James Allen
87f3e5e809
Remove old migration code
2016-08-23 11:03:37 +01:00
James Allen
a1bac9719a
Update secondary timeout for multi commands too
2016-07-12 11:45:10 +01:00
James Allen
d0c54f1be6
Increase secondary timeout to 600ms
2016-07-12 11:11:55 +01:00
James Allen
4f34cb7363
Only set keys in migration if they exist
2016-07-08 11:45:36 +01:00
James Allen
8b090c0a60
Don't call sadd with blank args
2016-07-07 16:41:38 +01:00
James Allen
6bc78ccf77
Add in some null checks on comparing backend results
2016-07-07 11:24:33 +01:00
James Allen
1c62a1c5a1
Don't propagate error on secondary timeout
2016-07-07 09:46:15 +01:00
James Allen
edf1e1ab49
Initialize primaryResult to an array, not null
2016-07-06 15:02:05 +01:00
James Allen
59883023ca
Ignore different order of results from smembers
2016-07-06 14:24:27 +01:00
James Allen
89f90c1b04
Timeout secondary requests if they take longer than 200ms
2016-07-06 12:25:36 +01:00
James Allen
508a95c19b
Use the main redis instance for locks
2016-07-06 11:50:02 +01:00
James Allen
e912ccc562
Wait for both backends to return so that they are always in sync
2016-07-05 16:07:47 +01:00
James Allen
dfd45bd23c
Add timers to time how long each redis request takes
2016-07-04 11:14:23 +01:00
James Allen
149351aa49
Remove debugging log lines
2016-06-29 21:18:13 +01:00
James Allen
fde334acc4
Create script that will ensure data is migrate between redis backends
2016-06-29 12:57:56 +01:00
James Allen
dbe03e2757
Track metrics of when backends match or disagree
2016-06-28 16:49:15 +01:00
James Allen
da89ff7172
Add in external health check rather than internal
2016-06-23 18:00:03 +01:00
James Allen
414ab5d6a9
Refactor TCP redis monitoring to be more explicit
2016-06-23 15:59:25 +01:00
James Allen
8ef03c3d2f
Add in application layer monitoring of the health of each cluster node
2016-06-23 15:38:51 +01:00
James Allen
39d920682a
Only pass through keys to redis that are explicitly set in the config
2016-06-17 14:11:22 +01:00
James Allen
40b7da8edb
Support endpoints and masterName in redis config
2016-06-17 14:05:54 +01:00
James Allen
bc00aab7b1
Explicitly separate calls to web and docupdater redis instances
2016-06-17 12:17:22 +01:00
James Allen
9ff3026807
Merge branch 'master' into ja_redis_cluster
2016-06-09 10:00:58 +01:00
James Allen
48a92b28e5
Don't run redis commands in parallel for easier consistency reasoning
2016-06-09 09:54:13 +01:00
James Allen
c823e06912
Don't run redis commands in parallel for easier consistency reasoning
2016-06-08 16:42:09 +01:00
James Allen
c6605ed5f0
Fix misapplied arguments to multi commands
2016-06-08 16:41:58 +01:00
James Allen
b4936f62af
Check that return values from different redis backends match
2016-06-08 12:18:37 +01:00
James Allen
ef43e2b325
Configure backend database as redis cluster
2016-06-07 18:38:32 +01:00
James Allen
1db6f8f159
Pull out rclient into RedisBackend that supports sending requests to multiple rclients
2016-06-07 17:58:18 +01:00
James Allen
ea8f4a6d7a
Remove unused getDocOpsLength method
2016-06-01 11:58:15 +01:00
James Allen
05b09a447e
Ensure that all multi call keys will hash to the same node in cluster
2016-06-01 11:40:12 +01:00
James Allen
f707783aba
Remove old methods of triggering doc updates
...
Remove the old pub/sub listener which is no longer used.
Also remove the DocsWithPendingUpdates set, which used to track
docs waiting to be updated. This was necessary incase messages were
missed on the pub/sub channel, so we knew which docs still had pending
updates. However, now we use the BLPOP queue, these updates just sit in
the queue until a consumer comes back to continue consuming them.
2016-06-01 11:28:23 +01:00
James Allen
b2e58266ae
Remove unused total doc counting redis commands
2016-06-01 11:03:39 +01:00
James Allen
32ce44f9c6
Remove unused gzipping logic
2016-06-01 10:53:50 +01:00
James Allen
11667e8250
Return semantic error if the requested ops are not available when getting a document
2016-05-31 13:24:19 +01:00
Shane Kilkelly
048fd19418
Add one to size of line.
...
To account for newline characters in the original document
2016-05-12 09:26:50 +01:00
Shane Kilkelly
82d5a7fafd
check total size of lines, rather than content-length
2016-05-11 15:55:21 +01:00
Shane Kilkelly
210a61112b
reject payloads larger than 2MB for setDoc
2016-04-29 15:08:21 +01:00
James Allen
4ab4e27724
Include COUNT in lock and make RND per-process incase randomBytes blocks
2016-04-13 16:27:53 +01:00
James Allen
d15a22be26
Remove misleading log line about lock value
2016-04-13 12:20:53 +01:00
James Allen
945c728db2
Use signed locks so only the locking party can remove their lock
2016-04-13 11:59:56 +01:00
James Allen
6c79ab4321
Don't let HTTP calls take longer than 5 seconds since we're inside a 30 second lock
2016-04-12 17:10:39 +01:00
James Allen
af5d01e440
Flush track changes when unloading data from redis
2016-01-20 15:05:31 +00:00
James Allen
184c9031cd
Delete doc from redis after a full set if it wasn't in redis before
2016-01-20 14:31:25 +00:00
Brian Gough
e8f09f3357
added comment about flushing to track changes
2015-12-01 16:03:05 +00:00
Brian Gough
6e97521971
Revert "Merge pull request #4 from sharelatex/flush-ops-from-redis"
...
This reverts commit 84a9ad8b67b860a92271cf7bb669cf21cc0b5b86, reversing
changes made to a6d4649f4f2db06862e8cbc5e132a073ce8c20b2.
2015-12-01 15:58:40 +00:00
Brian Gough
13e22e1802
Revert "only flush to track-changes when ops are queued"
...
This reverts commit 0baa8f989481ca263fa71e90af21f13b29c29504.
2015-12-01 15:48:49 +00:00
Brian Gough
7a9577e081
only flush to track-changes when ops are queued
2015-11-30 15:17:11 +00:00
Brian Gough
529a061567
Merge pull request #4 from sharelatex/flush-ops-from-redis
...
push ops to track changes when document is flushed
2015-11-30 10:46:21 +00:00
James Allen
2589e2d417
Gracefully return when an op has already been submitted
...
It is not a fatal error if an op has already been submitted. We just
need to send an ack back to the client that submitted it and continue.
If we detect a duplicate op, set dup: true on the op and pass it back
to real-time for distributing. The dup: true flag will ensure it only
gets acknowledged to the submitting client, not everyone.
2015-11-19 10:54:28 +00:00
