUT - expired_sec}]"
|
|
)
|
|
if await_finished:
|
|
GHActions.print_in_group(
|
|
f"Finished jobs, round [{round_cnt}]:",
|
|
[f"{job}:{batches}" for job, batches in await_finished.items()],
|
|
)
|
|
GHActions.print_in_group(
|
|
"Remaining jobs:",
|
|
[f"{job}:{params['batches']}" for job, params in jobs_with_params.items()],
|
|
)
|
|
return await_finished
|
|
|
|
|
|
@dataclass
|
|
class CiOptions:
|
|
# job will be included in the run if any keyword from the list matches job name
|
|
include_keywords: Optional[List[str]] = None
|
|
# job will be excluded in the run if any keyword from the list matches job name
|
|
exclude_keywords: Optional[List[str]] = None
|
|
|
|
# list of specified preconfigured ci sets to run
|
|
ci_sets: Optional[List[str]] = None
|
|
# list of specified jobs to run
|
|
ci_jobs: Optional[List[str]] = None
|
|
|
|
# btaches to run for all multi-batch jobs
|
|
job_batches: Optional[List[int]] = None
|
|
|
|
do_not_test: bool = False
|
|
no_ci_cache: bool = False
|
|
no_merge_commit: bool = False
|
|
|
|
def as_dict(self) -> Dict[str, Any]:
|
|
return asdict(self)
|
|
|
|
@staticmethod
|
|
def create_from_run_config(run_config: Dict[str, Any]) -> "CiOptions":
|
|
return CiOptions(**run_config["ci_options"])
|
|
|
|
@staticmethod
|
|
def create_from_pr_message(
|
|
debug_message: Optional[str], update_from_api: bool
|
|
) -> "CiOptions":
|
|
"""
|
|
Creates CiOptions instance based on tags found in PR body and/or commit message
|
|
@commit_message - may be provided directly for debugging purposes, otherwise it will be retrieved from git.
|
|
"""
|
|
res = CiOptions()
|
|
pr_info = PRInfo()
|
|
if (
|
|
not pr_info.is_pr and not debug_message
|
|
): # if commit_message is provided it's test/debug scenario - do not return
|
|
# CI options can be configured in PRs only
|
|
# if debug_message is provided - it's a test
|
|
return res
|
|
message = debug_message or GitRunner(set_cwd_to_git_root=True).run(
|
|
f"{GIT_PREFIX} log {pr_info.sha} --format=%B -n 1"
|
|
)
|
|
|
|
pattern = r"(#|- \[x\] +<!---)(\w+)"
|
|
matches = [match[-1] for match in re.findall(pattern, message)]
|
|
print(f"CI tags from commit message: [{matches}]")
|
|
|
|
if not debug_message: # to be skipped if debug/test
|
|
pr_info = PRInfo(
|
|
pr_event_from_api=update_from_api
|
|
) # Fetch updated PR body from GH API
|
|
matches_pr = [match[-1] for match in re.findall(pattern, pr_info.body)]
|
|
print(f"CI tags from PR body: [{matches_pr}]")
|
|
matches = list(set(matches + matches_pr))
|
|
|
|
if "do not test" in pr_info.labels:
|
|
# do_not_test could be set in GH labels
|
|
res.do_not_test = True
|
|
|
|
for match in matches:
|
|
if match.startswith("job_"):
|
|
if not res.ci_jobs:
|
|
res.ci_jobs = []
|
|
res.ci_jobs.append(match.removeprefix("job_"))
|
|
elif match.startswith("ci_set_") and match in CILabels:
|
|
if not res.ci_sets:
|
|
res.ci_sets = []
|
|
res.ci_sets.append(match)
|
|
elif match.startswith("ci_include_"):
|
|
if not res.include_keywords:
|
|
res.include_keywords = []
|
|
res.include_keywords.append(
|
|
normalize_check_name(match.removeprefix("ci_include_"))
|
|
)
|
|
elif match.startswith("ci_exclude_"):
|
|
if not res.exclude_keywords:
|
|
res.exclude_keywords = []
|
|
res.exclude_keywords.append(
|
|
normalize_check_name(match.removeprefix("ci_exclude_"))
|
|
)
|
|
elif match == CILabels.NO_CI_CACHE:
|
|
res.no_ci_cache = True
|
|
print("NOTE: CI Cache will be disabled")
|
|
elif match == CILabels.DO_NOT_TEST_LABEL:
|
|
res.do_not_test = True
|
|
elif match == CILabels.NO_MERGE_COMMIT:
|
|
res.no_merge_commit = True
|
|
print("NOTE: Merge Commit will be disabled")
|
|
elif match.startswith("batch_"):
|
|
batches = []
|
|
try:
|
|
batches = [
|
|
int(batch) for batch in match.removeprefix("batch_").split("_")
|
|
]
|
|
except Exception:
|
|
print(f"ERROR: failed to parse commit tag [{match}] - skip")
|
|
if batches:
|
|
if not res.job_batches:
|
|
res.job_batches = []
|
|
res.job_batches += batches
|
|
res.job_batches = list(set(res.job_batches))
|
|
else:
|
|
print(
|
|
f"WARNING: Invalid tag in commit message or PR body [{match}] - skip"
|
|
)
|
|
|
|
return res
|
|
|
|
def apply(
|
|
self,
|
|
jobs_to_do: List[str],
|
|
jobs_to_skip: List[str],
|
|
jobs_params: Dict[str, Dict[str, Any]],
|
|
pr_info: PRInfo,
|
|
) -> Tuple[List[str], List[str], Dict[str, Dict[str, Any]]]:
|
|
"""
|
|
Applies specified options on CI Run Config
|
|
Returns updated jobs_to_do, jobs_to_skip, jobs_params
|
|
"""
|
|
jobs_to_do_requested = [] # type: List[str]
|
|
|
|
# -1. Handle "ci_exclude_" tags if any
|
|
if self.exclude_keywords:
|
|
new_jobs_to_do = list(jobs_to_do)
|
|
for job in jobs_to_do:
|
|
found = False
|
|
for keyword in self.exclude_keywords:
|
|
if keyword in normalize_check_name(job):
|
|
print(
|
|
f"Job [{job}] matches Exclude keyword [{keyword}] - remove"
|
|
)
|
|
found = True
|
|
break
|
|
if found:
|
|
new_jobs_to_do.remove(job)
|
|
jobs_to_do = new_jobs_to_do
|
|
|
|
# 0. Handle "ci_include_" tags if any
|
|
if self.include_keywords:
|
|
for job in jobs_to_do:
|
|
found = False
|
|
for keyword in self.include_keywords:
|
|
if keyword in normalize_check_name(job):
|
|
print(f"Job [{job}] matches Include keyword [{keyword}] - add")
|
|
found = True
|
|
break
|
|
if found:
|
|
job_with_parents = CI_CONFIG.get_job_with_parents(job)
|
|
for job in job_with_parents:
|
|
if job in jobs_to_do and job not in jobs_to_do_requested:
|
|
jobs_to_do_requested.append(job)
|
|
print(
|
|
f"WARNING: Include tags are set but no job configured - Invalid tags, probably [{self.include_keywords}]"
|
|
)
|
|
if JobNames.STYLE_CHECK not in jobs_to_do_requested:
|
|
# Style check must not be omitted
|
|
jobs_to_do_requested.append(JobNames.STYLE_CHECK)
|
|
|
|
# FIXME: to be removed in favor of include/exclude
|
|
# 1. Handle "ci_set_" tags if any
|
|
if self.ci_sets:
|
|
for tag in self.ci_sets:
|
|
label_config = CI_CONFIG.get_label_config(tag)
|
|
assert label_config, f"Unknown tag [{tag}]"
|
|
print(
|
|
f"NOTE: CI Set's tag: [{tag}], add jobs: [{label_config.run_jobs}]"
|
|
)
|
|
jobs_to_do_requested += label_config.run_jobs
|
|
|
|
# FIXME: to be removed in favor of include/exclude
|
|
# 2. Handle "job_" tags if any
|
|
if self.ci_jobs:
|
|
for job in self.ci_jobs:
|
|
job_with_parents = CI_CONFIG.get_job_with_parents(job)
|
|
print(
|
|
f"NOTE: CI Job's tag: [#job_{job}], add jobs: [{job_with_parents}]"
