コード例 #1
0
def main() -> None:
    parser = argparse.ArgumentParser(prog="monocle")
    parser.add_argument("--loglevel", help="logging level", default="INFO")
    parser.add_argument(
        "--elastic-timeout",
        help="Elasticsearch connection retry timeout",
        default=10,
        type=int,
    )
    parser.add_argument("--elastic-conn",
                        help="Elasticsearch connection info",
                        default="localhost:9200")
    parser.add_argument(
        "--use-ssl",
        help="Use https protocol for communication with Elasticsearch",
        action="store_true",
    )
    parser.add_argument(
        "--insecure",
        help="Skip SSL CA cert validation",
        action="store_false",
    )
    parser.add_argument(
        "--ssl_show_warn",
        help="Skip showing a SSL warning message if it is not signed "
        "by CA authority",
        action="store_false",
    )
    parser.add_argument(
        "--elastic-user",
        help="Username for Elasticsearch authorization",
    )
    parser.add_argument(
        "--elastic-password",
        help="Password for Elasticsearch authorization",
    )
    subparsers = parser.add_subparsers(title="Subcommands",
                                       description="valid subcommands",
                                       dest="command")

    parser_crawler = subparsers.add_parser("crawler",
                                           help="Threaded crawlers pool")
    parser_crawler.add_argument("--config",
                                help="Configuration file of the crawlers pool",
                                required=True)

    parser_dbmanage = subparsers.add_parser("dbmanage",
                                            help="Database manager")
    parser_dbmanage.add_argument("--config",
                                 help="Configuration file",
                                 required=False)
    parser_dbmanage.add_argument(
        "--delete-repository",
        help="Delete events related to a repository (regexp)",
    )
    parser_dbmanage.add_argument(
        "--delete-index",
        help="Delete the index",
        action="store_true",
    )
    parser_dbmanage.add_argument("--index",
                                 help="The Elastisearch index name",
                                 required=True)
    parser_dbmanage.add_argument(
        "--run-migrate",
        help="Run the migration process",
    )

    parser_dbmanage.add_argument(
        "--update-idents",
        help="Update identities",
        action="store_true",
    )

    parser_dbquery = subparsers.add_parser(
        "dbquery", help="Run an existsing query on stored events")
    parser_dbquery.add_argument("--index",
                                help="The Elastisearch index name",
                                required=True)
    parser_dbquery.add_argument("--name", help="The query name", required=True)
    parser_dbquery.add_argument(
        "--repository",
        help="Scope to events of repositories (regexp)",
        required=True)
    parser_dbquery.add_argument(
        "--target-branch",
        help="Scope to events of a target branches (regexp)")
    parser_dbquery.add_argument("--gte",
                                help="Scope to events created after date")
    parser_dbquery.add_argument("--lte",
                                help="Scope to events created before date")
    parser_dbquery.add_argument(
        "--on_cc_gte",
        help="Scope to events related to changes created after date")
    parser_dbquery.add_argument(
        "--on_cc_lte",
        help="Scope to events related to changes created before date")
    parser_dbquery.add_argument(
        "--ec-same-date",
        help="Scope to events related to changes created during the "
        "same date bondaries defined by gte/lte arguments",
        action="store_true",
    )
    parser_dbquery.add_argument(
        "--type", help="Scope to events types list (comma separated)")
    parser_dbquery.add_argument(
        "--files", help="Scope to changes containing this file regexp")
    parser_dbquery.add_argument(
        "--state",
        help="Scope to changes with state (comma separated)",
    )
    parser_dbquery.add_argument("--change-ids",
                                help="Scope to change ids (comma separated)")
    parser_dbquery.add_argument("--authors",
                                help="Scope to authors (comma separated)")
    parser_dbquery.add_argument(
        "--approvals",
        help="Scope to objects with approvals (comma separated)")
    parser_dbquery.add_argument(
        "--exclude-approvals", help="Approvals exclude list (comma separated)")
    parser_dbquery.add_argument("--size",
                                help="Return maximum of size results",
                                default=10)
    parser_dbquery.add_argument(
        "--from", help="Starting index of the elements to retrieve", default=0)
    parser_dbquery.add_argument("--exclude-authors",
                                help="Authors exclude list (comma separated)")
    parser_dbquery.add_argument(
        "--tests-included",
        help="Scope to changes containing tests",
        action="store_true",
    )
    parser_dbquery.add_argument(
        "--self-merged",
        help="Scope to changes merged by their authors",
        action="store_true",
    )
    parser_dbquery.add_argument(
        "--has-issue-tracker-links",
        help="Scope to changes containing an issue tracker link",
        choices=["generic", "github.com", "altassian.net"],
    )
    parser_dbquery.add_argument(
        "--task-priority",
        help="Scope to changes related to task priorities (comma separated)",
    )
    parser_dbquery.add_argument(
        "--task-severity",
        help="Scope to changes related to task severities (comma separated)",
    )
    parser_dbquery.add_argument(
        "--task-issue-type",
        help="Scope to changes related to task type (comma separated)",
    )

