Files
scylladb/test/pylib/runner.py
Patryk Jędrzejczak 46b7170347 Merge 'test/pylib: centralize timeout scaling and propagate build_mode in LWT helpers' from Alex Dathskovsky
This series improves timeout handling consistency across the test framework and makes build-mode effects explicit in LWT tests. (starting with LWT test that got flaky)

1. Centralize timeout scaling
Introduce scale_timeout(timeout) fixture in runner.py to provide a single, consistent mechanism for scaling test timeouts based on build mode.
Previously, timeout adjustments were done in an ad-hoc manner across different helpers and tests. Centralizing the logic:
Ensures consistent behavior across the test suite
Simplifies maintenance and reasoning about timeout behavior
Reduces duplication and per-test scaling logic
This becomes increasingly important as tests run on heterogeneous hardware configurations, where different build modes (especially debug) can significantly impact execution time.

2. Make scale_timeout explicit in LWT helpers
Propagate scale_timeout explicitly through BaseLWTTester and Worker, validating it at construction time instead of relying on implicit pytest fixture injection inside helper classes.
Additionally:
Update wait_for_phase_ops() and wait_for_tablet_count() to use scale_timeout_by_mode() for consistent polling behavior across modes
Update all LWT test call sites to pass build_mode explicitly
Increase default timeout values, as the previous defaults were too short and prone to flakiness, particularly under slower configurations such as debug builds

Overall, this series improves determinism, reduces flakiness, and makes the interaction between build mode and test timing explicit and maintainable.

backport: not required just an enhansment for test.py infra

Closes scylladb/scylladb#28840

* https://github.com/scylladb/scylladb:
  test/auth_cluster: align service-level timeout expectations with scaled config
  test/lwt: propagate scale_timeout through LWT helpers; scale resize waits Pass scale_timeout explicitly through BaseLWTTester and Worker, validating it at construction time instead of relying on implicit pytest fixture injection inside helper classes. Update wait_for_phase_ops() and wait_for_tablet_count() to use scale_timeout_by_mode() so polling behavior remains consistent across build modes. Adjust LWT test call sites to pass scale_timeout explicitly. Increase default timeout values, as the previous defaults were too short and prone to flakiness under slower configurations (notably debug/dev builds).
  test/pylib: introduce scale_timeout fixture helper
2026-03-09 10:28:19 +01:00

489 lines
21 KiB
Python

#
# Copyright (C) 2025-present ScyllaDB
#
# SPDX-License-Identifier: LicenseRef-ScyllaDB-Source-Available-1.0
#
from __future__ import annotations
import asyncio
import logging
import os
import pathlib
import platform
import random
import sys
from argparse import BooleanOptionalAction
from collections import defaultdict
from itertools import chain, count, product
from functools import cache, cached_property
from pathlib import Path
from random import randint
from typing import TYPE_CHECKING, Callable
import pytest
import xdist
import yaml
from _pytest.junitxml import xml_key
from test import ALL_MODES, DEBUG_MODES, TEST_RUNNER, TOP_SRC_DIR, TESTPY_PREPARED_ENVIRONMENT, HOST_ID
from test.pylib.scylla_cluster import merge_cmdline_options
from test.pylib.suite.base import (
SUITE_CONFIG_FILENAME,
PYTEST_TESTS_LOGS_FOLDER,
TestSuite,
get_testpy_test,
prepare_environment,
init_testsuite_globals,
)
from test.pylib.util import get_modes_to_run, scale_timeout_by_mode
if TYPE_CHECKING:
from collections.abc import Generator
import _pytest.nodes
import _pytest.scope
from test.pylib.suite.base import Test
TEST_CONFIG_FILENAME = "test_config.yaml"
PYTEST_LOG_FOLDER = "pytest_log"
REPEATING_FILES = pytest.StashKey[set[pathlib.Path]]()
BUILD_MODE = pytest.StashKey[str]()
RUN_ID = pytest.StashKey[int]()
PYTEST_LOG_FILE = pytest.StashKey[str]()
EXIT_MAXFAIL_REACHED = 11
logger = logging.getLogger(__name__)
# Store pytest config globally so we can access it in hooks that only receive report
_pytest_config: pytest.Config | None = None
def pytest_addoption(parser: pytest.Parser) -> None:
parser.addoption('--mode', choices=ALL_MODES, action="append", dest="modes",
help="Run only tests for given build mode(s)")
parser.addoption('--tmpdir', action='store', default=str(TOP_SRC_DIR / 'testlog'),
help='Path to temporary test data and log files. The data is further segregated per build mode.')