|
|
)
|
|
# always add requested job itself, even if it could be skipped
|
|
jobs_to_do_requested.append(job_with_parents[0])
|
|
for parent in job_with_parents[1:]:
|
|
if parent in jobs_to_do and parent not in jobs_to_do_requested:
|
|
jobs_to_do_requested.append(parent)
|
|
|
|
# 3. Handle "do not test"
|
|
if self.do_not_test:
|
|
label_config = CI_CONFIG.get_label_config(CILabels.DO_NOT_TEST_LABEL)
|
|
assert label_config
|
|
print(
|
|
f"NOTE: CI 'do not test' setting applied, set jobs: [{label_config.run_jobs}]"
|
|
)
|
|
if jobs_to_do_requested:
|
|
print(
|
|
"WARNING: 'do not test' is used alongside with other CI modifying tags - 'do not test' prevails"
|
|
)
|
|
jobs_to_do_requested = list(label_config.run_jobs)
|
|
|
|
if jobs_to_do_requested:
|
|
jobs_to_do_requested = list(set(jobs_to_do_requested))
|
|
print(
|
|
f"NOTE: Only specific job(s) were requested by user's input: [{jobs_to_do_requested}]"
|
|
)
|
|
jobs_to_do = list(
|
|
set(job for job in jobs_to_do_requested if job not in jobs_to_skip)
|
|
)
|
|
# if requested job does not have params in jobs_params (it happens for "run_by_label" job)
|
|
# we need to add params - otherwise it won't run as "batches" list will be empty
|
|
for job in jobs_to_do:
|
|
if job not in jobs_params:
|
|
job_config = CI_CONFIG.get_job_config(job)
|
|
num_batches = job_config.num_batches
|
|
jobs_params[job] = {
|
|
"batches": list(range(num_batches)),
|
|
"num_batches": num_batches,
|
|
"run_if_ci_option_include_set": job_config.run_by_ci_option
|
|
and pr_info.is_pr,
|
|
}
|
|
|
|
# 4. Handle "batch_" tags
|
|
if self.job_batches:
|
|
print(
|
|
f"NOTE: Only specific job batches were requested [{self.job_batches}]"
|
|
)
|
|
for job, params in jobs_params.items():
|
|
if params["num_batches"] > 1:
|
|
params["batches"] = self.job_batches
|
|
|
|
for job in jobs_to_do[:]:
|
|
job_param = jobs_params[job]
|
|
if (
|
|
job_param["run_if_ci_option_include_set"]
|
|
and job not in jobs_to_do_requested
|
|
):
|
|
print(
|
|
f"Erasing job '{job}' from list because it's not in included set, but will run only by include"
|
|
)
|
|
jobs_to_skip.append(job)
|
|
jobs_to_do.remove(job)
|
|
|
|
return jobs_to_do, jobs_to_skip, jobs_params
|
|
|
|
|
|
def get_check_name(check_name: str, batch: int, num_batches: int) -> str:
|
|
res = check_name
|
|
if num_batches > 1:
|
|
res = f"{check_name} [{batch+1}/{num_batches}]"
|
|
return res
|
|
|
|
|
|
def normalize_check_name(check_name: str) -> str:
|
|
res = check_name.lower()
|
|
for r in ((" ", "_"), ("(", "_"), (")", "_"), (",", "_"), ("/", "_")):
|
|
res = res.replace(*r)
|
|
return res
|
|
|
|
|
|
def parse_args(parser: argparse.ArgumentParser) -> argparse.Namespace:
|
|
# FIXME: consider switching to sub_parser for configure, pre, run, post actions
|
|
parser.add_argument(
|
|
"--configure",
|
|
action="store_true",
|
|
help="Action that configures ci run. Calculates digests, checks job to be executed, generates json output",
|
|
)
|
|
parser.add_argument(
|
|
"--update-gh-statuses",
|
|
action="store_true",
|
|
help="Action that recreate success GH statuses for jobs that finished successfully in past and will be skipped this time",
|
|
)
|
|
parser.add_argument(
|
|
"--pre",
|
|
action="store_true",
|
|
help="Action that executes prerequesetes for the job provided in --job-name",
|
|
)
|
|
parser.add_argument(
|
|
"--run",
|
|
action="store_true",
|
|
help="Action that executes run action for specified --job-name. run_command must be configured for a given job name.",
|
|
)
|
|
parser.add_argument(
|
|
"--post",
|
|
action="store_true",
|
|
help="Action that executes post actions for the job provided in --job-name",
|
|
)
|
|
parser.add_argument(
|
|
"--mark-success",
|
|
action="store_true",
|
|
help="Action that marks job provided in --job-name (with batch provided in --batch) as successful",
|
|
)
|
|
parser.add_argument(
|
|
"--job-name",
|
|
default="",
|
|
type=str,
|
|
help="Job name as in config",
|
|
)
|
|
parser.add_argument(
|
|
"--run-command",
|
|
default="",
|
|
type=str,
|
|
help="A run command to run in --run action. Will override run_command from a job config if any",
|
|
)
|
|
parser.add_argument(
|
|
"--batch",
|
|
default=-1,
|
|
type=int,
|
|
help="Current batch number (required for --mark-success), -1 or omit for single-batch job",
|
|
)
|
|
parser.add_argument(
|
|
"--infile",
|
|
default="",
|
|
type=str,
|
|
help="Input json file or json string with ci run config",
|
|
)
|
|
parser.add_argument(
|
|
"--outfile",
|
|
default="",
|
|
type=str,
|
|
required=False,
|
|
help="output file to write json result to, if not set - stdout",
|
|
)
|
|
parser.add_argument(
|
|
"--pretty",
|
|
action="store_true",
|
|
default=False,
|
|
help="makes json output pretty formatted",
|
|
)
|
|
parser.add_argument(
|
|
"--skip-docker",
|
|
action="store_true",
|
|
default=False,
|
|
help="skip fetching docker data from dockerhub, used in --configure action (for debugging)",
|
|
)
|
|
parser.add_argument(
|
|
"--docker-digest-or-latest",
|
|
action="store_true",
|
|
default=False,
|
|
help="temporary hack to fallback to latest if image with digest as a tag is not on docker hub",
|
|
)
|
|
parser.add_argument(
|
|
"--skip-jobs",
|
|
action="store_true",
|
|
default=False,
|
|
help="skip fetching data about job runs, used in --configure action (for debugging and nigthly ci)",
|
|
)
|
|
parser.add_argument(
|
|
"--force",
|
|
action="store_true",
|
|
default=False,
|
|
help="Used with --run, force the job to run, omitting the ci cache",
|
|
)
|
|
# FIXME: remove, not used
|
|
parser.add_argument(
|
|
"--rebuild-all-binaries",
|
|
action="store_true",
|
|
default=False,
|
|
help="[DEPRECATED. to be removed, once no wf use it] will create run config without skipping build jobs in any case, used in --configure action (for release branches)",
|
|
)
|
|
parser.add_argument(
|
|
"--commit-message",
|
|
default="",
|
|
help="debug option to test commit message processing",
|
|
)
|
|
return parser.parse_args()
|
|
|
|
|
|
# FIXME: rewrite the docker job as regular reusable_test job and move interaction with docker hub inside job script
|
|
# that way run config will be more clean, workflow more generic and less api calls to dockerhub
|
|
def check_missing_images_on_dockerhub(
|
|
image_name_tag: Dict[str, str], arch: Optional[str] = None
|
|
) -> Dict[str, str]:
|
|
"""
|
|
Checks missing images on dockerhub.
