3e1b77efc2
Merge branch 'erikj/cached_keyword_args' of github.com:matrix-org/synapse into erikj/dictionary_cache
2015-08-05 16:45:56 +01:00
b52b4a84ec
Merge branch 'develop' of github.com:matrix-org/synapse into erikj/dictionary_cache
2015-08-05 15:41:20 +01:00
a89559d797
Use LRU cache by default
2015-08-05 15:39:47 +01:00
07507643cb
Use dictionary cache to do group -> state fetching
2015-08-05 15:11:42 +01:00
4d6cb8814e
Speed up event filtering (for ACL) logic
2015-08-04 09:32:23 +01:00
39e21ea51c
Add support for using keyword arguments with cached functions
2015-07-27 13:57:29 +01:00
80a61330ee
Add basic storage functions for handling of receipts
2015-07-01 17:19:12 +01:00
d8866d7277
Caches should be bound to instances.
...
Before, caches were global and so different instances of the stores
would share caches. This caused problems in the unit tests.
2015-06-03 14:45:17 +01:00
4429e720ae
Merge branch 'master' of github.com:matrix-org/synapse into develop
2015-05-22 10:33:00 +01:00
53447e9cd3
Add caches for things requested by the pushers
2015-05-21 16:41:39 +01:00
d61ce3f670
Add a cache for get_current_state with state_key
2015-05-21 16:41:39 +01:00
65878a2319
Remove unused metric
2015-05-18 14:06:30 +01:00
c71176858b
Newline, remove debug logging
2015-05-18 10:11:14 +01:00
c3b37abdfd
PEP8
2015-05-15 16:59:58 +01:00
aa32bd38e4
Add a wait
2015-05-15 11:35:04 +01:00
a2c4f3f150
Fix daedlock
2015-05-15 10:54:04 +01:00
1d566edb81
Remove race condition
2015-05-14 16:54:35 +01:00
142934084a
Count and loop
2015-05-14 15:40:21 +01:00
7cd6a6f6cf
Awful idea for speeding up fetching of events
2015-05-14 15:34:02 +01:00
386b7330d2
Move from _base to events
2015-05-14 14:45:22 +01:00
7d6a1dae31
Jump out early
2015-05-14 14:27:58 +01:00
656223fbd3
Actually, we probably want to run this in a transaction
2015-05-14 14:26:35 +01:00
2f7f8e1c2b
Preemptively jump into a transaction if we ask for get_prev_content
2015-05-14 14:17:36 +01:00
e1e9f0c5b2
loop -> gatherResults
2015-05-14 13:58:49 +01:00
ab78a8926e
Err, we probably want a bigger limit
2015-05-14 13:47:16 +01:00
cdb3757942
Refactor _get_events
2015-05-14 13:31:55 +01:00
36ea26c5c0
Merge branch 'develop' of github.com:matrix-org/synapse into erikj/perf
2015-05-14 12:01:38 +01:00
7c549dd557
Add ID generator for push_rules_enable to #resolve SYN-378
2015-05-14 11:44:03 +01:00
968b01a91a
Actually use async method
2015-05-13 17:02:46 +01:00
4071f29653
Fetch events from events_id in their own transactions
2015-05-13 16:59:41 +01:00
a988361aea
Typo
2015-05-13 15:44:15 +01:00
8888982db3
Don't insert None
2015-05-13 15:43:32 +01:00
cf706cc6ef
Don't return None
2015-05-13 15:31:25 +01:00
5971d240d4
Limit batch size
2015-05-13 15:26:49 +01:00
ca4f458787
Fetch events in bulk
2015-05-13 15:13:42 +01:00
6edff11a88
Don't fetch redaction and rejection stuff for each event, so we can use index only scan
2015-05-13 14:39:05 +01:00
409bcc76bd
Load events for state group seperately
2015-05-13 11:13:31 +01:00
da6a7bbdde
Merge branch 'develop' of github.com:matrix-org/synapse into erikj/logging_context
2015-05-12 13:10:42 +01:00
cd525c0f5a
push_rules table expects an 'id' field
2015-05-11 11:24:01 +01:00
476899295f
Change the way we do logging contexts so that they survive divergences
2015-05-08 16:32:18 +01:00
ed2584050f
Merge branch 'develop' of github.com:matrix-org/synapse into erikj/executemany
2015-05-05 18:15:20 +01:00
31049c4d72
Merge pull request #139 from matrix-org/bugs/SYN-369
...
