graphite1001 has cpu to spare, and we're currently running only one statsdlb process which often spikes to using one core at 100%, with SO_REUSEADDR already in statsdlb we should be able to have multiple processes all receiving incoming udp traffic and share the load
Description
Description
Details
Details
Status | Subtype | Assigned | Task | ||
---|---|---|---|---|---|
Resolved | fgiunchedi | T105218 check_graphite - "UNKNOWN: More than half of the datapoints are undefined " | |||
Resolved | fgiunchedi | T101141 UDP rcvbuferrors and inerrors on graphite hosts | |||
Resolved | fgiunchedi | T126447 add more statsdlb instances for more throughput |
Event Timeline
Comment Actions
Change 273920 had a related patch set uploaded (by Filippo Giunchedi):
set SO_REUSEADDR before bind()
Comment Actions
Change 282356 had a related patch set uploaded (by Faidon Liambotis):
Add a new statsd_proxy module and replace statsdlb
Comment Actions
Change 296920 had a related patch set uploaded (by Filippo Giunchedi):
use SO_REUSEPORT in code, not SO_REUSEADDR
Comment Actions
Change 296920 merged by Filippo Giunchedi:
use SO_REUSEPORT in code, not SO_REUSEADDR
Comment Actions
Change 282356 merged by Filippo Giunchedi:
Add a new statsd_proxy module and replace statsdlb
Comment Actions
Resolving this since it essentially happened by replacing statsdlb with statsd_proxy, followup on T101141