Harden exporter state and add follow-up TAP tests

[iskakaushik]

Summary

• clear the shared bgworker PID on shutdown and treat stale flush targets as "background worker not running"
• guard exporter failure stats before the exporter is initialized and use UINT64_FORMAT in shutdown logging
• add TAP coverage for stale bgworker flush handling and squashed IN-list normalization
• exclude benchmarking-related local work from this PR

Testing

• ./scripts/run-tests.sh ../postgres/install_tap tap 009
• ./scripts/run-tests.sh ../postgres/install_tap tap 031

[Copilot]

Pull request overview

Hardens pg_stat_ch’s background worker state handling (especially around stale PIDs and shutdown) and adds TAP coverage for new query-normalization behavior introduced in PostgreSQL 18.

Changes:

• Clear stored bgworker PID on worker shutdown and treat stale flush signal targets as “background worker not running”.
• Guard exporter failure-stat access before exporter initialization; improve shutdown logging format for 64-bit counters.
• Add TAP tests for stale bgworker flush behavior and PG18+ “squashed IN-list” query normalization.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
t/031_squashed_in_list_normalization.pl	Adds PG18+ TAP coverage for squashed IN-list normalization output.
t/009_bgworker.pl	Adds TAP test ensuring flush reports “bgworker not running” (not stale-PID signal failure) after worker exit.
src/worker/bgworker.cc	Clears stored bgworker PID on exit; improves flush signaling warnings and stale-PID handling.
src/queue/shmem.cc	Uses UINT64_FORMAT for shutdown counter logging.
src/export/stats_exporter.cc	Guards consecutive-failure access when exporter isn’t initialized.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

In the kill() failure path, only ESRCH is treated as a stale PID. If the bgworker PID is reused by a non-Postgres process owned by another user, kill() can fail with EPERM; in that case the shared bgworker_pid remains set and pg_stat_ch_flush() will keep warning and never recover. Consider additionally clearing the stored PID when the PID is no longer a Postgres backend (e.g., BackendPidGetProc/IsBackendPid check) and treating that as “background worker not running”, instead of leaving a permanently stale PID.

[Copilot]

This assertion doesn’t actually prove the IN-list was squashed to a single placeholder: checking that the query does not contain "$5" can still pass if the list incorrectly expands to only a few placeholders (e.g., $1..$4). It would be more robust to assert that the IN(...) clause contains exactly one $N placeholder plus the squashed comment marker (similar to the stricter regex used in the earlier numerics subtest).

Suggested change

	# The normalized form should be compact, not contain 20 separate placeholders
	unlike($q, qr/\$5/, 'No excessive placeholders (list is squashed, not expanded)');
	# The normalized IN-list should contain exactly one placeholder plus
	# the squashed comment marker, not an expanded series of placeholders.
	like($q, qr/WHERE\s+id\s+IN\s*\(\s*\$\d+\s+\/\*, \.\.\. \*\/\s*\)/,
	'IN list is squashed to a single placeholder');

[Copilot]

This regex can match placeholder numbers as well (e.g., "$7") because word boundaries exist between "$" and a digit. That can make the test fail even when normalization is correct if placeholder numbering ever changes. Consider tightening the pattern to specifically target raw constants from the original IN list (e.g., match digits that aren’t immediately preceded by '$') or assert the exact normalized NOT IN form instead.

Suggested change

	unlike($q, qr/\b[7-9]\b/, 'NOT IN constants removed');
	unlike($q, qr/(?<!\$)\b[7-9]\b/, 'NOT IN constants removed');