|
|
Works concurrently for all given images.
|
|
Docker must be logged in.
|
|
"""
|
|
|
|
def run_docker_command(
|
|
image: str, image_digest: str, arch: Optional[str] = None
|
|
) -> Dict:
|
|
"""
|
|
aux command for fetching single docker manifest
|
|
"""
|
|
command = [
|
|
"docker",
|
|
"manifest",
|
|
"inspect",
|
|
f"{image}:{image_digest}" if not arch else f"{image}:{image_digest}-{arch}",
|
|
]
|
|
|
|
process = subprocess.run(
|
|
command,
|
|
stdout=subprocess.PIPE,
|
|
stderr=subprocess.PIPE,
|
|
text=True,
|
|
check=False,
|
|
)
|
|
|
|
return {
|
|
"image": image,
|
|
"image_digest": image_digest,
|
|
"arch": arch,
|
|
"stdout": process.stdout,
|
|
"stderr": process.stderr,
|
|
"return_code": process.returncode,
|
|
}
|
|
|
|
result: Dict[str, str] = {}
|
|
with concurrent.futures.ThreadPoolExecutor() as executor:
|
|
futures = [
|
|
executor.submit(run_docker_command, image, tag, arch)
|
|
for image, tag in image_name_tag.items()
|
|
]
|
|
|
|
responses = [
|
|
future.result() for future in concurrent.futures.as_completed(futures)
|
|
]
|
|
for resp in responses:
|
|
name, stdout, stderr, digest, arch = (
|
|
resp["image"],
|
|
resp["stdout"],
|
|
resp["stderr"],
|
|
resp["image_digest"],
|
|
resp["arch"],
|
|
)
|
|
if stderr:
|
|
if stderr.startswith("no such manifest"):
|
|
result[name] = digest
|
|
else:
|
|
print(f"Error: Unknown error: {stderr}, {name}, {arch}")
|
|
elif stdout:
|
|
if "mediaType" in stdout:
|
|
pass
|
|
else:
|
|
print(f"Error: Unknown response: {stdout}")
|
|
assert False, "FIXME"
|
|
else:
|
|
print(f"Error: No response for {name}, {digest}, {arch}")
|
|
assert False, "FIXME"
|
|
return result
|
|
|
|
|
|
def _pre_action(s3, indata, pr_info):
|
|
CommitStatusData.cleanup()
|
|
JobReport.cleanup()
|
|
BuildResult.cleanup()
|
|
ci_cache = CiCache(s3, indata["jobs_data"]["digests"])
|
|
|
|
# for release/master branches reports must be from the same branches
|
|
report_prefix = normalize_string(pr_info.head_ref) if pr_info.number == 0 else ""
|
|
print(
|
|
f"Use report prefix [{report_prefix}], pr_num [{pr_info.number}], head_ref [{pr_info.head_ref}]"
|
|
)
|
|
reports_files = ci_cache.download_build_reports(file_prefix=report_prefix)
|
|
print(f"Pre action done. Report files [{reports_files}] have been downloaded")
|
|
|
|
|
|
def _mark_success_action(
|
|
s3: S3Helper,
|
|
indata: Dict[str, Any],
|
|
pr_info: PRInfo,
|
|
job: str,
|
|
batch: int,
|
|
) -> None:
|
|
ci_cache = CiCache(s3, indata["jobs_data"]["digests"])
|
|
job_config = CI_CONFIG.get_job_config(job)
|
|
num_batches = job_config.num_batches
|
|
# if batch is not provided - set to 0
|
|
batch = 0 if batch == -1 else batch
|
|
assert (
|
|
0 <= batch < num_batches
|
|
), f"--batch must be provided and in range [0, {num_batches}) for {job}"
|
|
|
|
# FIXME: find generic design for propagating and handling job status (e.g. stop using statuses in GH api)
|
|
# now job ca be build job w/o status data, any other job that exit with 0 with or w/o status data
|
|
if CI_CONFIG.is_build_job(job):
|
|
# there is no CommitStatus for build jobs
|
|
# create dummy status relying on JobReport
|
|
# FIXME: consider creating commit status for build jobs too, to treat everything the same way
|
|
job_report = JobReport.load() if JobReport.exist() else None
|
|
if job_report and job_report.status == SUCCESS:
|
|
CommitStatusData(
|
|
SUCCESS,
|
|
"dummy description",
|
|
"dummy_url",
|
|
pr_num=pr_info.number,
|
|
sha=pr_info.sha,
|
|
).dump_status()
|
|
|
|
job_status = None
|
|
if CommitStatusData.exist():
|
|
# normal scenario
|
|
job_status = CommitStatusData.load_status()
|
|
else:
|
|
# apparently exit after rerun-helper check
|
|
# do nothing, exit without failure
|
|
print(f"ERROR: no status file for job [{job}]")
|
|
|
|
if job_config.run_always or job_config.run_by_label:
|
|
print(f"Job [{job}] runs always or by label in CI - do not cache")
|
|
else:
|
|
if pr_info.is_master:
