What is Search request rate panel in Marvel?

monitoring

(Vikas Gopal) #1

Hi Experts ,

I have an index and it has 43.5 million documents in 7 shards , I am monitoring marvel to check performance and this is what I got in marvel

Couple of queries

  1. What 4.500 per 1s search request rate means ? is it good or bad ?
  2. I got this result after 4 min of kibana dashboard , I mean when I open dashboard i got marvel result after 4 min , why marvel is taking 4 min to show all metrics ?
  3. Marvel IOPS never increases it goes max 2.5 when I open kibana dashboard .Is it because due to no data ingestion into index , it is static i am just doing search on top of it ?As per my understanding IOPS should increase when I open dashboard in kibana.

my Environment

Core :- 8
RAM :- 64 Gb
No of nodes :- 2(one each server)
ES instance RAM :- 9 gb each


(Chris Earle) #2
  1. What 4.500 per 1s search request rate means ? is it good or bad ?

The search request rate deals with the number of search threads that your node(s) are seeing for each time-bucket.

This is different from the number of search requests that your cluster is receiving. It's actually a fuzzy multiplier of the number of raw requests that you're receiving:

For example, imagine that you have an index that has 5 shards (the default). Each shard is going to receive a search request, so that's 5 search requests. If you query the index twice, that's 10. If you query two indices with 5 shards in the same request, that's still 10. Now if you have one index with 2 shards and another with 5 shards, then searching both will be 7 search requests.

The fact that you see a .5 in the response is just the nature of averaging out the response to make it more usable from the display. Remember that Marvel is only able to collect stats every 10 seconds (by default), so it can miss things that only live between the polling intervals (like Open Search Contexts).

So, the fact that it's 4.5 means that not much is happening. It's also important to note that, if Marvel is running in the same cluster as your data, then its own charts will cause search requests and the monitoring aspect of Marvel will cause data to be indexed as well.

I got this result after 4 min of kibana dashboard , I mean when I open dashboard i got marvel result after 4 min , why marvel is taking 4 min to show all metrics ?

This doesn't sound normal. Either something wasn't started yet (like the monitoring cluster) or the connection was blocked for 4 minutes.

Marvel IOPS never increases it goes max 2.5 when I open kibana dashboard .Is it because due to no data ingestion into index , it is static i am just doing search on top of it ?As per my understanding IOPS should increase when I open dashboard in kibana.

Your indexing rate looks low, so perhaps you're barely doing anything with the IOPS. The nice thing about search is that it uses the file system cache when it can, so it has only a very mild impact on IOPS.


(system) #3