Description
The gomemcached client currently enables dead-connection detection by default with a 2 minute interval - however this only ensures disconnection by the Producer in the event of a slow Consumer, and not vice-versa.
Currently the gomemcached client just responds to a NOOP request and goes back to listening for the next event.
Hence we need to implement the logic to disconnect as the Consumer to a slow Producer; and then that will be used to replace the CheckStuckness() logic flow.
Attachments
Issue Links
For Gerrit Dashboard: MB-62624 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
212380,6 | MB-62624: Utilise dead-connection detection for a "stuck" DCP Producer | master | gomemcached | Status: MERGED | +2 | +1 |
212381,4 | MB-62624: Utilise dead-connection detection for a "stuck" DCP Producer | master | goxdcr | Status: MERGED | +2 | +1 |
212511,5 | MB-62624: Remove plumbing for DCP Stats (from KV) | master | goxdcr | Status: MERGED | +2 | +1 |
213300,2 | MB-62624: Exports MemcachedConnection type | master | gomemcached | Status: MERGED | +2 | +1 |
213301,2 | MB-62624: Fix bucketTopologyService_test | master | goxdcr | Status: MERGED | +2 | +1 |
213353,2 | MB-62624: Remove 'xdcrDevBucketTopologyLegacyDelay' hidden setting | master | goxdcr | Status: MERGED | +2 | +1 |