|
|
pass
|
|
# delete method is disabled for ci_cache. need it?
|
|
# pending enabled for master branch jobs only
|
|
# ci_cache.delete_pending(job, batch, num_batches, release_branch=True)
|
|
if job_status and job_status.is_ok():
|
|
ci_cache.push_successful(
|
|
job, batch, num_batches, job_status, pr_info.is_release_branch
|
|
)
|
|
print(f"Job [{job}] is ok")
|
|
elif job_status and not job_status.is_ok():
|
|
ci_cache.push_failed(
|
|
job, batch, num_batches, job_status, pr_info.is_release_branch
|
|
)
|
|
print(f"Job [{job}] is failed with status [{job_status.status}]")
|
|
else:
|
|
job_status = CommitStatusData(
|
|
description="dummy description", status=ERROR, report_url="dummy url"
|
|
)
|
|
ci_cache.push_failed(
|
|
job, batch, num_batches, job_status, pr_info.is_release_branch
|
|
)
|
|
print(f"No CommitStatusData for [{job}], push dummy failure to ci_cache")
|
|
|
|
|
|
def _print_results(result: Any, outfile: Optional[str], pretty: bool = False) -> None:
|
|
if outfile:
|
|
with open(outfile, "w", encoding="utf-8") as f:
|
|
if isinstance(result, str):
|
|
print(result, file=f)
|
|
elif isinstance(result, dict):
|
|
print(json.dumps(result, indent=2 if pretty else None), file=f)
|
|
else:
|
|
raise AssertionError(f"Unexpected type for 'res': {type(result)}")
|
|
else:
|
|
if isinstance(result, str):
|
|
print(result)
|
|
elif isinstance(result, dict):
|
|
print(json.dumps(result, indent=2 if pretty else None))
|
|
else:
|
|
raise AssertionError(f"Unexpected type for 'res': {type(result)}")
|
|
|
|
|
|
def _configure_docker_jobs(docker_digest_or_latest: bool) -> Dict:
|
|
print("::group::Docker images check")
|
|
# generate docker jobs data
|
|
docker_digester = DockerDigester()
|
|
imagename_digest_dict = (
|
|
docker_digester.get_all_digests()
|
|
) # 'image name - digest' mapping
|
|
images_info = docker_images_helper.get_images_info()
|
|
|
|
# FIXME: we need login as docker manifest inspect goes directly to one of the *.docker.com hosts instead of "registry-mirrors" : ["http://dockerhub-proxy.dockerhub-proxy-zone:5000"]
|
|
# find if it's possible to use the setting of /etc/docker/daemon.json (https://github.com/docker/cli/issues/4484#issuecomment-1688095463)
|
|
docker_images_helper.docker_login()
|
|
missing_multi_dict = check_missing_images_on_dockerhub(imagename_digest_dict)
|
|
missing_multi = list(missing_multi_dict)
|
|
missing_amd64 = []
|
|
missing_aarch64 = []
|
|
if not docker_digest_or_latest:
|
|
# look for missing arm and amd images only among missing multiarch manifests @missing_multi_dict
|
|
# to avoid extra dockerhub api calls
|
|
missing_amd64 = list(
|
|
check_missing_images_on_dockerhub(missing_multi_dict, "amd64")
|
|
)
|
|
# FIXME: WA until full arm support: skip not supported arm images
|
|
missing_aarch64 = list(
|
|
check_missing_images_on_dockerhub(
|
|
{
|
|
im: digest
|
|
for im, digest in missing_multi_dict.items()
|
|
if not images_info[im]["only_amd64"]
|
|
},
|
|
"aarch64",
|
|
)
|
|
)
|
|
else:
|
|
if missing_multi:
|
|
assert False, f"Missing images [{missing_multi}], cannot proceed"
|
|
print("::endgroup::")
|
|
|
|
return {
|
|
"images": imagename_digest_dict,
|
|
"missing_aarch64": missing_aarch64,
|
|
"missing_amd64": missing_amd64,
|
|
"missing_multi": missing_multi,
|
|
}
|
|
|
|
|
|
def _configure_jobs(
|
|
job_digester: JobDigester,
|
|
s3: S3Helper,
|
|
pr_info: PRInfo,
|
|
ci_options: CiOptions,
|
|
) -> Dict:
|
|
## a. digest each item from the config
|
|
job_digester = JobDigester()
|
|
jobs_params: Dict[str, Dict] = {}
|
|
jobs_to_do: List[str] = []
|
|
jobs_to_skip: List[str] = []
|
|
digests: Dict[str, str] = {}
|
|
|
|
# FIXME: find better place for these config variables
|
|
DOCS_CHECK_JOBS = [JobNames.DOCS_CHECK, JobNames.STYLE_CHECK]
|
|
MQ_JOBS = [JobNames.STYLE_CHECK, JobNames.FAST_TEST]
|
|
# Must always calculate digest for these jobs for CI Cache to function (they define s3 paths where records are stored)
|
|
REQUIRED_DIGESTS = [JobNames.DOCS_CHECK, Build.PACKAGE_RELEASE]
|
|
if pr_info.has_changes_in_documentation_only():
|
|
print(f"WARNING: Only docs are changed - will run only [{DOCS_CHECK_JOBS}]")
|
|
if pr_info.is_merge_queue:
|
|
print(f"WARNING: It's a MQ run - will run only [{MQ_JOBS}]")
|
|
|
|
print("::group::Job Digests")
|
|
for job in CI_CONFIG.job_generator(pr_info.head_ref if CI else "dummy_branch_name"):
|
|
if (
|
|
pr_info.is_merge_queue
|
|
and job not in MQ_JOBS
|
|
and job not in REQUIRED_DIGESTS
|
|
):
|
|
# We still need digest for JobNames.DOCS_CHECK since CiCache depends on it (FIXME)
|
|
continue
|
|
if (
|
|
pr_info.has_changes_in_documentation_only()
|
|
and job not in DOCS_CHECK_JOBS
|
|
and job not in REQUIRED_DIGESTS
|
|
):
|
|
continue
|
|
digest = job_digester.get_job_digest(CI_CONFIG.get_digest_config(job))
|
|
digests[job] = digest
|
|
print(f" job [{job.rjust(50)}] has digest [{digest}]")
|
|
print("::endgroup::")
|
|
|
|
## b. check what we need to run
|
|
ci_cache = None
|
|
if not ci_options.no_ci_cache and CI:
|
|
ci_cache = CiCache(s3, digests).update()
|
|
ci_cache.print_status()
|
|
|
|
jobs_to_wait: Dict[str, Dict[str, Any]] = {}
|
|
randomization_buckets = {} # type: Dict[str, Set[str]]
|
|
|
|
for job, digest in digests.items():
|
|
job_config = CI_CONFIG.get_job_config(job)
|
|
num_batches: int = job_config.num_batches
|
|
batches_to_do: List[int] = []
|
|
add_to_skip = False
|
|
|
|
if pr_info.is_merge_queue and job not in MQ_JOBS:
|
|
continue
|
|
if job_config.pr_only and pr_info.is_release_branch:
|
|
continue
|
|
if (
|
|
job_config.release_only
|
|
and not job_config.run_by_ci_option
|
|
and not pr_info.is_release_branch
|
|
):
|
|
continue
|
|
|
|
# fill job randomization buckets (for jobs with configured @random_bucket property))
|
|
if job_config.random_bucket:
|
|
if not job_config.random_bucket in randomization_buckets:
|
|
randomization_buckets[job_config.random_bucket] = set()
|
|
randomization_buckets[job_config.random_bucket].add(job)
|
|
|
|
for batch in range(num_batches): # type: ignore
|
|
if job_config.run_by_label:
|
|
# this job controlled by label, add to todo if its label is set in pr
|
|
if job_config.run_by_label in pr_info.labels:
|
|
batches_to_do.append(batch)
|
|
elif job_config.run_always:
|
|
# always add to todo
|
|
batches_to_do.append(batch)
|
|
elif not ci_cache:
|
|
batches_to_do.append(batch)
|
|
elif not ci_cache.is_successful(
|
|
job,
|
|
batch,
|
|
num_batches,
|
|
release_branch=pr_info.is_release_branch
|
|
and job_config.required_on_release_branch,
|
|
):
|
|
# ci cache is enabled and job is not in the cache - add
|
|
batches_to_do.append(batch)
|
|
|
|
# check if it's pending in the cache
|
|
if ci_cache.is_pending(
|
|
job,
|
|
batch,
|
|
num_batches,
|
|
release_branch=pr_info.is_release_branch
|
|
and job_config.required_on_release_branch,
|
|
):
|
|
if job in jobs_to_wait:
|
|
jobs_to_wait[job]["batches"].append(batch)
|
|
else:
|
|
jobs_to_wait[job] = {
|
|
"batches": [batch],
|
|
"num_batches": num_batches,
|
|
}
|
|
else:
|
|
add_to_skip = True
|
|
|
|
if batches_to_do:
|
|
jobs_to_do.append(job)
|
|
jobs_params[job] = {
|
|
"batches": batches_to_do,
|
|
"num_batches": num_batches,
|
|
"run_if_ci_option_include_set": job_config.run_by_ci_option
|
|
and pr_info.is_pr,
|
|
}
|
|
elif add_to_skip:
|
|
# treat job as being skipped only if it's controlled by digest
|
|
jobs_to_skip.append(job)
|
|
|
|
if not pr_info.is_release_branch:
|
|
# randomization bucket filtering (pick one random job from each bucket, for jobs with configured random_bucket property)
|
|
for _, jobs in randomization_buckets.items():
|
|
jobs_to_remove_randomization = set()
|
|
bucket_ = list(jobs)
|
|
random.shuffle(bucket_)
|
|
while len(bucket_) > 1:
|
|
random_job = bucket_.pop()
|
|
if random_job in jobs_to_do:
|
|
jobs_to_remove_randomization.add(random_job)
|
|
if jobs_to_remove_randomization:
|
|
print(
|
|
f"Following jobs will be removed due to randomization bucket: [{jobs_to_remove_randomization}]"
|
|
)
|
|
jobs_to_do = [
|
|
job for job in jobs_to_do if job not in jobs_to_remove_randomization
|
|
]
|
|
|
|
jobs_to_do, jobs_to_skip, jobs_params = ci_options.apply(
|
|
jobs_to_do, jobs_to_skip, jobs_params, pr_info
|
|
)
|
|
|
|
return {
|
|
"digests": digests,
|
|
"jobs_to_do": jobs_to_do,
|
|
"jobs_to_skip": jobs_to_skip,
|
|
"jobs_to_wait": {
|
|
job: params for job, params in jobs_to_wait.items() if job in jobs_to_do
|
|
},
|
|
"jobs_params": {
|
|
job: params for job, params in jobs_params.items() if job in jobs_to_do
|
|
},
|
|
}
|
|
|
|
|
|
def _generate_ci_stage_config(jobs_data: Dict[str, Any]) -> Dict[str, Dict[str, Any]]:
|
|
"""
|
|
populates GH Actions' workflow with real jobs
|
|
"Builds_1": [{"job_name": NAME, "runner_type": RUNER_TYPE}]
|
|
"Tests_1": [{"job_name": NAME, "runner_type": RUNER_TYPE}]
|
|
...