parser.addoption('--run_id', action='store', default=None, help='Run id for the test run')
parser.addoption('--byte-limit', action="store", default=randint(0, 2000), type=int,
help="Specific byte limit for failure injection (random by default)")
parser.addoption("--gather-metrics", action=BooleanOptionalAction, default=False,
help='Switch on gathering cgroup metrics')
parser.addoption('--random-seed', action="store",
help="Random number generator seed to be used by boost tests")
# Following option is to use with bare pytest command.
#
# For compatibility with reasons need to run bare pytest with --test-py-init option
# to run a test.py-compatible pytest session.
#
# TODO: remove this when we'll completely switch to bare pytest runner.
parser.addoption('--test-py-init', action='store_true', default=False,
help='Run pytest session in test.py-compatible mode. I.e., start all required services, etc.')
# Options for compatibility with test.py
parser.addoption('--save-log-on-success', default=False,
dest="save_log_on_success", action="store_true",
help="Save test log output on success and skip cleanup before the run.")
parser.addoption('--coverage', action='store_true', default=False,
help="When running code instrumented with coverage support"
"Will route the profiles to `tmpdir`/mode/coverage/`suite` and post process them in order to generate "
"lcov file per suite, lcov file per mode, and an lcov file for the entire run, "
"The lcov files can eventually be used for generating coverage reports")
parser.addoption("--coverage-mode", action='append', type=str, dest="coverage_modes",
help="Collect and process coverage only for the modes specified. implies: --coverage, default: All built modes")
parser.addoption("--cluster-pool-size", type=int,
help="Set the pool_size for PythonTest and its descendants. Alternatively environment variable "
"CLUSTER_POOL_SIZE can be used to achieve the same")
parser.addoption("--extra-scylla-cmdline-options", default='',
help="Passing extra scylla cmdline options for all tests. Options should be space separated:"
" '--logger-log-level raft=trace --default-log-level error'")
parser.addoption('--x-log2-compaction-groups', action="store", default="0", type=int,
help="Controls number of compaction groups to be used by Scylla tests. Value of 3 implies 8 groups.")
parser.addoption('--repeat', action="store", default="1", type=int,
help="number of times to repeat test execution")
# Pass information about Scylla node from test.py to pytest.
parser.addoption("--scylla-log-filename",
help="Path to a log file of a ScyllaDB node (for suites with type: Python)")
parser.addoption('--exe-path', default=False,
dest="exe_path", action="store",
help="Path to the executable to run. Not working with `mode`")
parser.addoption('--exe-url', default=False,
dest="exe_url", action="store",
help="URL to download the relocatable executable. Not working with `mode`")
@pytest.fixture(autouse=True)
def print_scylla_log_filename(request: pytest.FixtureRequest) -> Generator[None]:
"""Print out a path to a ScyllaDB log.
This is a fixture for Python test suites, because they are using a single node clusters created inside test.py,
but it is handy to have this information printed to a pytest log.
"""
yield
if scylla_log_filename := request.config.getoption("--scylla-log-filename"):
logger.info("ScyllaDB log file: %s", scylla_log_filename)
def testpy_test_fixture_scope(fixture_name: str, config: pytest.Config) -> _pytest.scope._ScopeName:
"""Dynamic scope for fixtures which rely on a current test.py suite/test.
test.py runs tests file-by-file as separate pytest sessions, so, `session` scope is effectively close to be the
same as `module` (can be a difference in the order.) In case of running tests with bare pytest command, we
need to use `module` scope to maintain same behavior as test.py, since we run all tests in one pytest session.
"""
if getattr(config.option, "test_py_init", False):
return "module"
return "session"
testpy_test_fixture_scope.__test__ = False
@pytest.fixture(scope=testpy_test_fixture_scope, autouse=True)
def build_mode(request: pytest.FixtureRequest) -> str:
params_stash = get_params_stash(node=request.node)
if params_stash is None:
return request.config.build_modes[0]
return params_stash[BUILD_MODE]
@pytest.fixture(scope=testpy_test_fixture_scope)
def scale_timeout(build_mode: str) -> Callable[[int | float], int | float]:
def scale_timeout_inner(timeout: int | float) -> int | float:
return scale_timeout_by_mode(build_mode, timeout)
return scale_timeout_inner
@pytest.fixture(scope=testpy_test_fixture_scope)
async def testpy_test(request: pytest.FixtureRequest, build_mode: str) -> Test | None:
"""Create an instance of Test class for the current test.py test."""
