view toolfactory/galaxy-tool-test @ 100:c749364c2283 draft

Uploaded
author fubar
date Mon, 23 Nov 2020 03:12:37 +0000
parents d4d88d393285
children
line wrap: on
line source

#!/usr/bin/env python

import argparse
import datetime as dt
import json
import logging
import os
import sys
import tempfile
from collections import namedtuple
from concurrent.futures import thread, ThreadPoolExecutor

import yaml

from galaxy.tool_util.verify.interactor import (
    DictClientTestConfig,
    GalaxyInteractorApi,
    verify_tool,
)

DESCRIPTION = """Script to quickly run a tool test against a running Galaxy instance."""
DEFAULT_SUITE_NAME = "Galaxy Tool Tests"
ALL_TESTS = -1
ALL_TOOLS = "*"
ALL_VERSION = "*"
LATEST_VERSION = None


TestReference = namedtuple("TestReference", ["tool_id", "tool_version", "test_index"])
TestException = namedtuple("TestException", ["tool_id", "exception", "was_recorded"])


class Results:

    def __init__(self, default_suitename, test_json, append=False):
        self.test_json = test_json or "-"
        test_results = []
        test_exceptions = []
        suitename = default_suitename
        if append:
            assert test_json != "-"
            with open(test_json) as f:
                previous_results = json.load(f)
                test_results = previous_results["tests"]
                if "suitename" in previous_results:
                    suitename = previous_results["suitename"]
        self.test_results = test_results
        self.test_exceptions = test_exceptions
        self.suitename = suitename

    def register_result(self, result):
        self.test_results.append(result)

    def register_exception(self, test_exception):
        self.test_exceptions.append(test_exception)

    def already_successful(self, test_reference):
        test_id = _test_id_for_reference(test_reference)
        for test_result in self.test_results:
            if test_result.get('id') != test_id:
                continue

            has_data = test_result.get('has_data', False)
            if has_data:
                test_data = test_result.get("data", {})
                if 'status' in test_data and test_data['status'] == 'success':
                    return True

        return False

    def write(self):
        tests = sorted(self.test_results, key=lambda el: el['id'])
        n_passed, n_failures, n_skips = 0, 0, 0
        n_errors = len([e for e in self.test_exceptions if not e.was_recorded])
        for test in tests:
            has_data = test.get('has_data', False)
            if has_data:
                test_data = test.get("data", {})
                if 'status' not in test_data:
                    raise Exception(f"Test result data {test_data} doesn't contain a status key.")
                status = test_data['status']
                if status == "success":
                    n_passed += 1
                elif status == "error":
                    n_errors += 1
                elif status == "skip":
                    n_skips += 1
                elif status == "failure":
                    n_failures += 1
        report_obj = {
            'version': '0.1',
            'suitename': self.suitename,
            'results': {
                'total': n_passed + n_failures + n_skips + n_errors,
                'errors': n_errors,
                'failures': n_failures,
                'skips': n_skips,
            },
            'tests': tests,
        }
        if self.test_json == "-":
            print(json.dumps(report_obj))
        else:
            with open(self.test_json, "w") as f:
                json.dump(report_obj, f)

    def info_message(self):
        messages = []
        passed_tests = self._tests_with_status('success')
        messages.append("Passed tool tests ({}): {}".format(
            len(passed_tests),
            [t["id"] for t in passed_tests]
        ))
        failed_tests = self._tests_with_status('failure')
        messages.append("Failed tool tests ({}): {}".format(
            len(failed_tests),
            [t["id"] for t in failed_tests]
        ))
        skiped_tests = self._tests_with_status('skip')
        messages.append("Skipped tool tests ({}): {}".format(
            len(skiped_tests),
            [t["id"] for t in skiped_tests]
        ))
        errored_tests = self._tests_with_status('error')
        messages.append("Errored tool tests ({}): {}".format(
            len(errored_tests),
            [t["id"] for t in errored_tests]
        ))
        return "\n".join(messages)