|
|
"""
|
|
result = {} # type: Dict[str, Any]
|
|
stages_to_do = []
|
|
for job in jobs_data["jobs_to_do"]:
|
|
stage_type = CI_CONFIG.get_job_ci_stage(job)
|
|
if stage_type == CIStages.NA:
|
|
continue
|
|
if stage_type not in result:
|
|
result[stage_type] = []
|
|
stages_to_do.append(stage_type)
|
|
result[stage_type].append(
|
|
{"job_name": job, "runner_type": CI_CONFIG.get_runner_type(job)}
|
|
)
|
|
result["stages_to_do"] = stages_to_do
|
|
return result
|
|
|
|
|
|
def _create_gh_status(
|
|
commit: Any, job: str, batch: int, num_batches: int, job_status: CommitStatusData
|
|
) -> None:
|
|
print(f"Going to re-create GH status for job [{job}]")
|
|
assert job_status.status == SUCCESS, "BUG!"
|
|
commit.create_status(
|
|
state=job_status.status,
|
|
target_url=job_status.report_url,
|
|
description=format_description(
|
|
f"Reused from [{job_status.pr_num}-{job_status.sha[0:8]}]: "
|
|
f"{job_status.description}"
|
|
),
|
|
context=get_check_name(job, batch=batch, num_batches=num_batches),
|
|
)
|
|
|
|
|
|
def _update_gh_statuses_action(indata: Dict, s3: S3Helper) -> None:
|
|
if CiOptions.create_from_run_config(indata).no_ci_cache:
|
|
print("CI cache is disabled - skip restoring commit statuses from CI cache")
|
|
return
|
|
job_digests = indata["jobs_data"]["digests"]
|
|
jobs_to_skip = indata["jobs_data"]["jobs_to_skip"]
|
|
jobs_to_do = indata["jobs_data"]["jobs_to_do"]
|
|
ci_cache = CiCache(s3, job_digests).update().fetch_records_data().print_status()
|
|
|
|
# create GH status
|
|
pr_info = PRInfo()
|
|
commit = get_commit(GitHub(get_best_robot_token(), per_page=100), pr_info.sha)
|
|
|
|
def _concurrent_create_status(job: str, batch: int, num_batches: int) -> None:
|
|
job_status = ci_cache.get_successful(job, batch, num_batches)
|
|
if not job_status:
|
|
return
|
|
_create_gh_status(commit, job, batch, num_batches, job_status)
|
|
|
|
with concurrent.futures.ThreadPoolExecutor() as executor:
|
|
futures = []
|
|
for job in job_digests:
|
|
if job not in jobs_to_skip and job not in jobs_to_do:
|
|
# no need to create status for job that are not supposed to be executed
|
|
continue
|
|
if CI_CONFIG.is_build_job(job):
|
|
# no GH status for build jobs
|
|
continue
|
|
job_config = CI_CONFIG.get_job_config(job)
|
|
if not job_config:
|
|
# there might be a new job that does not exist on this branch - skip it
|
|
continue
|
|
for batch in range(job_config.num_batches):
|
|
future = executor.submit(
|
|
_concurrent_create_status, job, batch, job_config.num_batches
|
|
)
|
|
futures.append(future)
|
|
done, _ = concurrent.futures.wait(futures)
|
|
for future in done:
|
|
try:
|
|
_ = future.result()
|
|
except Exception as e:
|
|
raise e
|
|
print("Going to update overall CI report")
|
|
set_status_comment(commit, pr_info)
|
|
print("... CI report update - done")
|
|
|
|
|
|
def _fetch_commit_tokens(message: str, pr_info: PRInfo) -> List[str]:
|
|
pattern = r"(#|- \[x\] +<!---)(\w+)"
|
|
matches = [match[-1] for match in re.findall(pattern, message)]
|
|
res = [
|
|
match
|
|
for match in matches
|
|
if match in CILabels or match.startswith("job_") or match.startswith("batch_")
|
|
]
|
|
print(f"CI modifyers from commit message: [{res}]")
|
|
res_2 = []
|
|
if pr_info.is_pr:
|
|
matches = [match[-1] for match in re.findall(pattern, pr_info.body)]
|
|
res_2 = [
|
|
match
|
|
for match in matches
|
|
if match in CILabels
|
|
or match.startswith("job_")
|
|
or match.startswith("batch_")
|
|
]
|
|
print(f"CI modifyers from PR body: [{res_2}]")
|
|
return list(set(res + res_2))
|
|
|
|
|
|
def _upload_build_artifacts(
|
|
pr_info: PRInfo,
|
|
build_name: str,
|
|
ci_cache: CiCache,
|
|
job_report: JobReport,
|
|
s3: S3Helper,
|
|
s3_destination: str,
|
|
) -> str:
|
|
# There are ugly artifacts for the performance test. FIXME:
|
|
s3_performance_path = "/".join(
|
|
(
|
|
get_release_or_pr(pr_info, get_version_from_repo())[1],
|
|
pr_info.sha,
|
|
normalize_string(build_name),
|
|
"performance.tar.zst",
|
|
)
|
|
)
|
|
performance_urls = []
|
|
assert job_report.build_dir_for_upload, "Must be set for build job"
|
|
performance_path = Path(job_report.build_dir_for_upload) / "performance.tar.zst"
|
|
if performance_path.exists():
|
|
performance_urls.append(
|
|
s3.upload_build_file_to_s3(performance_path, s3_performance_path)
|
|
)
|
|
print(
|
|
"Uploaded performance.tar.zst to %s, now delete to avoid duplication",
|
|
performance_urls[0],
|
|
)
|
|
performance_path.unlink()
|
|
build_urls = (
|
|
s3.upload_build_directory_to_s3(
|
|
Path(job_report.build_dir_for_upload),
|
|
s3_destination,
|
|
keep_dirs_in_s3_path=False,
|
|
upload_symlinks=False,
|
|
)
|
|
+ performance_urls
|
|
)
|
|
print("::notice ::Build URLs: {}".format("\n".join(build_urls)))
|
|
log_path = Path(job_report.additional_files[0])
|
|
log_url = ""
|
|
if log_path.exists():
|
|
log_url = s3.upload_build_file_to_s3(
|
|
log_path, s3_destination + "/" + log_path.name
|
|
)
|
|
print(f"::notice ::Log URL: {log_url}")
|
|
|
|
# generate and upload build report
|
|
build_result = BuildResult(
|
|
build_name,
|
|
log_url,
|
|
build_urls,
|
|
job_report.version,
|
|
job_report.status,
|
|
int(job_report.duration),
|
|
GITHUB_JOB_API_URL(),
|
|
head_ref=pr_info.head_ref,
|
|
pr_number=pr_info.number,
|
|
)
|
|
report_url = ci_cache.upload_build_report(build_result)
|
|
print(f"Report file has been uploaded to [{report_url}]")
|
|
|
|
# Upload head master binaries
|
|
static_bin_name = CI_CONFIG.build_config[build_name].static_binary_name
|
|
if pr_info.is_master and static_bin_name:
|
|
# Full binary with debug info:
|
|
s3_path_full = "/".join((pr_info.base_ref, static_bin_name, "clickhouse-full"))
|
|
binary_full = Path(job_report.build_dir_for_upload) / "clickhouse"
|
|
url_full = s3.upload_build_file_to_s3(binary_full, s3_path_full)
|
|