Fix race with cache invalidation. SYN-369
2015-05-05 17:46:13 +01:00
deb0237166
Add some doc-string
2015-05-05 17:45:11 +01:00
d18f37e026
Collect the invalidate callbacks on the transaction object rather than passing around a separate list
2015-05-05 17:32:21 +01:00
9951542393
Add a comment about the zip(*[zip(sorted(...),...)])
2015-05-05 17:06:55 +01:00
041b6cba61
SYN-369: Add comments to the sequence number logic in the cache
2015-05-05 16:32:44 +01:00
63075118a5
Add debug flag in synapse/storage/_base.py for debugging the cache logic by comparing what is in the cache with what was in the database on every access
2015-05-05 16:24:04 +01:00
43c2e8deae
Add support for using executemany
2015-05-05 15:13:25 +01:00
1692dc019d
Don't call 'encode_parameter' no-op
2015-05-05 15:00:30 +01:00
261d809a47
Sequence the modifications to the cache so that selects don't race with inserts
2015-05-05 14:13:50 +01:00
657298cebd
Don't lock user_ips table for upsert.
2015-05-01 10:46:48 +01:00
fabb7acd45
Fix bug where we reconnected to the database on every query.
2015-05-01 10:24:24 +01:00
2732be83d9
Shuffle operations so that locking upsert happens last in the txn. This ensures the lock is held for the least amount of time possible.
2015-04-27 13:22:30 +01:00
e4c4664d73
Handle the fact that postgres databases can be restarted from under us
2015-04-27 12:40:49 +01:00
b8092fbc82
Go back to storing JSON in TEXT
2015-04-16 11:17:52 +01:00
5b31afcbd1
Remove debug logging
2015-04-15 16:27:04 +01:00
ed26e4012b
pushers table requires a unique id.
2015-04-15 16:24:14 +01:00
a5c72780e6
Don't pass in removed flag
2015-04-15 15:13:22 +01:00
4af32a2817
Postgres does not allow you to continue using a cursor after a DB exception has been raised, so move _simple_insert or_ignore flag out of transaction
2015-04-15 14:51:21 +01:00
58d8339966
Add support for postgres instead of mysql. Change sql accourdingly. blob + varbinary -> bytea. No support for UNSIGNED or CREATE INDEX IF NOT EXISTS.
2015-04-14 13:53:20 +01:00
7ed2ec3061
Handle the fact that in sqlite binary data might be stored as unicode or bytes
2015-04-10 13:41:54 +01:00
8bf285e082
Merge branch 'develop' of github.com:matrix-org/synapse into mysql
2015-04-09 13:13:26 +01:00
8ad0f4912e
Stream ordering and out of order insertions.
...
Handle the fact that events can be persisted out of order, and so to get
the "current max" stream token becomes non trivial - as we need to make
sure that *all* stream tokens less than the current max have also
successfully been persisted.
2015-04-09 11:41:36 +01:00
83f5125d52
Merge pull request #114 from matrix-org/improve_get_event_cache
...
Improve get event cache
2015-04-08 16:50:27 +01:00
3887350e47
Merge branch 'develop' of github.com:matrix-org/synapse into mysql
2015-04-08 14:11:07 +01:00
c1b34af441
Move database timer logging to seperate logger
2015-04-08 13:31:19 +01:00
9a05795619
Retry transaction, not SQL query
2015-04-08 13:11:28 +01:00
4fe95094d1
Merge branch 'develop' of github.com:matrix-org/synapse into mysql
2015-04-07 18:05:39 +01:00
49d6aa1394
Retry on deadlock
2015-04-07 15:28:37 +01:00
0af5f5efaf
Don't use multiple UNIQUE constraints; it will cause deadlocks
2015-04-07 12:08:35 +01:00
c8d3f6486d
Implement or_ignore flag on inserts
2015-04-07 12:06:22 +01:00
304111afd0
Don't use AUTOINCREMENT, use an in memory version
2015-04-07 12:05:36 +01:00
779f7b0f44
Fix unicode support
2015-04-02 10:06:22 +01:00
9236136f3a
Make work in both Maria and SQLite. Fix tests
2015-04-01 14:12:33 +01:00
a198894bf7
Appease pep8
2015-03-26 11:53:58 +00:00
953e40f9dc
Implement the main getEvent cache using Cache() instead of a custom application of LruCache; also unify its two-level structure into just one
2015-03-25 19:12:16 +00:00
f173d40a32
Use FrozenEvent's reject_reason to decide whether to return it; don't include allow_rejected in the main getEvents cache key
2015-03-25 19:06:05 +00:00
1b988b051b
Store the rejected reason in (Frozen)Event structs
2015-03-25 19:06:05 +00:00
033a517feb
Indirect invalidations of _get_event_cache via a helper method to keep all uses of the cache lexically within one .py file
2015-03-25 19:06:05 +00:00
9ba6487b3f
Allow a choice of LRU behaviour for Cache() by using LruCache() or OrderedDict()
2015-03-25 19:05:34 +00:00
0e8f5095c7
Fix unicode database support
2015-03-25 17:15:20 +00:00
9e98f1022a
Don't order by rowid
2015-03-24 16:19:01 +00:00
6e7131f02f
Remove uses of REPLACE and ON CONFLICT IGNORE to make the SQL more portable.
