chore: reduce LOC by ~20% — drop dead benchmark/perf/unit-test infra

[iskakaushik]

Summary

• −20.6% code LOC (13,556 → 10,766, net −3,674 file bytes across 35 files)
• Removes entirely dead/duplicative infra; no functional changes to the extension

What's deleted

Benchmark + perf infra (loadgen package, perf Docker rig, entrypoints):

• benchmark/ (uv-managed Python loadgen pkg, docker-compose, init SQL)
• scripts/bench-otel.sh, scripts/run-perf.sh
• docker/perf.Dockerfile, docker/perf-entrypoint.sh, docker/docker-compose.perf.yml
• CI job bench-otel, mise task perf

C++ unit tests (gtest harness):

• test/unit/arrow_batch_test.cc (522 LOC) + test/unit/hostname_test.cc (65 LOC)
• gtest removed from vcpkg.json
• CI job unit-tests, mise task test:unit

Overlapping TAP tests:

• 028/029/030 query_normalization variants folded into 027 as subtests (preserves all unique assertions: error-path leak, escape_string_warning dedupe, nested SPI normalization)
• Dropped 003_buffer_overflow (covered by overflow_race.spec + 017_toctou_race.pl)
• Dropped 013_buffer_metrics (ClickHouse-gated tautologies; buffer capture is already in buffers.sql)
• Dropped 014_cpu_metrics (no unique signal beyond >= 0 checks)
• Dropped 019_reset_atomicity (reset during concurrent writes covered by overflow_race + ring_buffer_concurrent specs)

Kept (unique coverage the specs can't reproduce):

• 017_toctou_race.pl — double-checked-locking race under real pgbench contention
• 022_overflow_deadlock.pl — emit_log_hook → PschEnqueueEvent re-entry deadlock

Test plan

• cmake --build build clean against PG 18.3 local install (TAP-enabled)
• Regression suite (10/10 pass): basic, version, guc, stats, utility, buffers, cmd_type, client_info, error_capture, drop_database_barrier
• Isolation suite (3/3 pass): ring_buffer_concurrent, ring_buffer_boundary, overflow_race
• CI green (will verify after push)

Not included

overlay-ports/cityhash/ — discussed but deferred. The overlay patches a real aarch64 config.guess bug upstream hasn't fixed, and clickhouse-cpp pulls cityhash transitively via vcpkg. Full submodule vendoring would need an additional clickhouse-cpp port patch to drop the transitive dep, which adds LOC rather than removing it. Happy to do this as a follow-up if you want a specific shape.

[Copilot]

Pull request overview

Removes dead/duplicative benchmark, perf, and unit-test infrastructure to substantially reduce LOC while keeping the extension’s runtime behavior unchanged.

Changes:

• Drops perf/benchmark Docker + scripts and the Python benchmark/loadgen package.
• Removes opt-in C++ gtest unit tests and related build/CI/task wiring (including gtest vcpkg dep).
• Consolidates overlapping TAP query-normalization tests into t/027_query_normalization.pl and deletes redundant TAP files.

Reviewed changes

Copilot reviewed 33 out of 35 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
vcpkg.json	Removes gtest dependency now that C++ unit tests are dropped.
test/unit/hostname_test.cc	Deletes gtest-based hostname unit test.
test/unit/arrow_batch_test.cc	Deletes gtest-based Arrow batch unit test.
t/030_nested_query_normalization.pl	Removes standalone nested normalization TAP test (folded into t/027).
t/029_query_normalization_warning.pl	Removes standalone warning-dedup TAP test (folded into t/027).
t/028_query_normalization_error_leak.pl	Removes standalone error-path leak TAP test (folded into t/027).
t/027_query_normalization.pl	Adds new subtests to absorb coverage from removed 028/029/030 variants.
t/019_reset_atomicity.pl	Deletes TAP reset atomicity test deemed redundant with isolation specs.
t/014_cpu_metrics.pl	Deletes CPU metrics TAP test.
t/013_buffer_metrics.pl	Deletes buffer/WAL metrics TAP test.
t/003_buffer_overflow.pl	Deletes overflow TAP test deemed redundant with isolation/spec coverage.
scripts/run-perf.sh	Removes perf benchmark runner script.
scripts/bench-otel.sh	Removes OTel throughput benchmark script.
mise.toml	Removes test:unit and perf tasks.
docs/guides/testing.md	Updates testing guide to remove deleted TAP tests from the list.
docker/perf.Dockerfile	Removes perf benchmark Docker image build.
docker/perf-entrypoint.sh	Removes perf benchmark container entrypoint.
docker/docker-compose.perf.yml	Removes perf docker-compose stack.
benchmark/loadgen/uv.lock	Deletes lockfile for removed Python load generator.
benchmark/loadgen/src/loadgen/workloads.py	Deletes workload definitions for removed load generator.
benchmark/loadgen/src/loadgen/runner.py	Deletes runner for removed load generator.
benchmark/loadgen/src/loadgen/cli.py	Deletes CLI for removed load generator.
benchmark/loadgen/src/loadgen/__init__.py	Deletes package init for removed load generator.
benchmark/loadgen/pyproject.toml	Deletes Python project definition for removed load generator.
benchmark/loadgen/README.md	Deletes documentation for removed load generator.
benchmark/loadgen/.python-version	Deletes Python version pin for removed load generator.
benchmark/loadgen/.gitignore	Deletes subproject ignore rules for removed load generator.
benchmark/init/01-create-extension.sql	Deletes benchmark DB init SQL.
benchmark/docker-compose.yml	Deletes benchmark docker-compose stack.
benchmark/README.md	Deletes benchmark environment documentation.
benchmark/.env.example	Deletes benchmark env template.
META.json	Removes benchmark from no_index directories after benchmark removal.
CMakeLists.txt	Removes PSCH_BUILD_UNIT_TESTS option and unit test targets.
.gitignore	Drops ignores related to removed unit/perf/benchmark infra.
.github/workflows/ci.yml	Removes CI jobs for unit tests and perf benchmark.

---

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

[Copilot]

In the new error-path leak subtest, the assertion that the error event exports an empty query can pass even if the error event was never exported/inserted into ClickHouse (the SELECT may return no rows, yielding an empty string). Add an explicit wait/assert that at least one error row exists (e.g., poll count() on err_message != '') and/or wait for a minimum exported count before checking query == '' so the test actually validates the behavior and isn’t timing-dependent.