diff --git a/swh/lister/core/lister_base.py b/swh/lister/core/lister_base.py index a742453..22a35f8 100644 --- a/swh/lister/core/lister_base.py +++ b/swh/lister/core/lister_base.py @@ -1,549 +1,545 @@ # Copyright (C) 2015-2018 the Software Heritage developers # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import abc import datetime import gzip import json import logging import os import re import time from sqlalchemy import create_engine, func from sqlalchemy.orm import sessionmaker from swh.core import config from swh.scheduler import get_scheduler, utils from swh.storage import get_storage from .abstractattribute import AbstractAttribute logger = logging.getLogger(__name__) def utcnow(): return datetime.datetime.now(tz=datetime.timezone.utc) class FetchError(RuntimeError): def __init__(self, response): self.response = response def __str__(self): return repr(self.response) class SWHListerBase(abc.ABC, config.SWHConfig): """Lister core base class. Generally a source code hosting service provides an API endpoint for listing the set of stored repositories. A Lister is the discovery service responsible for finding this list, all at once or sequentially by parts, and queueing local tasks to fetch and ingest the referenced repositories. The core method in this class is ingest_data. Any subclasses should be calling this method one or more times to fetch and ingest data from API endpoints. See swh.lister.core.lister_base.SWHIndexingLister for example usage. This class cannot be instantiated. Any instantiable Lister descending from SWHListerBase must provide at least the required overrides. (see member docstrings for details): Required Overrides: MODEL def transport_request def transport_response_to_string def transport_response_simplified def transport_quota_check Optional Overrides: def filter_before_inject def is_within_bounds """ MODEL = AbstractAttribute('Subclass type (not instance)' ' of swh.lister.core.models.ModelBase' ' customized for a specific service.') LISTER_NAME = AbstractAttribute("Lister's name") - @abc.abstractmethod def transport_request(self, identifier): """Given a target endpoint identifier to query, try once to request it. Implementation of this method determines the network request protocol. Args: identifier (string): unique identifier for an endpoint query. e.g. If the service indexes lists of repositories by date and time of creation, this might be that as a formatted string. Or it might be an integer UID. Or it might be nothing. It depends on what the service needs. Returns: the entire request response Raises: Will catch internal transport-dependent connection exceptions and raise swh.lister.core.lister_base.FetchError instead. Other non-connection exceptions should propagate unchanged. """ pass - @abc.abstractmethod def transport_response_to_string(self, response): """Convert the server response into a formatted string for logging. Implementation of this method depends on the shape of the network response object returned by the transport_request method. Args: response: the server response Returns: a pretty string of the response """ pass - @abc.abstractmethod def transport_response_simplified(self, response): """Convert the server response into list of a dict for each repo in the response, mapping columns in the lister's MODEL class to repo data. Implementation of this method depends on the server API spec and the shape of the network response object returned by the transport_request method. Args: response: response object from the server. Returns: list of repo MODEL dicts ( eg. [{'uid': r['id'], etc.} for r in response.json()] ) """ pass - @abc.abstractmethod def transport_quota_check(self, response): """Check server response to see if we're hitting request rate limits. Implementation of this method depends on the server communication protocol and API spec and the shape of the network response object returned by the transport_request method. Args: response (session response): complete API query response Returns: 1) must retry request? True/False 2) seconds to delay if True """ pass def filter_before_inject(self, models_list): """Function run after transport_response_simplified but before injection into the local db and creation