print(f"::notice ::Binary static URL (with debug info): {url_full}")
|
|
|
|
# Stripped binary without debug info:
|
|
s3_path_compact = "/".join((pr_info.base_ref, static_bin_name, "clickhouse"))
|
|
binary_compact = Path(job_report.build_dir_for_upload) / "clickhouse-stripped"
|
|
url_compact = s3.upload_build_file_to_s3(binary_compact, s3_path_compact)
|
|
print(f"::notice ::Binary static URL (compact): {url_compact}")
|
|
|
|
return log_url
|
|
|
|
|
|
def _upload_build_profile_data(
|
|
pr_info: PRInfo,
|
|
build_name: str,
|
|
job_report: JobReport,
|
|
git_runner: GitRunner,
|
|
ch_helper: ClickHouseHelper,
|
|
) -> None:
|
|
ci_logs_credentials = CiLogsCredentials(Path("/dev/null"))
|
|
if ci_logs_credentials.host:
|
|
instance_type = get_instance_type()
|
|
instance_id = get_instance_id()
|
|
query = f"""INSERT INTO build_time_trace
|
|
(
|
|
pull_request_number,
|
|
commit_sha,
|
|
check_start_time,
|
|
check_name,
|
|
instance_type,
|
|
instance_id,
|
|
file,
|
|
library,
|
|
time,
|
|
pid,
|
|
tid,
|
|
ph,
|
|
ts,
|
|
dur,
|
|
cat,
|
|
name,
|
|
detail,
|
|
count,
|
|
avgMs,
|
|
args_name
|
|
)
|
|
SELECT {pr_info.number}, '{pr_info.sha}', '{job_report.start_time}', '{build_name}', '{instance_type}', '{instance_id}', *
|
|
FROM input('
|
|
file String,
|
|
library String,
|
|
time DateTime64(6),
|
|
pid UInt32,
|
|
tid UInt32,
|
|
ph String,
|
|
ts UInt64,
|
|
dur UInt64,
|
|
cat String,
|
|
name String,
|
|
detail String,
|
|
count UInt64,
|
|
avgMs UInt64,
|
|
args_name String')
|
|
FORMAT JSONCompactEachRow"""
|
|
|
|
auth = {
|
|
"X-ClickHouse-User": "ci",
|
|
"X-ClickHouse-Key": ci_logs_credentials.password,
|
|
}
|
|
url = f"https://{ci_logs_credentials.host}/"
|
|
profiles_dir = Path(TEMP_PATH) / "profiles_source"
|
|
profiles_dir.mkdir(parents=True, exist_ok=True)
|
|
print(
|
|
"Processing profile JSON files from %s",
|
|
Path(REPO_COPY) / "build_docker",
|
|
)
|
|
git_runner(
|
|
"./utils/prepare-time-trace/prepare-time-trace.sh "
|
|
f"build_docker {profiles_dir.absolute()}"
|
|
)
|
|
profile_data_file = Path(TEMP_PATH) / "profile.json"
|
|
with open(profile_data_file, "wb") as profile_fd:
|
|
for profile_source in profiles_dir.iterdir():
|
|
if profile_source.name not in (
|
|
"binary_sizes.txt",
|
|
"binary_symbols.txt",
|
|
):
|
|
with open(profiles_dir / profile_source, "rb") as ps_fd:
|
|
profile_fd.write(ps_fd.read())
|
|
|
|
print(
|
|
"::notice ::Log Uploading profile data, path: %s, size: %s, query: %s",
|
|
profile_data_file,
|
|
profile_data_file.stat().st_size,
|
|
query,
|
|
)
|
|
try:
|
|
ch_helper.insert_file(url, auth, query, profile_data_file)
|
|
except InsertException:
|
|
logging.error("Failed to insert profile data for the build, continue")
|
|
|
|
query = f"""INSERT INTO binary_sizes
|
|
(
|
|
pull_request_number,
|
|
commit_sha,
|
|
check_start_time,
|
|
check_name,
|
|
instance_type,
|
|
instance_id,
|
|
file,
|
|
size
|
|
)
|
|
SELECT {pr_info.number}, '{pr_info.sha}', '{job_report.start_time}', '{build_name}', '{instance_type}', '{instance_id}', file, size
|
|
FROM input('size UInt64, file String')
|
|
SETTINGS format_regexp = '^\\s*(\\d+) (.+)$'
|
|
FORMAT Regexp"""
|
|
|
|
binary_sizes_file = profiles_dir / "binary_sizes.txt"
|
|
|
|
print(
|
|
"::notice ::Log Uploading binary sizes data, path: %s, size: %s, query: %s",
|
|
binary_sizes_file,
|
|
binary_sizes_file.stat().st_size,
|
|
query,
|
|
)
|
|
try:
|
|
ch_helper.insert_file(url, auth, query, binary_sizes_file)
|
|
except InsertException:
|
|
logging.error("Failed to insert binary_sizes_file for the build, continue")
|
|
|
|
query = f"""INSERT INTO binary_symbols
|
|
(
|
|
pull_request_number,
|
|
commit_sha,
|
|
check_start_time,
|
|
check_name,
|
|
instance_type,
|
|
instance_id,
|
|
file,
|
|
address,
|
|
size,
|
|
type,
|
|
symbol
|
|
)
|
|
SELECT {pr_info.number}, '{pr_info.sha}', '{job_report.start_time}', '{build_name}', '{instance_type}', '{instance_id}',
|
|
file, reinterpretAsUInt64(reverse(unhex(address))), reinterpretAsUInt64(reverse(unhex(size))), type, symbol
|
|
FROM input('file String, address String, size String, type String, symbol String')
|
|
SETTINGS format_regexp = '^([^ ]+) ([0-9a-fA-F]+)(?: ([0-9a-fA-F]+))? (.) (.+)$'
|
|
FORMAT Regexp"""
|
|
|
|
binary_symbols_file = profiles_dir / "binary_symbols.txt"
|
|
|
|
print(
|
|
"::notice ::Log Uploading binary symbols data, path: %s, size: %s, query: %s",
|
|
binary_symbols_file,
|
|
binary_symbols_file.stat().st_size,
|
|
query,
|
|
)
|
|
try:
|
|
ch_helper.insert_file(url, auth, query, binary_symbols_file)
|
|
except InsertException:
|
|
logging.error(
|
|
"Failed to insert binary_symbols_file for the build, continue"
|
|
)
|
|
|
|
|
|
def _add_build_to_version_history(
|
|
pr_info: PRInfo,
|
|
job_report: JobReport,
|
|
version: str,
|
|
docker_tag: str,
|
|
ch_helper: ClickHouseHelper,
|
|
) -> None:
|
|
# with some probability we will not silently break this logic
|
|
assert pr_info.sha and pr_info.commit_html_url and pr_info.head_ref and version
|
|
|
|
data = {
|
|
"check_start_time": job_report.start_time,
|
|
"pull_request_number": pr_info.number,
|
|
"pull_request_url": pr_info.pr_html_url,
|
|
"commit_sha": pr_info.sha,
|
|
"commit_url": pr_info.commit_html_url,
|
|
"version": version,
|
|
"docker_tag": docker_tag,
|
|
"git_ref": pr_info.head_ref,
|
|
}
|
|
|
|
print(f"::notice ::Log Adding record to versions history: {data}")
|
|
|
|
ch_helper.insert_event_into(db="default", table="version_history", event=data)
|
|
|
|
|
|
def _run_test(job_name: str, run_command: str) -> int:
|
|
assert (
|
|
run_command or CI_CONFIG.get_job_config(job_name).run_command
|
|
), "Run command must be provided as input argument or be configured in job config"
|
|
|
|
if CI_CONFIG.get_job_config(job_name).timeout:
|
|
os.environ["KILL_TIMEOUT"] = str(CI_CONFIG.get_job_config(job_name).timeout)