if request.scope == "module":
return await get_testpy_test(path=request.path, options=request.config.option, mode=build_mode)
return None
@pytest.fixture(scope="function")
def scylla_binary(testpy_test) -> Path:
return testpy_test.suite.scylla_exe
def pytest_collection_modifyitems(items: list[pytest.Item]) -> None:
for item in items:
modify_pytest_item(item=item)
suites_order = defaultdict(count().__next__) # number suites in order of appearance
def sort_key(item: pytest.Item) -> tuple[int, bool]:
suite = item.stash[TEST_SUITE]
return suites_order[suite], suite and item.path.stem not in suite.cfg.get("run_first", [])
items.sort(key=sort_key)
def pytest_sessionstart(session: pytest.Session) -> None:
# test.py starts S3 mock and create/cleanup testlog by itself. Also, if we run with --collect-only option,
# we don't need this stuff.
if TEST_RUNNER != "pytest" or session.config.getoption("--collect-only"):
return
if not session.config.getoption("--test-py-init"):
return
# Check if this is an xdist worker
is_xdist_worker = xdist.is_xdist_worker(request_or_session=session)
# Always initialize globals in xdist workers (they run in separate processes)
# For the main process, only init if test.py hasn't done so already
if is_xdist_worker or TESTPY_PREPARED_ENVIRONMENT not in os.environ:
init_testsuite_globals()
TestSuite.artifacts.add_exit_artifact(None, TestSuite.hosts.cleanup)
# Run stuff just once for the main pytest process (not in xdist workers).
# Only prepare the environment if it hasn't been prepared by test.py
if not is_xdist_worker and TESTPY_PREPARED_ENVIRONMENT not in os.environ:
temp_dir = pathlib.Path(session.config.getoption("--tmpdir")).absolute()
prepare_environment(
tempdir_base=temp_dir,
modes=get_modes_to_run(session.config),
gather_metrics=session.config.getoption("--gather-metrics"),
save_log_on_success=session.config.getoption("--save-log-on-success"),
toxiproxy_byte_limit=session.config.getoption("--byte-limit"),
)
@pytest.hookimpl(tryfirst=True)
def pytest_runtest_logreport(report):
"""Add custom XML attributes to JUnit testcase elements.
This hook wraps the node_reporter's to_xml method to add custom attributes
when the XML element is created. This approach works with pytest-xdist because
it modifies the XML element directly when it's generated, rather than trying
to modify attrs before finalize() is called.
Attributes added:
- function_path: The function path of the test case (excluding parameters).
Uses tryfirst=True to run before LogXML's hook has created the node_reporter to avoid double recording.
"""
# Get the XML reporter
config = _pytest_config
if config is None:
return
xml = config.stash.get(xml_key, None)
if xml is None:
return
node_reporter = xml.node_reporter(report)
# Only wrap once to avoid multiple wrapping (check on the node_reporter object itself)
if not getattr(node_reporter, '__reporter_modified', False):
function_path = f'test/{report.nodeid.rsplit('.', 2)[0].rsplit('[', 1)[0]}'
# Wrap the to_xml method to add custom attributes to the element
original_to_xml = node_reporter.to_xml
def custom_to_xml():
"""Wrapper that adds custom attributes to the testcase element."""
element = original_to_xml()
element.set("function_path", function_path)
return element
node_reporter.to_xml = custom_to_xml
node_reporter.__reporter_modified = True
def pytest_sessionfinish(session: pytest.Session) -> None:
if not session.config.getoption("--test-py-init"):
return
is_xdist_worker = xdist.is_xdist_worker(request_or_session=session)
# If all tests passed, remove the log file to save space and avoid confusion with logs from failed runs.
# We check this at the end of the session to ensure that we have the complete log available for any failed tests.
if not (not is_xdist_worker and TESTPY_PREPARED_ENVIRONMENT in os.environ): # If this is not an xdist worker and test.py has prepared the environment, there is no separate xdist main process and no pytest_main.log file
if session.testsfailed == 0 and not session.config.getoption("--save-log-on-success"):
os.remove(_pytest_config.stash[PYTEST_LOG_FILE])
# Check if this is an xdist worker - workers should not clean up (only the main process should)
# Check if test.py has already prepared the environment, so it should clean up
if is_xdist_worker or TESTPY_PREPARED_ENVIRONMENT in os.environ:
return
# we only clean up when running with pure pytest
if getattr(TestSuite, "artifacts", None) is not None:
asyncio.run(TestSuite.artifacts.cleanup_before_exit())
