otelriver: Track inserts skipped as duplicate due to UniqueOpts

[echatman]

What

Adds a skipped_as_duplicate boolean attribute to river.insert_count data points and a duplicate_skipped_count span attribute on the river.insert_many span, so callers can tell how many of the submitted jobs were dropped by UniqueOpts vs. actually enqueued.

After this change, each InsertMany call emits two river.insert_count data points:

• river.insert_count{skipped_as_duplicate:false} — jobs that were actually enqueued
• river.insert_count{skipped_as_duplicate:true} — jobs dropped because UniqueOpts matched an in-flight or pending duplicate

The sum across both data points still equals len(manyParams), so dashboards or monitors that sum the metric without filtering see no change. The span also carries the count directly:

span.SetAttributes(attribute.Int64("duplicate_skipped_count", skipped))

Why

Today every job submitted to InsertMany is counted in river.insert_count regardless of whether River accepted it or dropped it via UniqueOpts. That makes the insert metric unable to distinguish "we accepted 100 new jobs" from "we accepted 0 new jobs because all 100 were dedup'd against in-flight work" — both look identical on a chart.

I've been carrying a custom River middleware that emits a separate …insert_duplicate_skipped_count counter just to recover this information. Pushing this into otelriver removes the need for that workaround and gives every otelriver user duplicate-rate visibility out of the box.

Tests

Existing InsertMany, InsertManyError, and InsertManyPanic tests updated to assert against the new partitioned data points. Two new tests cover the new behaviour:

• InsertManyAllDuplicates — 2 submitted, both flagged UniqueSkippedAsDuplicate: true
• InsertManyMixedDuplicates — 5 submitted, 2 dedup'd / 3 enqueued; also asserts the grand-total invariant

A small test helper requireSumByAttrs was added (alongside the existing requireSum) for asserting against a specific subset of data points by attributes, since the partitioned metric now emits multiple data points per measurement.

make test and make lint both pass; tests also pass under -race.

Notes for reviewers

• Sibling PR — same area, independent change: otelriver: treat JobSnoozeError as flow control, not a failure #59 (otelriver: treat JobSnoozeError as flow control, not a failure)
• No public API changes; all additions are observability attributes.

---

Made with Cursor on behalf of @echatman.

[brandur]

Could you modify this and the two instances below to prefix with unique_* like the row field name? I kind of like that unique_* is on there because for someone not familiar with the property already it's strongly suggestive that it's related to unique jobs.

[echatman]

Sure. I was trying to keep it short but already failed at that, so may as well go with fully verbose for clarity. Fix incoming.

[echatman]

Done. I also added a CHANGELOG entry; let me know if you prefer to leave small changes like this out of the changelog and I'll remove it

[brandur]

Thx!

Nah, every change should ideally have a changelog entry. I didn't mention it yet because I knew it'd conflict with the addition in #59 ;)

With #59 in now, can you rebase and re-push?

[echatman]

done

[brandur]

Published v0.9.0.