Make metrics aggregation opt-in for batches where sink supports multiple timestamps

[moschroe]

A note for the community

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request
• If you are interested in working on this issue or have submitted a pull request, please leave a comment

Use Cases

Bulk import of existing data. There are log files of legacy systems/impromptu solutions in csv format.

Attempted to import into influxdb, which completely annihilated any granularity due to events from file occurring at high rate and being aggregated by wall clock time.

Attempted Solutions

batch.max_events: 1 did faithfully import all events, although at significant performance penalty.

Proposal

While keeping aggregate mode as default, document it, and provide the option to keep all events unaltered until batch criteria (like event count) are met.

Alternatively, batching by timestamp should regard the event time, so that, for example, 15s aggregations of 1/s metrics could be processed at 1000 events/s from a file, with the resulting metrics having 15s granularity. Ingesting an event backlog and aggregating by current wall clock time is meaningless.

But this would not solve any case where many events with high temporal resolution should be processed verbatim.

This should be done for any sink that allows multiple values per batch with individual timestamps.

References

• Document that metric sinks aggregate metrics within a batch #7938

Version

vector 0.32.2 (x86_64-unknown-linux-gnu)

[StephenWakely]

We've just changed the Prometheus Remote Write sink so you can configure it to not aggregate within a batch #18676. We should do the same when we update the Influx metrics sink #9261.

[StephenWakely]

We need to do an audit of sinks to ensure this, but it looks like currently InfluxDB is the only sink that does aggregation of metrics. This needs to be done as part of #19102.