    parser_dbquery.add_argument(
        "--task-score",
        help="Scope to changes related to task score '<op>: <val>'",
    )

    args = parser.parse_args()

    logging.basicConfig(
        level=getattr(logging, args.loglevel.upper()),
        format="%(asctime)s - %(name)s - %(thread)d - %(threadName)s - " +
        "%(levelname)s - %(message)s",
    )
    log = logging.getLogger(__name__)

    if not args.command:
        parser.print_usage()
        sys.exit(1)

    if args.command == "crawler":
        realpath = os.path.expanduser(args.config)
        if not os.path.isfile(realpath):
            log.error("Unable to access config: %s" % realpath)
            sys.exit(1)
        configdata = yaml.safe_load(open(realpath).read())
        config.validate(configdata, config.schema)
        tpool: List[Union[Crawler, GroupCrawler]] = []
        group = {}
        app = None
        if os.getenv("APP_ID") and os.getenv("APP_KEY_PATH"):
            app = application.get_app(os.getenv("APP_ID"),
                                      os.getenv("APP_KEY_PATH"))
        for tenant in configdata["tenants"]:
            idents_config = config.get_idents_config(configdata,
                                                     tenant["index"])
            for crawler_item in tenant.get("crawler",
                                           {}).get("github_orgs", []):
                tg = pullrequest.TokenGetter(crawler_item["name"],
                                             crawler_item.get("token"), app)
                github_c_args = pullrequest.GithubCrawlerArgs(
                    command="github_crawler",
                    org=crawler_item["name"],
                    updated_since=crawler_item["updated_since"],
                    loop_delay=tenant["crawler"]["loop_delay"],
                    repository=crawler_item.get("repository"),
                    base_url=utils.strip_url(crawler_item["base_url"]),
                    token_getter=tg,
                    db=ELmonocleDB(
                        elastic_conn=args.elastic_conn,
                        index=tenant["index"],
                        timeout=args.elastic_timeout,
                        user=args.elastic_user,
                        password=args.elastic_password,
                        use_ssl=args.use_ssl,
                        verify_certs=args.insecure,
                        ssl_show_warn=args.ssl_show_warn,
                    ),
                    idents_config=idents_config,
                )
                gid = crawler_item.get("token")
                if not gid:
                    if app:
                        # No token, if we have a app then get the token from the app
                        gid = app.get_token(org=crawler_item["name"])
                    else:
                        log.info("Skip crawler because no token: %s" %
                                 github_c_args)
                        continue
                if gid not in group:
                    group[gid] = GroupCrawler()
                    tpool.append(group[gid])
                if github_c_args.repository:
                    repositories = [github_c_args.repository]
                else:
                    log.info("Discovering repositories in %s ..." %
                             github_c_args.org)
                    # No repository specified for that organization so
                    # try to discover all of them
                    rf = organization.RepositoriesFetcher(
                        graphql.GithubGraphQLQuery(token_getter=tg))
                    repos = rf.get(github_c_args.org)
                    repositories = [
                        repo["name"] for repo in repos
                        if not repo["isArchived"]
                    ]
                    log.info("Found %s repositories in %s ..." %
                             (len(repositories), github_c_args.org))
                for repository in repositories:
                    github_c_args.repository = repository
                    group[gid].add_crawler(Runner(github_c_args))
            for crawler_item in tenant.get("crawler",
                                           {}).get("gerrit_repositories", []):
                gerrit_c_args = review.GerritCrawlerArgs(
                    command="gerrit_crawler",
                    repository=crawler_item["name"],
                    updated_since=crawler_item["updated_since"],
                    loop_delay=tenant["crawler"]["loop_delay"],
                    base_url=utils.strip_url(crawler_item["base_url"]),
                    insecure=crawler_item.get("insecure", False),
                    login=crawler_item.get("login"),
                    password=crawler_item.get("password"),
                    db=ELmonocleDB(
                        elastic_conn=args.elastic_conn,
                        index=tenant["index"],
                        timeout=args.elastic_timeout,
                        user=args.elastic_user,
                        password=args.elastic_password,
                        use_ssl=args.use_ssl,
                        verify_certs=args.insecure,
                        ssl_show_warn=args.ssl_show_warn,
                    ),
                    prefix=crawler_item.get("prefix"),
                    idents_config=idents_config,
                )
                tpool.append(Crawler(gerrit_c_args))
        log.info("%d configured threads" % len(tpool))
        for cthread in tpool:
            cthread.start()

