The production cluster needs to be load tested to ensure that
- the new hardware is solid, and
- there are no stability issues with Debian Jessie (testing so far was with the older Ubuntu Trusty).
Description
Description
| Status | Subtype | Assigned | Task | |
|---|---|---|---|---|
| · · · | ||||
| Resolved | • GWicke | T89481 RESTBase beta release (revision storage / content API) | ||
| Resolved | • GWicke | T89567 Load test production cluster | ||
| Resolved | fgiunchedi | T89366 Access to restbase / cassandra cluster | ||
| · · · |
Event Timeline
Comment Actions
Things have been fairly smooth on the testing front.
- we encountered no issues from the trusty -> jessie migration
- restbase1006 turned out to have a faulty disk controller: T89639
- peak performance we measured is around 10k HTML revision req/s with a mean latency of ~80ms
- we processed ~7k req/s for a week without issues
- deliberate cassandra node restarts did not impact availability
- creating 12k tables for 800 wikis did not go so well; this was addressed in T91188
- we are currently loading current revisions for all 280 wikipedias into storage, again without issues so far