    @property
    def success_count(self):
        self._tests_with_status('success')

    @property
    def skip_count(self):
        self._tests_with_status('skip')

    @property
    def error_count(self):
        return self._tests_with_status('error') + len(self.test_exceptions)

    @property
    def failure_count(self):
        return self._tests_with_status('failure')

    def _tests_with_status(self, status):
        return [t for t in self.test_results if t.get("data", {}).get("status") == status]


def test_tools(
    galaxy_interactor,
    test_references,
    results,
    log=None,
    parallel_tests=1,
    history_per_test_case=False,
    no_history_cleanup=False,
    retries=0,
    verify_kwds=None,
):
    """Run through tool tests and write report.

    Refactor this into Galaxy in 21.01.
    """
    verify_kwds = (verify_kwds or {}).copy()
    tool_test_start = dt.datetime.now()
    history_created = False
    if history_per_test_case:
        test_history = None
    else:
        history_created = True
        test_history = galaxy_interactor.new_history(history_name=f"History for {results.suitename}")
    verify_kwds.update({
        "no_history_cleanup": no_history_cleanup,
        "test_history": test_history,
    })
    with ThreadPoolExecutor(max_workers=parallel_tests) as executor:
        try:
            for test_reference in test_references:
                _test_tool(
                    executor=executor,
                    test_reference=test_reference,
                    results=results,
                    galaxy_interactor=galaxy_interactor,
                    log=log,
                    retries=retries,
                    verify_kwds=verify_kwds,
                )
        finally:
            # Always write report, even if test was cancelled.
            try:
                executor.shutdown(wait=True)
            except KeyboardInterrupt:
                executor._threads.clear()
                thread._threads_queues.clear()
            results.write()
            if log:
                log.info("Report written to '%s'", os.path.abspath(results.test_json))
                log.info(results.info_message())
                log.info("Total tool test time: {}".format(dt.datetime.now() - tool_test_start))
            if history_created and not no_history_cleanup:
                galaxy_interactor.delete_history(test_history)


def _test_id_for_reference(test_reference):
    tool_id = test_reference.tool_id
    tool_version = test_reference.tool_version
    test_index = test_reference.test_index

    if tool_version and tool_id.endswith("/" + tool_version):
        tool_id = tool_id[:-len("/" + tool_version)]

    label_base = tool_id
    if tool_version:
        label_base += "/" + str(tool_version)

    test_id = label_base + "-" + str(test_index)
    return test_id


def _test_tool(
    executor,
    test_reference,
    results,
    galaxy_interactor,
    log,
    retries,
    verify_kwds,
):
    tool_id = test_reference.tool_id
    tool_version = test_reference.tool_version
    test_index = test_reference.test_index
    # If given a tool_id with a version suffix, strip it off so we can treat tool_version
    # correctly at least in client_test_config.
    if tool_version and tool_id.endswith("/" + tool_version):
        tool_id = tool_id[:-len("/" + tool_version)]

    test_id = _test_id_for_reference(test_reference)

    def run_test():
        run_retries = retries
        job_data = None
        job_exception = None

        def register(job_data_):
            nonlocal job_data
            job_data = job_data_

        try:
            while run_retries >= 0:
                job_exception = None
                try:
                    if log:
                        log.info("Executing test '%s'", test_id)
                    verify_tool(
                        tool_id, galaxy_interactor, test_index=test_index, tool_version=tool_version,
                        register_job_data=register, **verify_kwds
                    )
                    if log:
                        log.info("Test '%s' passed", test_id)
                    break
                except Exception as e:
                    if log:
                        log.warning("Test '%s' failed", test_id, exc_info=True)

                    job_exception = e
                    run_retries -= 1
        finally:
            if job_data is not None:
                results.register_result({
                    "id": test_id,
                    "has_data": True,
                    "data": job_data,
                })
            if job_exception is not None:
                was_recorded = job_data is not None
                test_exception = TestException(tool_id, job_exception, was_recorded)
                results.register_exception(test_exception)

