We have an API implemented in Tornado using CouchDB as it's backend.
Usually, queries to CouchDB finishes in less than 50ms.
But lately, queries takes up to 250ms, sometimes more. After some New Year's Day troubleshooting, we figured out that replication tanks performance so much, we disabled it, and performance rose again.
Last night around midnight (UTC+7) we reenabled replication, no impact. But this morning when the API starts to see high loads, the CouchDB performance tanked again.
- Is there a way to troubleshoot this performance issue?
- If replication turns out to be the culprit, how to optimize replication?
