Over the weekend, the kafka* servers saw a sharp increase in tracked connections (T131028)., which has now settled down to normal levels. We should have a monitoring metric (Diamond collector?) which reads the current table usage from /proc/sys/net/netfilter/nf_conntrack_count, so that we can track down and correlate changes.
Description
Description
Details
Details
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Add diamond nf_conntrack counter. | operations/puppet | production | +35 -0 |
Related Objects
Related Objects
- Mentioned Here
- T131028: nf_conntrack warnings for kafka hosts
Event Timeline
Comment Actions
Change 280265 had a related patch set uploaded (by Elukey):
Add diamond nf_conntrack counter.
Comment Actions
Already added, the only remaining step was to decide if the collector needs to be deployed everywhere or not, but it is not in the scope of the ticket. Closing!