    executor.submit(run_test)


def build_case_references(
    galaxy_interactor,
    tool_id=ALL_TOOLS,
    tool_version=LATEST_VERSION,
    test_index=ALL_TESTS,
    page_size=0,
    page_number=0,
    check_against=None,
    log=None,
):
    test_references = []
    if tool_id == ALL_TOOLS:
        tests_summary = galaxy_interactor.get_tests_summary()
        for tool_id, tool_versions_dict in tests_summary.items():
            for tool_version, summary in tool_versions_dict.items():
                for test_index in range(summary["count"]):
                    test_reference = TestReference(tool_id, tool_version, test_index)
                    test_references.append(test_reference)
    else:
        assert tool_id
        tool_test_dicts = galaxy_interactor.get_tool_tests(tool_id, tool_version=tool_version) or {}
        for i, tool_test_dict in enumerate(tool_test_dicts):
            this_tool_version = tool_test_dict.get("tool_version", tool_version)
            this_test_index = i
            if test_index == ALL_TESTS or i == test_index:
                test_reference = TestReference(tool_id, this_tool_version, this_test_index)
                test_references.append(test_reference)

    if check_against:
        filtered_test_references = []
        for test_reference in test_references:
            if check_against.already_successful(test_reference):
                if log is not None:
                    log.debug(f"Found successful test for {test_reference}, skipping")
                continue
            filtered_test_references.append(test_reference)
        log.info(f"Skipping {len(test_references)-len(filtered_test_references)} out of {len(test_references)} tests.")
        test_references = filtered_test_references

    if page_size > 0:
        slice_start = page_size * page_number
        slice_end = page_size * (page_number + 1)
        test_references = test_references[slice_start:slice_end]

    return test_references


def main(argv=None):
    if argv is None:
        argv = sys.argv[1:]

    args = _arg_parser().parse_args(argv)
    log = setup_global_logger(__name__, verbose=args.verbose)
    client_test_config_path = args.client_test_config
    if client_test_config_path is not None:
        log.debug(f"Reading client config path {client_test_config_path}")
        with open(client_test_config_path) as f:
            client_test_config = yaml.full_load(f)
    else:
        client_test_config = {}

    def get_option(key):
        arg_val = getattr(args, key, None)
        if arg_val is None and key in client_test_config:
            val = client_test_config.get(key)
        else:
            val = arg_val
        return val

    output_json_path = get_option("output_json")
    galaxy_interactor_kwds = {
        "galaxy_url": get_option("galaxy_url"),
        "master_api_key": get_option("admin_key"),
        "api_key": get_option("key"),
        "keep_outputs_dir": args.output,
        "download_attempts": get_option("download_attempts"),
        "download_sleep": get_option("download_sleep"),
    }
    tool_id = args.tool_id
    tool_version = args.tool_version
    tools_client_test_config = DictClientTestConfig(client_test_config.get("tools"))
    verbose = args.verbose

    galaxy_interactor = GalaxyInteractorApi(**galaxy_interactor_kwds)
    results = Results(args.suite_name, output_json_path, append=args.append)
    check_against = None if not args.skip_successful else results
    test_references = build_case_references(
        galaxy_interactor,
        tool_id=tool_id,
        tool_version=tool_version,
        test_index=args.test_index,
        page_size=args.page_size,
        page_number=args.page_number,
        check_against=check_against,
        log=log,
    )
    log.debug(f"Built {len(test_references)} test references to executed.")
    verify_kwds = dict(
        client_test_config=tools_client_test_config,
        force_path_paste=args.force_path_paste,
        skip_with_reference_data=not args.with_reference_data,
        quiet=not verbose,
    )
    test_tools(
        galaxy_interactor,
        test_references,
        results,
        log=log,
        parallel_tests=args.parallel_tests,
        history_per_test_case=args.history_per_test_case,
        no_history_cleanup=args.no_history_cleanup,
        verify_kwds=verify_kwds,
    )
    exceptions = results.test_exceptions
    if exceptions:
        exception = exceptions[0]
        if hasattr(exception, "exception"):
            exception = exception.exception
        raise exception