    if args.command == "dbmanage":

        if args.update_idents and not args.config:
            log.error("Please provide the --config option")
            sys.exit(1)
        if args.update_idents:
            idents_config = config.get_idents_config(
                yaml.safe_load(open(args.config)), args.index)
        else:
            idents_config = []
        db = ELmonocleDB(
            elastic_conn=args.elastic_conn,
            index=args.index,
            idents_config=idents_config,
            user=args.elastic_user,
            password=args.elastic_password,
            use_ssl=args.use_ssl,
            verify_certs=args.insecure,
            ssl_show_warn=args.ssl_show_warn,
        )
        if args.delete_repository:
            db.delete_repository(args.delete_repository)
        if args.delete_index:
            db.delete_index()
        if args.update_idents:
            db.update_idents()
        if args.run_migrate:
            try:
                migrate.run_migrate(args.run_migrate, args.elastic_conn,
                                    args.index)
            except migrate.NotAvailableException:
                log.error("Error: %s is not a valid migration process" %
                          args.run_migrate)

    if args.command == "dbquery":
        db = ELmonocleDB(
            elastic_conn=args.elastic_conn,
            index=args.index,
            user=args.elastic_user,
            password=args.elastic_password,
            use_ssl=args.use_ssl,
            verify_certs=args.insecure,
            ssl_show_warn=args.ssl_show_warn,
        )
        params = utils.set_params(args)
        try:
            ret = db.run_named_query(args.name, args.repository.lstrip("^"),
                                     params)
        except UnknownQueryException as err:
            log.error("Unable to run query: %s" % err)
            sys.exit(1)
        pprint(ret)
コード例 #2
0
def main():
    parser = argparse.ArgumentParser(prog='monocle')
    parser.add_argument('--loglevel', help='logging level', default='INFO')
    parser.add_argument(
        '--elastic-timeout',
        help='Elasticsearch connection retry timeout',
        default=10,
        type=int,
    )
    parser.add_argument(
        '--elastic-conn', help='Elasticsearch connection info', default='localhost:9200'
    )
    subparsers = parser.add_subparsers(
        title='Subcommands', description='valid subcommands', dest="command"
    )

    parser_crawler = subparsers.add_parser('crawler', help='Threaded crawlers pool')
    parser_crawler.add_argument(
        '--config', help='Configuration file of the crawlers pool', required=True
    )

    parser_dbmanage = subparsers.add_parser('dbmanage', help='Database manager')
    parser_dbmanage.add_argument(
        '--delete-repository', help='Delete events related to a repository (regexp)',
    )
    parser_dbmanage.add_argument(
        '--delete-index', help='Delete the index', action='store_true',
    )
    parser_dbmanage.add_argument(
        '--index', help='The Elastisearch index name', required=True
    )

