[amalsaidov]

What this PR does:

MergeSlicesParallel merges the per-ingester label name and label value slices concurrently. The distributor calls it with parallelism 8 on every LabelValues and LabelNames query, so the slices are split into several batches merged by separate goroutines.

The function keeps only the first error, over a one-slot channel, but it sent to that channel with a blocking send:

if e != nil {
    errCh <- e   // blocks once the channel is full
    wg.Done()
    return
}

When a query is canceled, more than one batch can see the canceled context and return an error at the same time (each batch checks ctx.Err() once it has merged CheckContextEveryNIterations = 128 elements). The first send fills the channel, the receiver takes that error and returns, and every other worker that also errored blocks forever on its send. The goroutine that waits on the WaitGroup before closing the channels then never returns either, so the whole set leaks for the lifetime of the process. Canceled or timed-out label queries are routine, so this leaks goroutines steadily in production.

The fix sends the error without blocking, so a worker that cannot report its error drops it and exits. The first error is still the one returned, and the successful path is unchanged.

The new regression test cancels the context up front and uses goleak to assert nothing is left behind. It fails before the change (two workers parked on chan send plus the WaitGroup waiter) and passes after.

Which issue(s) this PR fixes:

No issue filed; found while reading the label-query merge path.

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated
• docs/configuration/v1-guarantees.md updated if this PR introduces experimental flags

---

AI-assisted; directed, reviewed and tested by the author (amalsaidov).