diff options
author | drebs <drebs@leap.se> | 2014-03-14 15:13:42 -0300 |
---|---|---|
committer | drebs <drebs@leap.se> | 2014-03-17 11:59:44 -0300 |
commit | 35bf9f390db15e74ad4e37b6fc7fac9d6d7ca658 (patch) | |
tree | 12fecfdccf58e496806e38e8c3bf0cd449fc2275 /common/src | |
parent | 5a2e9ac138faca940e10920be008a229a7a54cca (diff) |
Parallelize get_docs on couch backend (#5008).
Diffstat (limited to 'common/src')
-rw-r--r-- | common/src/leap/soledad/common/couch.py | 78 |
1 files changed, 78 insertions, 0 deletions
diff --git a/common/src/leap/soledad/common/couch.py b/common/src/leap/soledad/common/couch.py index d16563d3..a4f30cf3 100644 --- a/common/src/leap/soledad/common/couch.py +++ b/common/src/leap/soledad/common/couch.py @@ -26,6 +26,7 @@ import binascii import socket import time import sys +import threading from StringIO import StringIO @@ -334,6 +335,47 @@ class CouchDatabase(CommonBackend): A U1DB implementation that uses CouchDB as its persistence layer. """ + # We spawn threads to parallelize the CouchDatabase.get_docs() method + MAX_GET_DOCS_THREADS = 20 + + class _GetDocThread(threading.Thread): + """ + A thread that gets a document from a database. + + TODO: switch this for a twisted deferred to thread. This depends on + replacing python-couchdb for paisley in this module. + """ + + def __init__(self, db, doc_id, check_for_conflicts, + release_fun): + """ + :param db: The database from where to get the document. + :type db: u1db.Database + :param doc_id: The doc_id of the document to be retrieved. + :type doc_id: str + :param check_for_conflicts: Whether the get_doc() method should + check for existing conflicts. + :type check_for_conflicts: bool + :param release_fun: A function that releases a semaphore, to be + called after the document is fetched. + :type release_fun: function + """ + threading.Thread.__init__(self) + self._db = db + self._doc_id = doc_id + self._check_for_conflicts = check_for_conflicts + self._release_fun = release_fun + self._doc = None + + def run(self): + """ + Fetch the document, store it as a property, and call the release + function. + """ + self._doc = self._db._get_doc( + self._doc_id, self._check_for_conflicts) + self._release_fun() + @classmethod def open_database(cls, url, create): """ @@ -401,6 +443,9 @@ class CouchDatabase(CommonBackend): self._set_replica_uid(replica_uid) if ensure_ddocs: self.ensure_ddocs_on_db() + # initialize a thread pool for parallelizing get_docs() + self._sem_pool = threading.BoundedSemaphore( + value=self.MAX_GET_DOCS_THREADS) def ensure_ddocs_on_db(self): """ @@ -1331,6 +1376,39 @@ class CouchDatabase(CommonBackend): old_doc.has_conflicts = doc.has_conflicts return state, self._get_generation() + def get_docs(self, doc_ids, check_for_conflicts=True, + include_deleted=False): + """ + Get the JSON content for many documents. + + :param doc_ids: A list of document identifiers. + :type doc_ids: list + :param check_for_conflicts: If set to False, then the conflict check + will be skipped, and 'None' will be + returned instead of True/False. + :type check_for_conflictsa: bool + :param include_deleted: If set to True, deleted documents will be + returned with empty content. Otherwise deleted + documents will not be included in the results. + :return: iterable giving the Document object for each document id + in matching doc_ids order. + :rtype: iterable + """ + # spawn threads to retrieve docs + threads = [] + for doc_id in doc_ids: + self._sem_pool.acquire() + t = self._GetDocThread(self, doc_id, check_for_conflicts, + self._sem_pool.release) + t.start() + threads.append(t) + # join threads and yield docs + for t in threads: + t.join() + if t._doc.is_tombstone() and not include_deleted: + continue + yield t._doc + class CouchSyncTarget(CommonSyncTarget): """ |