2015-03-23 15:38:56 +00:00
0f86312c4c
Pull out the cache logic from the @cached wrapper into its own class we can reuse
2015-03-20 18:25:42 +00:00
b1022ed8b5
func(*EXPR) is valid Python syntax, really...
2015-03-20 17:47:45 +00:00
f6583796fe
Merge branch 'develop' of github.com:matrix-org/synapse into mysql
2015-03-20 16:31:48 +00:00
80cd08c190
PEP8
2015-03-20 16:03:25 +00:00
9517f4da4d
Merge branch 'develop' of github.com:matrix-org/synapse into store_rearrangement
2015-03-20 16:02:47 +00:00
dc0c989ef4
Give sensible names for '_simple_...' transactions
2015-03-20 15:59:18 +00:00
7e282a53a5
Tidy up _simple_... methods
2015-03-20 15:05:10 +00:00
91cb46191d
Allow @cached-wrapped functions to have more or fewer than 1 argument; assert on the total count of them though
2015-03-20 14:59:45 +00:00
87db64b839
Rearrange storage modules
2015-03-20 14:11:38 +00:00
cb8162d3d1
Rearrange storage modules
2015-03-20 13:52:56 +00:00
d7a0496f3e
Convert storage layer to be mysql compatible
2015-03-19 15:59:48 +00:00
1489521ee5
Be polite and ensure we use @functools.wraps() when creating a function decorator
2015-03-17 17:19:22 +00:00
be170b1426
Add a metric for the scheduling latency of SQL queries
2015-03-16 17:21:59 +00:00
e75fa8bbbf
Bugfix to sql_txn_timer increment - add only the per-TXN duration, not the total time ever spent since boot
2015-03-12 16:24:51 +00:00
63cb7ece62
Rename the timer metrics exported by synapse.storage to append _time, so the meaning of ':total' is clearer
2015-03-12 16:24:51 +00:00
f1fbe3e09f
Rename TimerMetric to DistributionMetric; as it could count more than just time
2015-03-12 16:24:51 +00:00
642f725fd7
Pretend the 'getEvent' cache is just another cache in the set of all the others for metric
2015-03-12 16:24:51 +00:00
b0cf867319
Use _ instead of . as a metric namespacing separator, for Prometheus
2015-03-12 16:24:51 +00:00
f9478e475b
Rename Metrics' "keys" to "labels"
2015-03-12 16:24:51 +00:00
fa319a5786
Add TimerMetrics to shadow the PerformanceCounters in synapse.storage; with the view to eventually replacing them entirely
2015-03-12 16:24:51 +00:00
25187ab674
Collect per-SQL-verb timer stats on query execution time
2015-03-12 16:24:50 +00:00
e1a7e3564f
Delete a couple of TODO markers of monitoring stats now done
2015-03-12 16:24:50 +00:00
8664599af7
Rename CacheCounterMetric to just CacheMetric; add a CallbackMetric component to give the size of the cache
2015-03-12 16:24:50 +00:00
b0cdf097f4
Sprinkle some CacheCounterMetrics around the synapse.storage layer
2015-03-12 16:24:50 +00:00
59a5f012cc
Also give _execute() a description
2015-03-11 17:19:17 +00:00
099e4b88d8
Add a description to storage layer's _execute_and_decode()
2015-03-11 17:08:57 +00:00
3d73383d18
Modify _simple_select_list to allow an empty WHERE clause. Use it for get_all_rooms and get_all_users.