|
|
|
|
if not run_command:
|
|
run_command = "/".join(
|
|
(os.path.dirname(__file__), CI_CONFIG.get_job_config(job_name).run_command)
|
|
)
|
|
if ".py" in run_command and not run_command.startswith("python"):
|
|
run_command = "python3 " + run_command
|
|
print("Use run command from a job config")
|
|
else:
|
|
print("Use run command from the workflow")
|
|
os.environ["CHECK_NAME"] = job_name
|
|
print(f"Going to start run command [{run_command}]")
|
|
process = subprocess.run(
|
|
run_command,
|
|
stdout=sys.stdout,
|
|
stderr=sys.stderr,
|
|
text=True,
|
|
check=False,
|
|
shell=True,
|
|
)
|
|
|
|
if process.returncode == 0:
|
|
print(f"Run action done for: [{job_name}]")
|
|
exit_code = 0
|
|
else:
|
|
print(
|
|
f"Run action failed for: [{job_name}] with exit code [{process.returncode}]"
|
|
)
|
|
exit_code = process.returncode
|
|
return exit_code
|
|
|
|
|
|
def _get_ext_check_name(check_name: str) -> str:
|
|
run_by_hash_num = int(os.getenv("RUN_BY_HASH_NUM", "0"))
|
|
run_by_hash_total = int(os.getenv("RUN_BY_HASH_TOTAL", "0"))
|
|
if run_by_hash_total > 1:
|
|
check_name_with_group = (
|
|
check_name + f" [{run_by_hash_num + 1}/{run_by_hash_total}]"
|
|
)
|
|
else:
|
|
check_name_with_group = check_name
|
|
return check_name_with_group
|
|
|
|
|
|
def main() -> int:
|
|
logging.basicConfig(level=logging.INFO)
|
|
exit_code = 0
|
|
parser = argparse.ArgumentParser(
|
|
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
|
|
)
|
|
args = parse_args(parser)
|
|
|
|
if args.mark_success or args.pre or args.run:
|
|
assert args.infile, "Run config must be provided via --infile"
|
|
assert args.job_name, "Job name must be provided via --job-name"
|
|
|
|
indata: Optional[Dict[str, Any]] = None
|
|
if args.infile:
|
|
if os.path.isfile(args.infile):
|
|
with open(args.infile, encoding="utf-8") as jfd:
|
|
indata = json.load(jfd)
|
|
else:
|
|
indata = json.loads(args.infile)
|
|
assert indata and isinstance(indata, dict), "Invalid --infile json"
|
|
|
|
result: Dict[str, Any] = {}
|
|
s3 = S3Helper()
|
|
pr_info = PRInfo()
|
|
git_runner = GitRunner(set_cwd_to_git_root=True)
|
|
|
|
### CONFIGURE action: start
|
|
if args.configure:
|
|
ci_options = CiOptions.create_from_pr_message(
|
|
args.commit_message or None, update_from_api=True
|
|
)
|
|
|
|
# tokens = _fetch_commit_tokens(message, pr_info)
|
|
if ci_options.no_merge_commit and CI:
|
|
git_runner.run(f"{GIT_PREFIX} checkout {pr_info.sha}")
|
|
git_ref = git_runner.run(f"{GIT_PREFIX} rev-parse HEAD")
|
|
|
|
docker_data = {}
|
|
git_ref = git_runner.run(f"{GIT_PREFIX} rev-parse HEAD")
|
|
|
|
# let's get CH version
|
|
version = get_version_from_repo(git=Git(True)).string
|
|
print(f"Got CH version for this commit: [{version}]")
|
|
|
|
docker_data = (
|
|
_configure_docker_jobs(args.docker_digest_or_latest)
|
|
if not args.skip_docker
|
|
else {}
|
|
)
|
|
|
|
job_digester = JobDigester()
|
|
build_digest = job_digester.get_job_digest(
|
|
CI_CONFIG.get_digest_config("package_release")
|
|
)
|
|
docs_digest = job_digester.get_job_digest(
|
|
CI_CONFIG.get_digest_config(JobNames.DOCS_CHECK)
|
|
)
|
|
jobs_data = (
|
|
_configure_jobs(
|
|
job_digester,
|
|
s3,
|
|
pr_info,
|
|
ci_options,
|
|
)
|
|
if not args.skip_jobs
|
|
else {}
|
|
)
|
|
|
|
if not args.skip_jobs:
|
|
ci_cache = CiCache(s3, jobs_data["digests"])
|
|
|
|
if pr_info.is_master:
|
|
# wait for pending jobs to be finished, await_jobs is a long blocking call
|
|
# wait pending jobs (for now only on release/master branches)
|
|
ready_jobs_batches_dict = ci_cache.await_jobs(
|
|
jobs_data.get("jobs_to_wait", {}), pr_info.is_release_branch
|
|
)
|
|
jobs_to_do = jobs_data["jobs_to_do"]
|
|
jobs_to_skip = jobs_data["jobs_to_skip"]
|
|
jobs_params = jobs_data["jobs_params"]
|
|
for job, batches in ready_jobs_batches_dict.items():
|
|
if job not in jobs_params:
|
|
print(f"WARNING: Job [{job}] is not in the params list")
|
|
continue
|
|
for batch in batches:
|
|
jobs_params[job]["batches"].remove(batch)
|
|
if not jobs_params[job]["batches"]:
|
|
jobs_to_do.remove(job)
|
|
jobs_to_skip.append(job)
|
|
del jobs_params[job]
|
|
|
|
# set planned jobs as in-progress in CI cache
|
|
for job in jobs_data["jobs_to_do"]:
|
|
config = CI_CONFIG.get_job_config(job)
|
|
if config.run_always or config.run_by_label:
|
|
continue
|
|
job_params = jobs_data["jobs_params"][job]
|
|
ci_cache.push_pending(
|
|
job,
|
|
job_params["batches"],
|
|
config.num_batches,
|
|
release_branch=pr_info.is_release_branch,
|
|
)
|
|
|
|
if "jobs_to_wait" in jobs_data:
|
|
del jobs_data["jobs_to_wait"]
|
|
|
|
# conclude results
|
|
result["git_ref"] = git_ref
|
|
result["version"] = version
|
|
result["build"] = build_digest
|
|
result["docs"] = docs_digest
|
|
result["ci_options"] = ci_options.as_dict()
|
|
if not args.skip_jobs:
|
|
result["stages_data"] = _generate_ci_stage_config(jobs_data)
|
|
result["jobs_data"] = jobs_data
|
|
result["docker_data"] = docker_data
|
|
### CONFIGURE action: end
|
|
|
|
### PRE action: start
|
|
elif args.pre:
|
|
assert indata, "Run config must be provided via --infile"
|
|
_pre_action(s3, indata, pr_info)
|
|
|
|
### RUN action: start
|
|
elif args.run:
|
|
assert indata
|
|
ci_options = CiOptions.create_from_run_config(indata)
|
|
check_name = args.job_name
|
|
check_name_with_group = _get_ext_check_name(check_name)
|
|
print(
|
|
f"Check if rerun for name: [{check_name}], extended name [{check_name_with_group}]"
|
|
)
|
|
previous_status = None
|
|
if CI_CONFIG.is_build_job(check_name):
|
|
# this is a build job - check if build report is present
|
|
build_result = (
|
|
BuildResult.load_any(check_name, pr_info.number, pr_info.head_ref)
|
|
if not ci_options.no_ci_cache
|
|
else None
|
|
)
|
|
if build_result:
|
|
if build_result.status == SUCCESS:
|
|
previous_status = build_result.status
|
|
else:
