What exactly is ep_latency_get_cmd?
I'm looking at the stats coming from my cluster and the ep_latency_get_cmd numbers run in lockstep with cmd_get (the values are the same). All of my cache is being served from memory.
Is this trying to measure the latency of the get command, but only to millisecond accuracy so it can't show anything less than 1ms?
If I use the timings argument to cbstats, I see a histogram with gets being served mostly in the 8-16 microseconds range. Is the ep_latency_get_cmd trying to show that same information, or something completely different?
The docs at http://www.couchbase.com/docs/couchbase-manual-1.8/cbstats-all-bucket-in... aren't detailed enough.
Thanks.
Can you expand on what you mean by "histogram"? The value returned is an integer, and not a list of integers that might represent buckets of timing ranges.
As you can see from the output below, the ep_latency_get_cmd value is identical to the number of cmd_gets that have been run. I can't tell if this means it took 1ms and that's the best accuracy that can be achieved (doubtful), or if what you are saying is true, 100% of the calls fell into a certain bucket, whose definition of the range of times is undefined.
$ /opt/couchbase/bin/cbstats localhost:11210 -b jobs-bps all | egrep "ep_latency_get_cmd|cmd_get" ; sleep 60 ; /opt/couchbase/bin/cbstats localhost:11210 -b jobs-bps all | egrep "ep_latency_get_cmd|cmd_get" cmd_get: 54358467 ep_latency_get_cmd: 54358467 cmd_get: 54367855 ep_latency_get_cmd: 54367855
Ok. I've looked at this a little bit closer and ep_latency_get_cmd is the same as cmd_get. For some reason this stat just displays the number of gets that have been recieved by the server. Clearly this labeling is not very good. What I was talking about was the timing stats and there is histogram in those stats which also happens to be labeled "get_cmd". You can see this histogram by running the command below and this will tell you how long your gets are taking.
/opt/couchbase/bin/cbstats localhost:11210 -b jobs-bps timings
Also, please note that we removed the ep_latency_get_cmd stat in 2.0.
cmd_get is the amount of gets that have been issued to the server and it is incremented each time to send a get request. ep_latency_get_cmd is a histogram that displays how long it takes the server to service those get requests.