fix: flaky localdns exporter e2e validation

[jingwenw15]

Summary

• Replace echo "$VAR" | grep/awk/sed pipes with grep/awk/sed <<< "$VAR" here-strings throughout the exporter validation script
• Capture ss -tln output once into a variable (SS_LISTEN_OUTPUT) instead of calling ss multiple times
• Replace echo "$METRICS" | grep ... | head -10 with awk one-liners that exit after 10 matches

Why: With set -euo pipefail enabled, echo "$LARGE_VAR" | grep -q ... can fail spuriously. If grep finds its match and exits early, echo receives SIGPIPE and returns a non-zero exit status, which pipefail treats as a command failure. Here-strings (<<<) avoid the pipe entirely — the shell passes the variable directly to the command's stdin via a temporary file, so there is no writer process to receive SIGPIPE.

This was the root cause of intermittent test failures in the exporter e2e where metrics were present but the validation still failed.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Fixes flakiness in the LocalDNS exporter E2E validation script by avoiding pipefail-related false negatives when grepping large metrics payloads.

Changes:

• Cache ss -tln output and reuse it for port checks and listen address detection.
• Replace echo "$VAR" | grep ... patterns with here-strings (grep ... <<< "$VAR") to avoid SIGPIPE/pipefail failures.
• Minor refactors to avoid pipelines (e.g., head, grep | sed) when inspecting $METRICS.

[yewmsft]

The here-string conversion is a sound fix for the pipefail/SIGPIPE rationale in the inline comment — grep <<<"$VAR" keeps grep's stdin a regular file descriptor instead of a pipe that grep can close early on first match. Caching the ss -tln output once per check is a nice bonus that also removes a class of TOCTOU between the TCP/UDP probes.

Two small suggestions:

1. Some of the swaps (e.g. head -n 1 → sed -n '1p') are pure style and don't change pipefail semantics — fewer churn lines = easier bisect if this validator regresses again. Worth keeping the diff scoped to the actual fix.

2. Add a brief assertion (even just a shell snippet in a comment, or a tiny shellspec case if there's a harness nearby) that exercises the cached-ss path with a large fake payload, so a future well-meaning refactor that reverts to echo … | grep gets caught instead of going flaky again under load.

[yewmsft]

stop re-review