fix(otelconf): report Prometheus HTTP server errors in v0.2.0 - #9653

Problem

go.opentelemetry.io/contrib/otelconf/v0.2.0 starts an HTTP server for the Prometheus pull exporter in a goroutine. The error check on server.Serve(lis) is inverted: a clean shutdown (http.ErrServerClosed) is reported through otel.Handle as "the Prometheus HTTP server exited unexpectedly", while every genuine Serve failure (accept errors, TLS/listener problems, etc.) is silently dropped. Operators therefore get a spurious error on every shutdown and no signal at all when the metrics endpoint actually stops serving.

Root cause

otelconf/v0.2.0/metric.go:369:

if err := server.Serve(lis); err != nil && errors.Is(err, http.ErrServerClosed) {

The condition is missing the negation. The sibling implementations in otelconf/v0.3.0/metric.go and otelconf/x/metric.go already use !errors.Is(err, http.ErrServerClosed).

Fix

• Negate the predicate (err != nil && !errors.Is(err, http.ErrServerClosed)), so the goroutine in prometheusReader is now identical to the one in v0.3.0 and x.
• Add a CHANGELOG.md entry under Unreleased / Fixed.

No public API changes.

The first revision extracted the check into a handleServeErr helper with a table test; it was inlined again in e769464 at review request so the three implementations stay the same.

How tested

The handler is the process-global otel.Handle, so this branch has no isolated unit test (see the discussion on #9097); the fix is the same code that v0.3.0 and x already run.

From otelconf/: go build ./..., go vet ./v0.2.0/..., go test -race ./v0.2.0/... (ok), golangci-lint run ./v0.2.0/... (0 issues).

Links

• Fixes otelconf: Prometheus metrics server errors are silently ignored #9056
• Supersedes fix(otelconf): surface Prometheus metrics server errors in v0.2.0 #9097 (closed by its author).

This change was prepared with an AI agent operated by KR-Ravindra, who reviewed and tested it.