# Modify exit code to reflect the number of failed tests for easier detection in CI.
maxfail = session.config.getoption("maxfail")
if 0 < maxfail <= session.testsfailed:
session.exitstatus = EXIT_MAXFAIL_REACHED
def pytest_configure(config: pytest.Config) -> None:
global _pytest_config
_pytest_config = config
if _pytest_config.getoption("--test-py-init"):
pytest_log_dir = pathlib.Path(_pytest_config.getoption("--tmpdir")).absolute() / PYTEST_LOG_FOLDER
worker_id = os.environ.get("PYTEST_XDIST_WORKER")
# If this is an xdist worker, set up logging to a separate file for this worker. Otherwise, set up logging for the main process.
if worker_id is not None:
_pytest_config.stash[PYTEST_LOG_FILE] = f"{pytest_log_dir}/pytest_{worker_id}_{HOST_ID}.log"
logging.basicConfig(
format=config.getini("log_file_format"),
filename=_pytest_config.stash[PYTEST_LOG_FILE],
level=config.getini("log_file_level"),
)
else:
# For the main process, we want to clean up old logs before the run, so we create the log directory and remove any existing log files.
pytest_log_dir.mkdir(parents=True, exist_ok=True)
if not _pytest_config.getoption("--save-log-on-success"):
for file in pytest_log_dir.glob("*"):
file.unlink()
_pytest_config.stash[PYTEST_LOG_FILE] = f"{pytest_log_dir}/pytest_main_{HOST_ID}.log"
logging.basicConfig(
format=config.getini("log_file_format"),
filename=_pytest_config.stash[PYTEST_LOG_FILE],
level=config.getini("log_file_level"),
)
if config.getoption("--exe-url") and config.getoption("--exe-path"):
raise RuntimeError("Can't use --exe-url and exe-path simultaneously.")
if config.getoption("--exe-path") or config.getoption("--exe-url"):
if config.getoption("--mode"):
raise RuntimeError("Can't use --mode with --exe-path or --exe-url.")
config.option.modes = ["custom_exe"]
os.environ["TOPOLOGY_RANDOM_FAILURES_TEST_SHUFFLE_SEED"] = os.environ.get("TOPOLOGY_RANDOM_FAILURES_TEST_SHUFFLE_SEED", str(random.randint(0, sys.maxsize)))
config.build_modes = get_modes_to_run(config)
repeat = int(config.getoption("--repeat"))
if testpy_run_id := config.getoption("--run_id"):
if repeat != 1:
raise RuntimeError("Can't use --run_id and --repeat simultaneously.")
config.run_ids = (testpy_run_id,)
else:
config.run_ids = tuple(range(1, repeat + 1))
@pytest.hookimpl(wrapper=True)
def pytest_collect_file(file_path: pathlib.Path,
parent: pytest.Collector) -> Generator[None, list[pytest.Collector], list[pytest.Collector]]:
collectors = yield
if len(collectors) == 1 and file_path not in parent.stash.setdefault(REPEATING_FILES, set()):
parent.stash[REPEATING_FILES].add(file_path)
build_modes = parent.config.build_modes
if suite_config := TestSuiteConfig.from_pytest_node(node=collectors[0]):
build_modes = (
mode for mode in build_modes
if not suite_config.is_test_disabled(build_mode=mode, path=file_path)
)
repeats = list(product(build_modes, parent.config.run_ids))
if not repeats:
return []
ihook = parent.ihook
collectors = list(chain(collectors, chain.from_iterable(
ihook.pytest_collect_file(file_path=file_path, parent=parent) for _ in range(1, len(repeats))
)))
for (build_mode, run_id), collector in zip(repeats, collectors, strict=True):
collector.stash[BUILD_MODE] = build_mode
collector.stash[RUN_ID] = run_id
collector.stash[TEST_SUITE] = suite_config
parent.stash[REPEATING_FILES].remove(file_path)
return collectors
@pytest.hookimpl(tryfirst=True, hookwrapper=True)
def pytest_runtest_makereport(item, call):
