frostfs-testcases/pytest_tests/testsuites/special/test_frostfs_logs.py

import os
import shutil
import time
from datetime import datetime

import allure
import pytest
from frostfs_testlib import reporter
from frostfs_testlib.hosting import Host
from frostfs_testlib.testing.cluster_test_base import Cluster
from frostfs_testlib.testing.parallel import parallel


def pytest_generate_tests(metafunc: pytest.Metafunc):
    metafunc.fixturenames.append("repo")
    metafunc.fixturenames.append("markers")
    metafunc.parametrize(
        "repo, markers",
        [("frostfs-testcases", metafunc.config.option.markexpr)],
    )


class TestLogs:
    @allure.title("Check logs from frostfs-testcases with marks '{request.config.option.markexpr}'")
    @pytest.mark.order(1000)
    def test_logs_after_session(
        self, temp_directory: str, cluster: Cluster, session_start_time: datetime, request: pytest.FixtureRequest
    ):
        """
        This test automatically added to any test run to check logs from cluster for critical errors.

        """

        end_time = datetime.utcnow()
        logs_dir = os.path.join(temp_directory, "logs")
        os.makedirs(logs_dir)
        # Using \b here because 'oom' and 'panic' can sometimes be found in OID or CID
        issues_regex = r"\bpanic\b|\boom\b|too many|insufficient funds|insufficient amount of gas|wallet passwd|secret \bkey\b|access \bkey\b|cannot assign requested address"
        exclude_filter = r"too many requests"

        time.sleep(2)

        futures = parallel(
            self._collect_logs_on_host,
            cluster.hosts,
            logs_dir,
            issues_regex,
            session_start_time,
            end_time,
            exclude_filter,
        )

        hosts_with_problems = [
            future.result() for future in futures if not future.exception() and future.result() is not None
        ]
        if hosts_with_problems:
            self._attach_logs(logs_dir)

        assert (
            not hosts_with_problems
        ), f"The following hosts contains contain critical errors in system logs: {', '.join(hosts_with_problems)}"

    def _collect_logs_on_host(
        self, host: Host, logs_dir: str, regex: str, since: datetime, until: datetime, exclude_filter: str
    ):
        with reporter.step(f"Get logs from {host.config.address}"):
            logs = host.get_filtered_logs(filter_regex=regex, since=since, until=until, exclude_filter=exclude_filter)

        if not logs:
            return None

        with open(os.path.join(logs_dir, f"{host.config.address}.log"), "w") as file:
            file.write(logs)

        return host.config.address

    def _attach_logs(self, logs_dir: str) -> None:
        # Zip all files and attach to Allure because it is more convenient to download a single
        # zip with all logs rather than mess with individual logs files per service or node
        logs_zip_file_path = shutil.make_archive(logs_dir, "zip", logs_dir)
        reporter.attach(logs_zip_file_path, "logs.zip")