[Marvel] What does the Search Rate mean?

monitoring

(Richard Leonard) #1

I am quite confused about this graph shown in marvel...

Actually, I don't know which one the Search Rate refers to
(considering I have 5 shards, 1 replica, totally 10 shards saved on 2 machines)
if I triggered 100 SEARCH requests per second using RESRful API
what the search rate will be?

  1. 100 / s (total search requests to the whole system)
  2. 500 / s (sum up all the requests on every shards, I think every search request will need infos on every shard)
  3. 1000 / s (case 2 with 1 replica?)

Anyone could help?
I am also confused on the Indexing Rate, Search Latency and Indexing Latency....


(Mark Walkom) #2

It's the rate of search requests being executed against the cluster.
Index rate is the number of indexing requests being made against the cluster.

Search and indexing latencies; I am not sure exactly what that is.


(Richard Leonard) #4

If it refers to requests towards the cluster.
I've made a trivial test. I've riggered 50000 Random SEARCH requests, all completed in 300s. So the Search rate should be 166/s ?
The graph on marvel seems to show a far more larger value than 166/s....


(Mark Walkom) #5

We're those requests spread over 300s, or you just issued them and they happened to finish in 300 s?


(Richard Leonard) #7

Pattern Like this:
Request, Get Response, Request, Get Response....
50000 x (Request, Get Response) takes 300s


(Yang Liu) #8

I'd also like to know how Search Rate is calculated. Basically it's much higher than our actual QPS. I even saw negative numbers during deployment.


(Ronald Plöger) #9

Hi, we would also like to know how the search rate displayed by marvel is calculated. We see a search rate of about 1,700 - 1,800 per second in Marvel and measure just over 500 request per second fired from our apps towards our elastic cluster.


(Yang Liu) #10

Seems that nobody would give an answer.

I summarized from my experience that Search Rate is the lookups running per second multiplied by the number of shards.
And the Search Latency means the search latency within each shard.

I have a brief write-up here. You may verify that with your index and mapping settings.


(system) #11