    parser_dbquery = subparsers.add_parser(
        'dbquery', help='Run an existsing query on stored events'
    )
    parser_dbquery.add_argument(
        '--index', help='The Elastisearch index name', required=True
    )
    parser_dbquery.add_argument('--name', help='The query name', required=True)
    parser_dbquery.add_argument(
        '--repository', help='Scope to events of repositories (regexp)', required=True
    )
    parser_dbquery.add_argument(
        '--target-branch', help='Scope to events of a target branches (regexp)'
    )
    parser_dbquery.add_argument('--gte', help='Scope to events created after date')
    parser_dbquery.add_argument('--lte', help='Scope to events created before date')
    parser_dbquery.add_argument(
        '--on_cc_gte', help='Scope to events related to changes created after date'
    )
    parser_dbquery.add_argument(
        '--on_cc_lte', help='Scope to events related to changes created before date'
    )
    parser_dbquery.add_argument(
        '--ec-same-date',
        help='Scope to events related to changes created during the '
        'same date bondaries defined by gte/lte arguments',
        action='store_true',
    )
    parser_dbquery.add_argument(
        '--type', help='Scope to events types list (comma separated)'
    )
    parser_dbquery.add_argument(
        '--files', help='Scope to changes containing this file regexp'
    )
    parser_dbquery.add_argument(
        '--state',
        help='Scope to changes having this state',
        choices=['OPEN', 'CLOSED', 'MERGED'],
    )
    parser_dbquery.add_argument(
        '--change-ids', help='Scope to change ids (comma separated)'
    )
    parser_dbquery.add_argument('--authors', help='Scope to authors (comma separated)')
    parser_dbquery.add_argument(
        '--approvals', help='Scope to objects with approvals (comma separated)'
    )
    parser_dbquery.add_argument(
        '--exclude-approvals', help='Approvals exclude list (comma separated)'
    )
    parser_dbquery.add_argument(
        '--size', help='Return maximum of size results', default=10
    )
    parser_dbquery.add_argument(
        '--from', help='Starting index of the elements to retrieve', default=0
    )
    parser_dbquery.add_argument(
        '--exclude-authors', help='Authors exclude list (comma separated)'
    )
    parser_dbquery.add_argument(
        '--tests-included',
        help='Scope to changes containing tests',
        action='store_true',
    )
    parser_dbquery.add_argument(
        '--has-issue-tracker-links',
        help='Scope to changes containing an issue tracker link',
        choices=['generic', 'github.com', 'altassian.net'],
    )

    args = parser.parse_args()

    logging.basicConfig(
        level=getattr(logging, args.loglevel.upper()),
        format="%(asctime)s - %(name)s - %(thread)d - %(threadName)s - "
        + "%(levelname)s - %(message)s",
    )
    log = logging.getLogger(__name__)

    if not args.command:
        parser.print_usage()
        return 1

    if args.command == "crawler":
        realpath = os.path.expanduser(args.config)
        if not os.path.isfile(realpath):
            log.error('Unable to access config: %s' % realpath)
            sys.exit(1)
        configdata = yaml.safe_load(open(realpath).read())
        validate(instance=configdata, schema=config.schema)
        tpool = []
        group = {}
        app = None
        if os.getenv('APP_ID') and os.getenv('APP_KEY_PATH'):
            app = application.get_app(os.getenv('APP_ID'), os.getenv('APP_KEY_PATH'))
        for tenant in configdata['tenants']:
            for crawler_item in tenant['crawler'].get('github_orgs', []):
                tg = pullrequest.TokenGetter(
                    crawler_item['name'], crawler_item.get('token'), app
                )
                c_args = pullrequest.GithubCrawlerArgs(
                    command='github_crawler',
                    org=crawler_item['name'],
                    updated_since=crawler_item['updated_since'],
                    loop_delay=tenant['crawler']['loop_delay'],
                    repository=crawler_item.get('repository'),
                    base_url=crawler_item['base_url'],
                    token_getter=tg,
                    db=ELmonocleDB(
                        elastic_conn=args.elastic_conn,
                        index=tenant['index'],
                        timeout=args.elastic_timeout,
                    ),
                )
                gid = crawler_item.get('token')
                if not gid:
                    if app:
                        # No token, if we have a app then get the token from the app
                        gid = app.get_token(org=crawler_item['name'])
                    else:
                        log.info('Skip crawler because no token: %s' % c_args)
                        continue
                if gid not in group:
                    group[gid] = GroupCrawler()
                    tpool.append(group[gid])
                if c_args.repository:
                    repositories = [c_args.repository]
                else:
                    log.info('Discovering repositories in %s ...' % c_args.org)
                    # No repository specified for that organization so
                    # try to discover all of them
                    rf = organization.RepositoriesFetcher(
                        graphql.GithubGraphQLQuery(token_getter=tg)
                    )
                    repos = rf.get(c_args.org)
                    repositories = [
                        repo['name'] for repo in repos if not repo['isArchived']
                    ]
                    log.info(
                        'Found %s repositories in %s ...'
                        % (len(repositories), c_args.org)
                    )
                for repository in repositories:
                    c_args.repository = repository
                    group[gid].add_crawler(Runner(c_args))
            for crawler_item in tenant['crawler'].get('gerrit_repositories', []):
                c_args = review.GerritCrawlerArgs(
                    command='gerrit_crawler',
                    repository=crawler_item['name'],
                    updated_since=crawler_item['updated_since'],
                    loop_delay=tenant['crawler']['loop_delay'],
                    base_url=crawler_item['base_url'],
                    insecure=crawler_item.get('insecure', False),
                    login=crawler_item.get('login'),
                    password=crawler_item.get('password'),
                    db=ELmonocleDB(
                        elastic_conn=args.elastic_conn,
                        index=tenant['index'],
                        timeout=args.elastic_timeout,
                    ),
                )
                tpool.append(Crawler(c_args))
        log.info('%d configured threads' % len(tpool))
        for cthread in tpool:
            cthread.start()