# This hook is used to capture test failures and save their details to a file in the pytest_tests_logs directory.
# We use tryfirst=True to ensure that this hook runs before any other hooks that might modify the report,
# and we use hookwrapper=True to allow us to access the report after it has been generated by other hooks.
outcome = yield
if _pytest_config.getoption("--test-py-init"):
rep = outcome.get_result()
# we only look at actual failing test calls, not setup/teardown
pytest_tests_logs = pathlib.Path(_pytest_config.getoption("--tmpdir")).absolute() / PYTEST_TESTS_LOGS_FOLDER
if rep.failed or _pytest_config.getoption("--save-log-on-success"):
mode = "a" if os.path.exists(pytest_tests_logs) else "w"
with open(pytest_tests_logs/ f"{item._nodeid.replace("::", "-").replace("/", "-")}-{rep.when}-{HOST_ID}.log",mode) as f:
f.write(rep.longreprtext + "\n")
for section in rep.sections:
f.write(section[0] + "\n")
f.write(section[1] + "\n")
class TestSuiteConfig:
def __init__(self, config_file: pathlib.Path):
self.path = config_file.parent
self.cfg = yaml.safe_load(config_file.read_text(encoding="utf-8"))
@cached_property
def name(self) -> str:
return self.path.name
@cached_property
def _run_in_specific_mode(self) -> set[str]:
return set(chain.from_iterable(self.cfg.get(f"run_in_{build_mode}", []) for build_mode in ALL_MODES))
@cache
def disabled_tests(self, build_mode: str) -> set[str]:
result = set(self.cfg.get("disable", []))
result.update(self.cfg.get(f"skip_in_{build_mode}", []))
if build_mode in DEBUG_MODES:
result.update(self.cfg.get("skip_in_debug_modes", []))
run_in_this_mode = set(self.cfg.get(f"run_in_{build_mode}", []))
result.update(self._run_in_specific_mode - run_in_this_mode)
return result
def is_test_disabled(self, build_mode: str, path: pathlib.Path) -> bool:
return str(path.relative_to(self.path).with_suffix("")) in self.disabled_tests(build_mode=build_mode)
@classmethod
def from_pytest_node(cls, node: _pytest.nodes.Node) -> TestSuiteConfig | None:
for config_file in (node.path / SUITE_CONFIG_FILENAME, node.path / TEST_CONFIG_FILENAME,):
if config_file.is_file():
suite = cls(config_file=config_file)
break
else:
if node.parent is None:
return None
suite = node.parent.stash.get(TEST_SUITE, None)
if suite is None:
suite = cls.from_pytest_node(node=node.parent)
if suite:
extra_opts = node.config.getoption("--extra-scylla-cmdline-options")
if extra_opts:
extra_cmd = suite.cfg.get('extra_scylla_cmdline_options', [])
extra_cmd = merge_cmdline_options(extra_cmd, extra_opts.split())
suite.cfg['extra_scylla_cmdline_options'] = extra_cmd
node.stash[TEST_SUITE] = suite
return suite
TEST_SUITE = pytest.StashKey[TestSuiteConfig | None]()
_STASH_KEYS_TO_COPY = BUILD_MODE, RUN_ID, TEST_SUITE
def get_params_stash(node: _pytest.nodes.Node) -> pytest.Stash | None:
parent = node.getparent(cls=pytest.File)
if parent is None:
return None
return parent.stash
def modify_pytest_item(item: pytest.Item) -> None:
params_stash = get_params_stash(node=item)
for key in _STASH_KEYS_TO_COPY:
item.stash[key] = params_stash[key]
suffix = f".{item.stash[BUILD_MODE]}.{item.stash[RUN_ID]}"
item._nodeid = f"{item._nodeid}{suffix}"
item.name = f"{item.name}{suffix}"
skip_marks = [
mark for mark in item.iter_markers("skip_mode")
if mark.name == "skip_mode"
]
for mark in skip_marks:
def __skip_test(mode, reason, platform_key=None):
modes = [mode] if isinstance(mode, str) else mode
for mode in modes:
if mode == item.stash[BUILD_MODE]:
if platform_key is None or platform_key in platform.platform():
item.add_marker(pytest.mark.skip(reason=reason))
try:
__skip_test(*mark.args, **mark.kwargs)
except TypeError as e:
raise TypeError(f"Failed to process skip_mode mark, {mark} for test {item}, error {e}")
if (any(mark.name == "xfail" for mark in item.iter_markers("xfail"))
and not any(mark.name == "nightly" for mark in item.iter_markers("nightly"))):
item.add_marker(pytest.mark.nightly)
if (any(mark.name in ("perf", "manual", "unstable") for mark in item.iter_markers())
and not any(mark.name == "non_gating" for mark in item.iter_markers("non_gating"))):
item.add_marker(pytest.mark.non_gating)