Details
-
Bug
-
Resolution: Won't Fix
-
Critical
-
Cheshire-Cat
-
Untriaged
-
1
-
Unknown
Description
When running n1ql covering array query, CPU usage is much higher.
http://perf.jenkins.couchbase.com/job/aether/191
Looking at CPU profile, a lot of time is being spent on mcall. There are 2 possibilities:
1) IO is saturated and log cleaning is falling behind. Writer is stalled due to tryThrottleForLSS(). This function can invoke mcall.
2) There is a lot of goroutines contending for the write buffer.
Attachments
For Gerrit Dashboard: MB-45960 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
152294,4 | MB-45960: Do not call Gosched in tryThrottleForLSS | unstable | plasma | Status: ABANDONED | 0 | -1 |
152388,1 | MB-45960: Limit number of writers/swappers for evicting dirty pages | unstable | plasma | Status: ABANDONED | 0 | -1 |
152411,1 | MB-45960: Use constant EvictRunInterval | unstable | plasma | Status: ABANDONED | 0 | -1 |
152414,1 | MB-45960: Compact | unstable | plasma | Status: ABANDONED | 0 | -1 |