|
|
# FIXME: Consider reusing failures for build jobs.
|
|
# Just remove this if/else - that makes build job starting and failing immediately
|
|
print(
|
|
"Build report found but status is unsuccessful - will try to rerun"
|
|
)
|
|
print("::group::Build Report")
|
|
print(build_result.as_json())
|
|
print("::endgroup::")
|
|
else:
|
|
# this is a test job - check if GH commit status or cache record is present
|
|
commit = get_commit(
|
|
GitHub(get_best_robot_token(), per_page=100), pr_info.sha
|
|
)
|
|
|
|
# rerun helper check
|
|
# FIXME: remove rerun_helper check and rely on ci cache only
|
|
if check_name not in (
|
|
# we might want to rerun reports' jobs - disable rerun check for them
|
|
JobNames.BUILD_CHECK,
|
|
JobNames.BUILD_CHECK_SPECIAL,
|
|
):
|
|
rerun_helper = RerunHelper(commit, check_name_with_group)
|
|
if rerun_helper.is_already_finished_by_status():
|
|
status = rerun_helper.get_finished_status()
|
|
assert status
|
|
previous_status = status.state
|
|
print("::group::Commit Status")
|
|
print(status)
|
|
print("::endgroup::")
|
|
|
|
# ci cache check
|
|
if not previous_status and not ci_options.no_ci_cache:
|
|
ci_cache = CiCache(s3, indata["jobs_data"]["digests"]).update()
|
|
job_config = CI_CONFIG.get_job_config(check_name)
|
|
if ci_cache.is_successful(
|
|
check_name,
|
|
args.batch,
|
|
job_config.num_batches,
|
|
job_config.required_on_release_branch,
|
|
):
|
|
job_status = ci_cache.get_successful(
|
|
check_name, args.batch, job_config.num_batches
|
|
)
|
|
assert job_status, "BUG"
|
|
_create_gh_status(
|
|
commit,
|
|
check_name,
|
|
args.batch,
|
|
job_config.num_batches,
|
|
job_status,
|
|
)
|
|
previous_status = job_status.status
|
|
GHActions.print_in_group("Commit Status Data", job_status)
|
|
|
|
if previous_status and not args.force:
|
|
print(
|
|
f"Commit status or Build Report is already present - job will be skipped with status: [{previous_status}]"
|
|
)
|
|
if previous_status == SUCCESS:
|
|
exit_code = 0
|
|
else:
|
|
exit_code = 1
|
|
else:
|
|
exit_code = _run_test(check_name, args.run_command)
|
|
### RUN action: end
|
|
|
|
### POST action: start
|
|
elif args.post:
|
|
job_report = JobReport.load() if JobReport.exist() else None
|
|
if job_report:
|
|
ch_helper = ClickHouseHelper()
|
|
check_url = ""
|
|
|
|
if CI_CONFIG.is_build_job(args.job_name):
|
|
assert (
|
|
indata
|
|
), f"--infile with config must be provided for POST action of a build type job [{args.job_name}]"
|
|
build_name = args.job_name
|
|
s3_path_prefix = "/".join(
|
|
(
|
|
get_release_or_pr(pr_info, get_version_from_repo())[0],
|
|
pr_info.sha,
|
|
build_name,
|
|
)
|
|
)
|
|
log_url = _upload_build_artifacts(
|
|
pr_info,
|
|
build_name,
|
|
ci_cache=CiCache(s3, indata["jobs_data"]["digests"]),
|
|
job_report=job_report,
|
|
s3=s3,
|
|
s3_destination=s3_path_prefix,
|
|
)
|
|
_upload_build_profile_data(
|
|
pr_info, build_name, job_report, git_runner, ch_helper
|
|
)
|
|
check_url = log_url
|
|
else:
|
|
# test job
|
|
gh = GitHub(get_best_robot_token(), per_page=100)
|
|
additional_urls = []
|
|
s3_path_prefix = "/".join(
|
|
(
|
|
get_release_or_pr(pr_info, get_version_from_repo())[0],
|
|
pr_info.sha,
|
|
normalize_string(
|
|
job_report.check_name or _get_ext_check_name(args.job_name)
|
|
),
|
|
)
|
|
)
|
|
if job_report.build_dir_for_upload:
|
|
additional_urls = s3.upload_build_directory_to_s3(
|
|
Path(job_report.build_dir_for_upload),
|
|
s3_path_prefix,
|
|
keep_dirs_in_s3_path=False,
|
|
upload_symlinks=False,
|
|
)
|
|
if job_report.test_results or job_report.additional_files:
|
|
check_url = upload_result_helper.upload_results(
|
|
s3,
|
|
pr_info.number,
|
|
pr_info.sha,
|
|
job_report.test_results,
|
|
job_report.additional_files,
|
|
job_report.check_name or _get_ext_check_name(args.job_name),
|
|
additional_urls=additional_urls or None,
|
|
)
|
|
commit = get_commit(gh, pr_info.sha)
|
|
post_commit_status(
|
|
commit,
|
|
job_report.status,
|
|
check_url,
|
|
format_description(job_report.description),
|
|
job_report.check_name or _get_ext_check_name(args.job_name),
|
|
pr_info,
|
|
dump_to_file=True,
|
|
)
|
|
|
|
print(f"Job report url: [{check_url}]")
|
|
prepared_events = prepare_tests_results_for_clickhouse(
|
|
pr_info,
|
|
job_report.test_results,
|
|
job_report.status,
|
|
job_report.duration,
|
|
job_report.start_time,
|
|
check_url or "",
|
|
job_report.check_name or _get_ext_check_name(args.job_name),
|
|
)
|
|
ch_helper.insert_events_into(
|
|
db="default", table="checks", events=prepared_events
|
|
)
|
|
|
|
if "DockerServerImage" in args.job_name and indata is not None:
|
|
_add_build_to_version_history(
|
|
pr_info,
|
|
job_report,
|
|
indata["version"],
|
|
indata["build"],
|
|
ch_helper,
|
|
)
|
|
else:
|
|
# no job report
|
|
print(f"No job report for {[args.job_name]} - do nothing")
|
|
### POST action: end
|
|
|
|
### MARK SUCCESS action: start
|
|
elif args.mark_success:
|
|
assert indata, "Run config must be provided via --infile"
|
|
_mark_success_action(s3, indata, pr_info, args.job_name, args.batch)
|
|
|
|
### UPDATE GH STATUSES action: start
|
|
elif args.update_gh_statuses:
|
|
assert indata, "Run config must be provided via --infile"
|
|
_update_gh_statuses_action(indata=indata, s3=s3)
|
|
|
|
### print results
|
|
_print_results(result, args.outfile, args.pretty)
|
|
|
|
return exit_code
|
|
|
|
|
|
if __name__ == "__main__":
|
|
sys.exit(main())
|