of workers. Can be used to eliminate some of the results if necessary. MAY BE OVERRIDDEN if an intermediate Lister class needs to filter results before injection without requiring every child class to do so. Args: models_list: list of dicts returned by transport_response_simplified. Returns: models_list with entries changed according to custom logic. """ return models_list def do_additional_checks(self, models_list): """Execute some additional checks on the model list. For example, to check for existing repositories in the db. MAY BE OVERRIDDEN if an intermediate Lister class needs to check some more the results before injection. Checks are fine by default, returns the models_list as is by default. Args: models_list: list of dicts returned by transport_response_simplified. Returns: models_list with entries if checks ok, False otherwise """ return models_list def is_within_bounds(self, inner, lower=None, upper=None): """See if a sortable value is inside the range [lower,upper]. MAY BE OVERRIDDEN, for example if the server indexable* key is technically sortable but not automatically so. * - ( see: swh.lister.core.indexing_lister.SWHIndexingLister ) Args: inner (sortable type): the value being checked lower (sortable type): optional lower bound upper (sortable type): optional upper bound Returns: whether inner is confined by the optional lower and upper bounds """ try: if lower is None and upper is None: return True elif lower is None: ret = inner <= upper elif upper is None: ret = inner >= lower else: ret = lower <= inner <= upper self.string_pattern_check(inner, lower, upper) except Exception as e: logger.error(str(e) + ': %s, %s, %s' % (('inner=%s%s' % (type(inner), inner)), ('lower=%s%s' % (type(lower), lower)), ('upper=%s%s' % (type(upper), upper))) ) raise return ret # You probably don't need to override anything below this line. DEFAULT_CONFIG = { 'storage': ('dict', { 'cls': 'remote', 'args': { 'url': 'http://localhost:5002/' }, }), 'scheduler': ('dict', { 'cls': 'remote', 'args': { 'url': 'http://localhost:5008/' }, }), 'lister': ('dict', { 'cls': 'local', 'args': { 'db': 'postgresql:///lister', }, }), } @property def CONFIG_BASE_FILENAME(self): # noqa: N802 return 'lister_%s' % self.LISTER_NAME @property def ADDITIONAL_CONFIG(self): # noqa: N802 return { 'credentials': ('dict', {}), 'cache_responses': ('bool', False), 'cache_dir': ('str', '~/.cache/swh/lister/%s' % self.LISTER_NAME), } INITIAL_BACKOFF = 10 MAX_RETRIES = 7 CONN_SLEEP = 10 def __init__(self, override_config=None): self.backoff = self.INITIAL_BACKOFF logger.debug('Loading config from %s' % self.CONFIG_BASE_FILENAME) self.config = self.parse_config_file( base_filename=self.CONFIG_BASE_FILENAME, additional_configs=[self.ADDITIONAL_CONFIG] ) self.config['cache_dir'] = os.path.expanduser(self.config['cache_dir']) if self.config['cache_responses']: config.prepare_folders(self.config, 'cache_dir') if override_config: self.config.update(override_config) logger.debug('%s CONFIG=%s' % (self, self.config)) self.storage = get_storage(**self.config['storage']) self.scheduler = get_scheduler(**self.config['scheduler']) self.db_engine = create_engine(self.config['lister']['args']['db']) self.mk_session = sessionmaker(bind=self.db_engine) self.db_session = self.mk_session() def reset_backoff(self): """Reset exponential backoff timeout to initial level.""" self.backoff = self.INITIAL_BACKOFF def back_off(self): """Get next exponential backoff timeout.""" ret = self.backoff self.backoff *= 10 return ret def safely_issue_request(self, identifier): """Make network request with retries, rate quotas, and response logs. Protocol is handled by the implementation of the transport_request method. Args: identifier: resource identifier Returns: server response """ retries_left = self.MAX_RETRIES do_cache = self.config['cache_responses'] r = None while retries_left > 0: try: r = self.transport_request(identifier) except FetchError: # network-level connection error, try again logger.warning( 'connection error on %s: sleep for %d seconds' % (identifier, self.CONN_SLEEP)) time.sleep(self.CONN_SLEEP) retries_left -= 1 continue if do_cache: self.save_response(r) # detect throttling must_retry, delay = self.transport_quota_check(r) if must_retry: logger.warning( 'rate limited on %s: sleep for %f seconds' % (identifier, delay)) time.sleep(delay) else: # request ok break retries_left -= 1 if not retries_left: logger.warning( 'giving up on %s: max retries exceeded' % identifier) return r def db_query_equal(self, key, value): """Look in the db for a row with key == value Args: key: column key to look at value: value to look for in that column Returns: sqlalchemy.ext.declarative.declarative_base object with the given key == value """ if isinstance(key, str): key = self.MODEL.