[exporter/elasticsearch] Truncate timestamp to millis for metrics to avoid TSDB grouping issue

[carsonip]

Component(s)

exporter/elasticsearch

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

See known issues in ES exporter readme and #37150 about version conflict in TSDB due to nanoseconds precision in metrics.

Describe the solution you'd like

As discussed in #37150 (comment) , it is better for ES exporter to handle this OOTB to truncate timestamp to milliseconds.

Scope is not decided yet - whether it is 1. for metrics only and 2. for OTel mapping mode only.

Describe alternatives you've considered

No response

Additional context

No response

[carsonip]

/label -needs-triage

[github-actions]

Pinging code owners:

• exporter/elasticsearch: @JaredTan95 @carsonip @lahsivjar

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

[felixbarny]

I think it should be for metrics only. IMO, we should also not modify the source events (so that we can use consumer.Capabilities{MutatesData: false}) but instead adjust the grouping hash and the timestamp serialization. We may also want to include a mode that emits timestamps in nanos precision for metrics. But I think that can come later in case there are users that need that. I'd assume that for the vast majority of metrics use cases, millisecond precision is fine. For reference, Prometheus also only supports ms precision timestamps.

[carsonip]

As discussed, this should be good to close, as workaround for majority of version conflict errors is stated in #37150

It is also useful to surface TSDB issue due to millis instead of dropping data in es exporter.