    if args.command == "dbmanage":
        db = ELmonocleDB(elastic_conn=args.elastic_conn, index=args.index)
        if args.delete_repository:
            db.delete_repository(args.delete_repository)
        if args.delete_index:
            db.delete_index()

    if args.command == "dbquery":
        db = ELmonocleDB(elastic_conn=args.elastic_conn, index=args.index)
        params = utils.set_params(args)
        try:
            ret = db.run_named_query(args.name, args.repository.lstrip('^'), params)
        except UnknownQueryException as err:
            log.error('Unable to run query: %s' % err)
            sys.exit(1)
        pprint(ret)
コード例 #3
0
ファイル: main.py プロジェクト: mhuin/monocle
def main():
    parser = argparse.ArgumentParser(prog='monocle')
    parser.add_argument('--loglevel', help='logging level', default='INFO')
    parser.add_argument(
        '--elastic-timeout',
        help='Elasticsearch connection retry timeout',
        default=10,
        type=int,
    )
    parser.add_argument('--elastic-conn',
                        help='Elasticsearch connection info',
                        default='localhost:9200')
    subparsers = parser.add_subparsers(title='Subcommands',
                                       description='valid subcommands',
                                       dest="command")

    parser_crawler = subparsers.add_parser('crawler',
                                           help='Threaded crawlers pool')
    parser_crawler.add_argument('--config',
                                help='Configuration file of the crawlers pool',
                                required=True)

    parser_dbmanage = subparsers.add_parser('dbmanage',
                                            help='Database manager')
    parser_dbmanage.add_argument(
        '--delete-repository',
        help='Delete events related to a repository (regexp)',
        required=True,
    )
    parser_dbmanage.add_argument('--index',
                                 help='The Elastisearch index name',
                                 required=True)