def setup_global_logger(name, log_file=None, verbose=False):
    formatter = logging.Formatter('%(asctime)s %(levelname)-5s - %(message)s')
    console = logging.StreamHandler()
    console.setFormatter(formatter)

    logger = logging.getLogger(name)
    logger.setLevel(logging.DEBUG if verbose else logging.INFO)
    logger.addHandler(console)

    if not log_file:
        # delete = false is chosen here because it is always nice to have a log file
        # ready if you need to debug. Not having the "if only I had set a log file"
        # moment after the fact.
        temp = tempfile.NamedTemporaryFile(prefix="ephemeris_", delete=False)
        log_file = temp.name
    file_handler = logging.FileHandler(log_file)
    logger.addHandler(file_handler)
    logger.info(f"Storing log file in: {log_file}")
    return logger


def _arg_parser():
    parser = argparse.ArgumentParser(description=DESCRIPTION)
    parser.add_argument('-u', '--galaxy-url', default="http://localhost:8080", help='Galaxy URL')
    parser.add_argument('-k', '--key', default=None, help='Galaxy User API Key')
    parser.add_argument('-a', '--admin-key', default=None, help='Galaxy Admin API Key')
    parser.add_argument('--force_path_paste', default=False, action="store_true", help='This requires Galaxy-side config option "allow_path_paste" enabled. Allows for fetching test data locally. Only for admins.')
    parser.add_argument('-t', '--tool-id', default=ALL_TOOLS, help='Tool ID')
    parser.add_argument('--tool-version', default=None, help='Tool Version (if tool id supplied). Defaults to just latest version, use * to test all versions')
    parser.add_argument('-i', '--test-index', default=ALL_TESTS, type=int, help='Tool Test Index (starting at 0) - by default all tests will run.')
    parser.add_argument('-o', '--output', default=None, help='directory to dump outputs to')
    parser.add_argument('--append', default=False, action="store_true", help="Extend a test record json (created with --output-json) with additional tests.")
    parser.add_argument('--skip-successful', default=False, action="store_true", help="When used with --append, skip previously run successful tests.")
    parser.add_argument('-j', '--output-json', default=None, help='output metadata json')
    parser.add_argument('--verbose', default=False, action="store_true", help="Verbose logging.")
    parser.add_argument('-c', '--client-test-config', default=None, help="Test config YAML to help with client testing")
    parser.add_argument('--suite-name', default=DEFAULT_SUITE_NAME, help="Suite name for tool test output")
    parser.add_argument('--with-reference-data', dest="with_reference_data", default=False, action="store_true")
    parser.add_argument('--skip-with-reference-data', dest="with_reference_data", action="store_false", help="Skip tests the Galaxy server believes use data tables or loc files.")
    parser.add_argument('--history-per-suite', dest="history_per_test_case", default=False, action="store_false", help="Create new history per test suite (all tests in same history).")
    parser.add_argument('--history-per-test-case', dest="history_per_test_case", action="store_true", help="Create new history per test case.")
    parser.add_argument('--no-history-cleanup', default=False, action="store_true", help="Perserve histories created for testing.")
    parser.add_argument('--parallel-tests', default=1, type=int, help="Parallel tests.")
    parser.add_argument('--retries', default=0, type=int, help="Retry failed tests.")
    parser.add_argument('--page-size', default=0, type=int, help="If positive, use pagination and just run one 'page' to tool tests.")
    parser.add_argument('--page-number', default=0, type=int, help="If page size is used, run this 'page' of tests - starts with 0.")
    parser.add_argument('--download-attempts', default=1, type=int, help="Galaxy may return a transient 500 status code for download if test results are written but not yet accessible.")
    parser.add_argument('--download-sleep', default=1, type=int, help="If download attempts is greater than 1, the amount to sleep between download attempts.")
    return parser


if __name__ == "__main__":
    main()