[
https://issues.apache.org/jira/browse/COUCHDB-3088?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15400044#comment-15400044
]
ASF GitHub Bot commented on COUCHDB-3088:
-----------------------------------------
GitHub user iilyak opened a pull request:
https://github.com/apache/couchdb-couch-replicator/pull/44
Inject random delays in scan_all_dbs
couch_replication_server scans filesystem to find all _replication
databases. For every database found it does
gen_server:cast(Server, {resume_scan, DbName})
Extract independent process where we do gen_server:cast after a random
delay.
This effectively removes stampede and randomizes the order in which we
process _replication databases.
COUCHDB-3088
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/cloudant/couchdb-couch-replicator
69914-insert-random-delays
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/couchdb-couch-replicator/pull/44.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #44
----
commit 5715c5e25dba61442834b08d7d7202b185341a87
Author: ILYA Khlopotov <[email protected]>
Date: 2016-07-29T21:32:02Z
Inject random delays in scan_all_dbs
couch_replication_server scans filesystem to find all _replication
databases. For every database found it does
gen_server:cast(Server, {resume_scan, DbName})
Extract independent process where we do gen_server:cast after a random
delay.
This effectively removes stampede and randomizes the order in which we
process _replication databases.
COUCHDB-3088
----
> restart of couch_replication_server causes a stampede
> -----------------------------------------------------
>
> Key: COUCHDB-3088
> URL: https://issues.apache.org/jira/browse/COUCHDB-3088
> Project: CouchDB
> Issue Type: Bug
> Reporter: ILYA
>
> couch_replication_server scans all files in database_dir searching for files
> matching "_replicator.<number>.couch". For every _replication db it does
> gen_server:cast(Server, {resume_scan, DbName}). This creates a stampede
> effect and causes sharp load spikes on the replication cluster. The problem
> get worse if you migrate from older version of couchdb. In this case there is
> a logic which injects validation ddoc into every _replication db. Causing a
> spike in [couchdb, database_writes] metric.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
