Details
-
Improvement
-
Resolution: Unresolved
-
Major
-
3.0.1
-
Security Level: Public
-
None
Description
Friends,
we have a 20 node cluster with 10 buckets with little work (10K requests per cluster).
we like that setup due to various reasons.
yet we feel that statistics collection polling by ns_server produces additional CPU load for both, ns_server and memcached processes.
we've came across a runtime-configurable parameter,
date; curl -u teligent:teligent -XPOST -d 'application_controller:get_env(ns_server, grab_stats_every_n_ticks).' http://localhost:8091/diag/eval
|
#default undefined, means 1, means 1 second (more precise: tick_interval milliseconds, which defaults to 1000)
|
date; curl -u teligent:teligent -XPOST -d 'application_controller:set_env(ns_server, grab_stats_every_n_ticks, 10).' http://localhost:8091/diag/eval
|
which does work and does reduce the polling and does reduce CPU usage for our setup.
now
- numbers under graphs are updated once per 10 seconds OK
- fragmentation when reaches auto-compaction, autocompaction gets triggered OK
- but that breaks dear web-console graphs.
Is there some way to fix web-console graphs?
What do you think of this approach to reduce load?