__dict__[key] return self.db_session.query(self.MODEL) \ .filter(key == value).first() def winnow_models(self, mlist, key, to_remove): """Given a list of models, remove any with matching some member of a list of values. Args: mlist (list of model rows): the initial list of models key (column): the column to filter on to_remove (list): if anything in mlist has column equal to one of the values in to_remove, it will be removed from the result Returns: A list of model rows starting from mlist minus any matching rows """ if isinstance(key, str): key = self.MODEL.__dict__[key] if to_remove: return mlist.filter(~key.in_(to_remove)).all() else: return mlist.all() def db_num_entries(self): """Return the known number of entries in the lister db""" return self.db_session.query(func.count('*')).select_from(self.MODEL) \ .scalar() def db_inject_repo(self, model_dict): """Add/update a new repo to the db and mark it last_seen now. Args: model_dict: dictionary mapping model keys to values Returns: new or updated sqlalchemy.ext.declarative.declarative_base object associated with the injection """ sql_repo = self.db_query_equal('uid', model_dict['uid']) if not sql_repo: sql_repo = self.MODEL(**model_dict) self.db_session.add(sql_repo) else: for k in model_dict: setattr(sql_repo, k, model_dict[k]) sql_repo.last_seen = utcnow() return sql_repo def origin_dict(self, origin_type, origin_url, **kwargs): """Return special dict format for the origins list Args: origin_type (string) origin_url (string) Returns: the same information in a different form """ return { 'type': origin_type, 'url': origin_url, } def task_dict(self, origin_type, origin_url, **kwargs): """Return special dict format for the tasks list Args: origin_type (string) origin_url (string) Returns: the same information in a different form """ _type = 'load-%s' % origin_type _policy = 'recurring' return utils.create_task_dict(_type, _policy, origin_url) def string_pattern_check(self, a, b, c=None): """When comparing indexable types in is_within_bounds, complex strings may not be allowed to differ in basic structure. If they do, it could be a sign of not understanding the data well. For instance, an ISO 8601 time string cannot be compared against its urlencoded equivalent, but this is an easy mistake to accidentally make. This method acts as a friendly sanity check. Args: a (string): inner component of the is_within_bounds method b (string): lower component of the is_within_bounds method c (string): upper component of the is_within_bounds method Returns: nothing Raises: TypeError if strings a, b, and c don't conform to the same basic pattern. """ if isinstance(a, str): a_pattern = re.sub('[a-zA-Z0-9]', '[a-zA-Z0-9]', re.escape(a)) if (isinstance(b, str) and (re.match(a_pattern, b) is None) or isinstance(c, str) and (re.match(a_pattern, c) is None)): logger.debug(a_pattern) raise TypeError('incomparable string patterns detected') def inject_repo_data_into_db(self, models_list): """Inject data into the db. Args: models_list: list of dicts mapping keys from the db model for each repo to be injected Returns: dict of uid:sql_repo pairs """ injected_repos = {} for m in models_list: injected_repos[m['uid']] = self.db_inject_repo(m) return injected_repos def create_missing_origins_and_tasks(self, models_list, injected_repos): """Find any newly created db entries that don't yet have tasks or origin objects assigned. Args: models_list: a list of dicts mapping keys in the db model for each repo injected_repos: dict of uid:sql_repo pairs that have just been created Returns: Nothing. Modifies injected_repos. """ origins = {} tasks = {} def _origin_key(m): _type = m.get('origin_type', m.get('type')) _url = m.get('origin_url', m.get('url')) return '%s-%s' % (_type, _url) def _task_key(m): return '%s-%s' % (m['type'], json.dumps(m['arguments'], sort_keys=True)) for m in models_list: ir = injected_repos[m['uid']] if not ir.origin_id: origin_dict = self.origin_dict(**m) origins[_origin_key(m)] = (ir, m, origin_dict) if not ir.task_id: task_dict = self.task_dict(**m) tasks[_task_key(task_dict)] = (ir, m, task_dict) new_origins = self.storage.origin_add( (origin_dicts for (_, _, origin_dicts) in origins.values())) for origin in new_origins: ir, m, _ = origins[_origin_key(origin)] ir.origin_id = origin['id'] new_tasks = self.scheduler.create_tasks( (task_dicts for (_, _, task_dicts) in tasks.values())) for task in new_tasks: ir, m, _ = tasks[_task_key(task)] ir.task_id = task['id'] def ingest_data(self, identifier, checks=False): """The core data fetch sequence. Request server endpoint. Simplify and filter response list of repositories. Inject repo information into local db. Queue loader tasks for linked repositories. Args: identifier: Resource identifier. checks (bool): Additional checks required """ # Request (partial?) list of repositories info response = self.safely_issue_request(identifier) if not response: return response, [] models_list = self.transport_response_simplified(response) models_list = self.filter_before_inject(models_list) if checks: models_list = self.do_additional_checks(models_list) if not models_list: return response, [] # inject into local db injected = self.inject_repo_data_into_db(models_list) # queue workers self.create_missing_origins_and_tasks(models_list, injected) return response, injected def save_response(self, response): """Log the response from a server request to a cache dir. Args: response: full server response cache_dir: system path for cache dir Returns: nothing """ datepath = utcnow().isoformat() fname = os.path.join( self.config['cache_dir'], datepath + '.gz', ) with gzip.open(fname, 'w') as f: f.write(bytes( self.transport_response_to_string(response), 'UTF-8' )) diff --git a/swh/lister/core/simple_lister.py b/swh/lister/core/simple_lister.py index 8455a79..11060eb 100644 --- a/swh/lister/core/simple_lister.py +++ b/swh/lister/core/simple_lister.py @@ -1,76 +1,74 @@ # Copyright (C) 2018 the Software Heritage developers # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information -import abc import logging from swh.core import utils from .lister_base import SWHListerBase class SimpleLister(SWHListerBase): """Lister* intermediate class for any service that follows the simple, 'list in oneshot information' pattern. - Client sends a request to list repositories in oneshot - Client receives structured (json/xml/etc) response with information and stores those in db """ - @abc.abstractmethod def list_packages(self, *args): """Listing packages method. """ pass def ingest_data(self, identifier, checks=False): """Rework the base ingest_data. Request server endpoint which gives all in one go. Simplify and filter response list of repositories. Inject repo information into local db. Queue loader tasks for linked repositories. Args: identifier: Resource identifier (unused) checks (bool): Additional checks required (unused) """ response = self.safely_issue_request(identifier) response = self.list_packages(response) if not response: return response, [] models_list = self.transport_response_simplified(response) models_list = self.filter_before_inject(models_list) all_injected = [] for models in utils.grouper(models_list, n=10000): models = list(models) logging.debug('models: %s' % len(models)) # inject into local db injected = self.inject_repo_data_into_db(models) # queue workers self.create_missing_origins_and_tasks(models, injected) all_injected.append(injected) # flush self.db_session.commit() self.db_session = self.mk_session() return response, all_injected def run(self): """Query the server which answers in one query. Stores the information, dropping actual redundant information we already have. Returns: nothing """ dump_not_used_identifier = 0 response, injected_repos = self.ingest_data(dump_not_used_identifier) if not response and not injected_repos: logging.info('No response from api server, stopping')