Like T711, we should deploy statsd plugin in production as well. It'll allow us to stop having to send this data to ganglia via the script we have now and will reduce the delta between prod and labs.
|Resolved||• chasemp||T90889 Deploy statsd plugin for production elasticsearch & logstash|
|Resolved||EBernhardson||T100500 Install elasticsearch-statsd-plugin to the beta cluster|
|Resolved||Gehel||T102159 [EPIC] Elastic search service takes many hours to restart|
|Resolved||Deskana||T101530 Upgrade Cirrus to Elasticsearch 1.6.0|
|Resolved||• Manybubbles||T102008 Import Elasticsearch 1.6.0 deb into wmf apt|
|Resolved||EBernhardson||T99244 CirrusSearch: Support pausing writes to Elasticsearch|
- Mentioned In
- T109089: EPIC: Cultivating the Elasticsearch garden (operational lessons from 1.7.1 upgrade)
T99763: Determine if poolcounter hits in CirrusSearch-Search:_elasticsearch_enwiki are due to increased number of queries, or queries taking longer to process.
- Mentioned Here
- T111573: Setup diamond for the elasticsearch cluster
T711: Deploy statsd plugin for beta elasticsearch.
Note that statsd plugin for elasticsearch was undeployed from beta cluster some time ago (not sure why).
FWIW feel free to add me on code reviews when deploying/adding the statsd plugin, mostly to assess how many metrics are going to be pushed
What statsd plugin is this? I do know diamond has a native elasticsearch poller as well.
This is https://github.com/ebernhardson/elasticsearch-statsd-plugin/tree/v0.3.3-wmf2 which is a forked from Automattic (which is forked from somewhere else). Its not the most maintained piece of software, we would basically be taking over. If there are better options i'm all ears.