Bulk "Service time" - total or per request

[boris.toninski]

Is "Service time" for bulk operation from the report the total time needed for insert requests to Elastic search? The reference says "period between start of request processing and receiving the complete response", so I'm a little confuse is it the time for only one request or for all of them.

[dliappis (Dimitrios Liappis)]

Hi Boris,

When Rally creates stats for the bulk operation, each measurement (code here) is done per bulk API call and for each of those the time taken==took value in the response. After the bulk operation is over, service time percentiles get calculated based on the collected stats per operation.

This part of the FAQ clarifies things a bit better: https://esrally.readthedocs.io/en/stable/faq.html?highlight=took#what-does-latency-and-service-time-mean-and-how-do-they-related-to-the-took-field-that-elasticsearch-returns

Dimitris

[boris.toninski]

Hi Dimitris,

thank you replying. It seems I will have to dive deeper to figure it out . But still let say that I'm testing with 1000 documents and bulk-size of 100, so there should be 10 bulk operations, would the "100th percentile service time" be the time spend for all of the 10 operations.

Thanks,
Boris