2

After configuring application package with:

    <proton>
        <tuning>
            <searchnode>
                <requestthreads>
                    <search>88</search>
                    <persearch>4</persearch>
                </requestthreads>
            </searchnode>
        </tuning>
    </proton>

And app package activation requests searchnode restart due to: Num searcher threads\nproton.numsearcherthreads has changed

The change is not being reflected when: vespa-get-config -n vespa.config.search.core.proton | grep search, the output is:

search.mmap.advise "NORMAL"
search.memory.limiter.maxthreads 0
search.memory.limiter.mincoverage 1
search.memory.limiter.minhits 1000000
numsearcherthreads 64
numthreadspersearch 1
distribution.searchablecopies 1

Is this a bug ? where can it be validated that the threads being used are indeed what was configured?

S.Glick
  • 63
  • 3

1 Answers1

1

Changing thread counts is a change that requires a restart of the searchnode process - the prepare/deploy feedback will tell you which services needs to be restarted.

Jo Kristian Bergum
  • 2,984
  • 5
  • 8
  • this is after restart – S.Glick Sep 07 '22 at 07:30
  • 3
    I suggest that you use `vespa-config-status` to check if all services are using the latest/expected config generation. The reason you are not getting expected output from `vespa-get-config` is that you have not supplied the '-i' option (use something like `vespa-configproxy-cmd| grep proton` to get correct value to use for '-i`option). – hmusum Sep 07 '22 at 07:36
  • I've updated the doc with an example of how to do what @hmusum suggests in https://docs.vespa.ai/en/operations/config-proxy.html#inspecting-config – Kristian Aune Sep 08 '22 at 11:20