topic Bucket/Timechart and Dedup in Splunk Search

Bucket/Timechart and Dedup

jbp4444 — Tue, 17 May 2011 00:33:49 GMT

I'm trying to plot total load-avg vs number of processors in a cluster (i.e. how loaded is the system). The following basically works:

numproc OR loadshort | transaction xid | bucket span=10m _time | timechart span=10m sum(numproc) sum(loadshort)

Except we occasionally see multiple data items posted in a given 10min window -- e.g. {host,load1,numproc,time1} and {host,load2,numproc,time2} both land in the same time-bucket. The above commands adds in the numproc value twice, which obscures the real load on the system.

Is there a way to dedup the data after its been bucketed? or, maybe said another way, to dedup the data within a single bucket?

Re: Bucket/Timechart and Dedup

gkanapathy — Tue, 17 May 2011 02:15:24 GMT

Might be easiest to just use first() instead, which will give you the most recent numproc in each bucket. I'll assume numproc doesn't change, though if it did, you might just use avg():

numproc OR loadshort | transaction xid | bucket span=10m _time | timechart span=10m first(numproc) sum(loadshort)

Re: Bucket/Timechart and Dedup

jbp4444 — Tue, 17 May 2011 13:04:33 GMT

Thanks for the quick reply gkanapathy -- that's definitely in the right direction. But there are multiple hosts producing the {numproc,loadshort} data, and I want to sum each item across all those machines. My understanding is that 'first' would give only one value from one host.

Maybe some combination of 'first .. by host' then a separate summation command?

Re: Bucket/Timechart and Dedup

jbp4444 — Tue, 17 May 2011 13:28:18 GMT

Looks like dedup might work after all -- I didn't realize you could dedup based on more than one field:

numproc OR loadshort | transaction xid | bucket span=10m time | dedup host time | timechart span=10m sum(loadshort) sum(numproc)

Since bucket discretized the timestamps, the {host,time} pairs are duplicates and dedup can take care of them (time should be underscore-time).