    parser_dbquery = subparsers.add_parser(
        'dbquery', help='Run an existsing query on stored events')
    parser_dbquery.add_argument('--index',
                                help='The Elastisearch index name',
                                required=True)
    parser_dbquery.add_argument('--name', help='The query name', required=True)
    parser_dbquery.add_argument(
        '--repository',
        help='Scope to events of repositories (regexp)',
        required=True)
    parser_dbquery.add_argument(
        '--target-branch',
        help='Scope to events of a target branches (regexp)')
    parser_dbquery.add_argument('--gte',
                                help='Scope to events created after date')
    parser_dbquery.add_argument('--lte',
                                help='Scope to events created before date')
    parser_dbquery.add_argument(
        '--on_cc_gte',
        help='Scope to events related to changes created after date')
    parser_dbquery.add_argument(
        '--on_cc_lte',
        help='Scope to events related to changes created before date')
    parser_dbquery.add_argument(
        '--ec-same-date',
        help='Scope to events related to changes created during the '
        'same date bondaries defined by gte/lte arguments',
        action='store_true',
    )
    parser_dbquery.add_argument(
        '--type', help='Scope to events types list (comma separated)')
    parser_dbquery.add_argument(
        '--files', help='Scope to changes containing this file regexp')
    parser_dbquery.add_argument(
        '--state',
        help='Scope to changes having this state',
        choices=['OPEN', 'CLOSED', 'MERGED'],
    )
    parser_dbquery.add_argument('--change-ids',
                                help='Scope to change ids (comma separated)')
    parser_dbquery.add_argument('--authors',
                                help='Scope to authors (comma separated)')
    parser_dbquery.add_argument('--approval',
                                help='Scope to events with approval')
    parser_dbquery.add_argument('--size',
                                help='Return maximum of size results',
                                default=10)
    parser_dbquery.add_argument(
        '--from', help='Starting index of the elements to retrieve', default=0)
    parser_dbquery.add_argument('--exclude-authors',
                                help='Authors exclude list (comma separated)')
    parser_dbquery.add_argument(
        '--tests-included',
        help='Scope to changes containing tests',
        action='store_true',
    )
    parser_dbquery.add_argument(
        '--has-issue-tracker-links',
        help='Scope to changes containing an issue tracker link',
        choices=['generic', 'github.com', 'altassian.net'],
    )

    args = parser.parse_args()

    logging.basicConfig(
        level=getattr(logging, args.loglevel.upper()),
        format="%(asctime)s - %(name)s - %(threadName)s - " +
        "%(levelname)s - %(message)s",
    )
    log = logging.getLogger(__name__)

    if not args.command:
        parser.print_usage()
        return 1

    if args.command == "crawler":
        realpath = os.path.expanduser(args.config)
        if not os.path.isfile(realpath):
            log.error('Unable to access config: %s' % realpath)
            sys.exit(1)
        configdata = yaml.safe_load(open(realpath).read())
        validate(instance=configdata, schema=config.schema)
        tpool = []
        group = {}
        for tenant in configdata['tenants']:
            for crawler_item in tenant['crawler'].get('github_orgs', []):
                c_args = pullrequest.GithubCrawlerArgs(
                    command='github_crawler',
                    index=tenant['index'],
                    org=crawler_item['name'],
                    updated_since=crawler_item['updated_since'],
                    loop_delay=tenant['crawler']['loop_delay'],
                    token=crawler_item['token'],
                    repository=crawler_item.get('repository'),
                    base_url=crawler_item['base_url'],
                )
                log.info('args=%s' % c_args)
                if crawler_item['token'] not in group:
                    group[crawler_item['token']] = GroupCrawler()
                    tpool.append(group[crawler_item['token']])
                group[crawler_item['token']].add_crawler(
                    Runner(
                        c_args,
                        elastic_conn=args.elastic_conn,
                        elastic_timeout=args.elastic_timeout,
                    ))
            for crawler_item in tenant['crawler'].get('gerrit_repositories',
                                                      []):
                c_args = review.GerritCrawlerArgs(
                    command='gerrit_crawler',
                    index=tenant['index'],
                    repository=crawler_item['name'],
                    updated_since=crawler_item['updated_since'],
                    loop_delay=tenant['crawler']['loop_delay'],
                    base_url=crawler_item['base_url'],
                )
                tpool.append(
                    Crawler(
                        c_args,
                        elastic_conn=args.elastic_conn,
                        elastic_timeout=args.elastic_timeout,
                    ))
        log.info('%d configured threads' % len(tpool))
        for cthread in tpool:
            cthread.start()

    if args.command == "dbmanage":
        if args.delete_repository:
            db = ELmonocleDB(elastic_conn=args.elastic_conn, index=args.index)
            db.delete_repository(args.delete_repository)

    if args.command == "dbquery":
        db = ELmonocleDB(elastic_conn=args.elastic_conn, index=args.index)
        params = utils.set_params(args)
        try:
            ret = db.run_named_query(args.name, args.repository.lstrip('^'),
                                     params)
        except UnknownQueryException as err:
            log.error('Unable to run query: %s' % err)
            sys.exit(1)
        pprint(ret)