2015-03-02 10:16:24 +00:00
9640510de2
Use OrderedDict for @cached backing store, so we can evict the oldest key unbiased
2015-02-23 18:41:58 +00:00
f53fcbce97
Use cache.pop() instead of a separate membership test + del []
2015-02-23 18:30:45 +00:00
27080698e7
Fix code style warning
2015-02-23 18:19:13 +00:00
a09e59a698
Pull the _get_event_cache.setdefault() call out of the try block, as it doesn't need to be there and is confusing
2015-02-23 16:55:57 +00:00
e76d485e29
Allow @cached-wrapped functions to have a prefill method for setting entries
2015-02-23 15:41:54 +00:00
55022d6ca5
Remove a TODO note
2015-02-19 18:38:09 +00:00
ebc3db295b
Take named arguments to @cached() decorator, add a 'max_entries' limit
2015-02-19 18:36:02 +00:00
077d200342
Move @cached decorator out into synapse.storage._base; add minimal docs
2015-02-19 17:29:39 +00:00
42bc56dad3
Merge branch 'develop' of github.com:matrix-org/synapse into use-simplejson
2015-02-11 17:01:38 +00:00
aff892ce79
Fix formatting
2015-02-11 15:02:35 +00:00
f5a70e0d2e
Add a cache for get_event
2015-02-11 15:01:15 +00:00
4ebbaf0d43
Blunty replace json with simplejson
2015-02-11 14:23:10 +00:00
b085fac735
Code-style fixes
2015-02-10 16:30:48 +00:00
fda4422bc9
Fix pyflakes
2015-02-10 14:54:07 +00:00
d7c7efb691
Add performance counters for different stages of loading events
2015-02-10 14:50:53 +00:00
0c4536da8f
Use the transaction 'desc' rather than 'name', increment the txn_ids in
...
txn names
2015-02-09 18:06:31 +00:00
347b497db0
Formatting
2015-02-09 17:57:09 +00:00
3a5ad7dbd5
Performance counters for database transaction names
2015-02-09 17:55:56 +00:00
c4ee4ce93e
Fix typo
2015-02-09 15:00:37 +00:00
66fde49f07
Log database time every 10s and log as percentage
2015-02-09 14:45:15 +00:00
75656712e3
Time how long we're spending on the database thread
2015-02-09 14:22:52 +00:00
2f4cb04f45
Be more specific in naming columns in selects.
2015-01-30 14:48:11 +00:00
e0b7c521cb
Merge branch 'develop' of github.com:matrix-org/synapse into rejections_storage
...
Conflicts:
synapse/storage/__init__.py
synapse/storage/schema/delta/v12.sql
2015-01-30 14:08:28 +00:00
6d485dd1c7
unnecessary newlines
2015-01-28 14:48:42 +00:00
fb0928097a
More magic commas (including the place I copied it from...)
2015-01-28 14:48:07 +00:00
0cbb6b0f52
Google doc style
2015-01-28 14:44:41 +00:00
b1b85753d7
Add support for storing rejected events in EventContext and data stores
2015-01-22 15:50:17 +00:00
c06a9063e1
Merge branch 'develop' into pushers
2015-01-13 13:15:51 +00:00
36a2a877e2
Use time.time() instead of time.clock()
2015-01-06 16:34:41 +00:00
52b2c6c9c7
Don't include None's in _get_events_txn
2015-01-06 14:56:57 +00:00
5e23a19204
Merge pull request #28 from matrix-org/erikj-perf
...
Database performance improvements.
2015-01-06 13:33:40 +00:00
adb04b1e57
Update copyright notices
2015-01-06 13:21:39 +00:00
af1c7c7808
PEP8
2015-01-06 13:13:17 +00:00
52d8519008
Don't do batching when getting events.
2015-01-06 13:10:27 +00:00
98933e3db6
Only fetch prev_content when a client is streaming/paginating. Use transactions for event streams.
2015-01-06 13:03:23 +00:00
3e26720e05
Temporarily turn off 'redacted_because' and 'prev_content' keys
2015-01-06 11:26:58 +00:00
f4ea78e9e2
More debug logging
2015-01-06 11:24:18 +00:00
d7e8ea67b3
Reformat
2015-01-06 11:18:02 +00:00
b56730bb6e
Merge branch 'develop' into pushers
...
Conflicts:
synapse/api/errors.py
synapse/server.py
synapse/storage/__init__.py
2014-12-18 15:15:22 +00:00
9728c305a3
after a few rethinks, a working implementation of pushers.
2014-12-18 14:49:22 +00:00
Erik Johnston