[processor/tailsamplingprocessor] Policy on span_count & min_spans

[yehaotian]

Component(s)

processor/tailsamplingprocessor - span_count

Is your feature request related to a problem? Please describe.

For now tail sampling processor policy on span_count only have min_spans to sample on traces contains minimal number of spans.
It is good to have max_spans to sample traces contains maximum number of spans, to help filter out giant traces. Usually we find out the giant traces are not helpful and takes a lot of data volume quota.

Describe the solution you'd like

Something similar

policies:
      [
# sample traces with maximum 20 spans.
         {
            name: test-policy-10,
            type: span_count,
            span_count: {max_spans: 20}
         },
# or not sample traces with more than 20 spans
         {
            name: test-policy-11,
            type: span_count,
            span_count: {min_spans: 20, invert_match: true}
         },

...
      ]

Describe alternatives you've considered

No response

Additional context

No response

[github-actions]

Pinging code owners:

• processor/tailsampling: @jpkrohling

See Adding Labels via Comments if you do not have permissions to add labels yourself.

[mizhexiaoxiao]

Seems reasonable to me.
Assign me if possible, i should be able to work on it

[atoulme]

@mizhexiaoxiao it's yours!

[yehaotian]

Thanks for the help!

[jpkrohling]

# sample traces with maximum 20 spans.
         {
            name: test-policy-10,
            type: span_count,
            span_count: {max_spans: 20}
         },
# or not sample traces with more than 20 spans
         {
            name: test-policy-11,
            type: span_count,
            span_count: {min_spans: 20, invert_match: true}
         },

Aren't those equivalent? Why not just have one of min_spans/max_spans? Or remove the invert match?

[yehaotian]

I mentioned or in the comment, either way works for me

[jpkrohling]

I think just a "num_spans" or just "num" should be fine. I know we use invert_match elsewhere, but perhaps we could use a different name? Like, operation=min?

[mizhexiaoxiao]

I think just a "num_spans" or just "num" should be fine. I know we use invert_match elsewhere, but perhaps we could use a different name? Like, operation=min?

I think adding max_spans configuration is enough. Users can set the span_count number they want to sample in the range of min_spans and max_spans. Of course, you can also set only one of them. This is similar to the min_value and max_value of the numeric_attribute policy

Changing the configuration name may not be friendly to users who are using the span_count policy, but adding max_spans will not affect it.

Possible configurations are as follows

# sample traces with minimum 5 spans.
{
    name: test-policy-1,
    type: span_count,
    span_count: {min_spans: 5}
}

# sample traces with maximum 20 spans.
{
    name: test-policy-1,
    type: span_count,
    span_count: {max_spans: 20}
}

# sample traces range of 5 and 20 spans.
{
    name: test-policy-1,
    type: span_count,
    span_count: {min_spans:5, max_spans: 20}
}

@jpkrohling What's your thoughts on this?

[jpkrohling]

This looks cleaner and users would not be confused about the semantics.

[jiaojialin]

hello, for this feature, looks very straightforward but useful, could we please help out to get it out asap? Thanks so much!

[yehaotian]

@mizhexiaoxiao @jpkrohling @TylerHelmuth Thank you all for the work!
Is there a way I can track the release on this? We are currently using opentelemetry-collector-contrib-dev:latest and would like to use the updated policy within a week.

[jpkrohling]

It will be available on the next release, scheduled for the week of 2023-03-20. If you are building your own distribution, you can use it already based on the main's current state.