Before & After Each Optimization (Apache, PHP-FPM & HAR)

Act as a senior Laravel, PHP-FPM, and Apache performance engineer. Audit only performance measurement in HolidayLandmark and propose a repeatable before/after process for every improvement step.

Enter Plan Mode first. Do not change code, logging configuration, or server processes until I review and explicitly approve your proposed approach. Begin with read-only inspection.

Read the project’s AGENTS.md and CLAUDE.md. Locate the performance report and identify every page and user action listed in section 3. If that section is unavailable, clearly identify the missing action inventory instead of inventing it.

Inspect these four areas separately:

1. Apache request timing:
   - Inspect active virtual hosts, access-log formats, and existing timing fields.
   - Verify whether %D is recorded and explain that it reports request duration in microseconds.
   - Propose a dedicated timing format where needed, including timestamp, method, safe request path, status, duration, and a correlation ID where supported.
   - Avoid logging credentials, tokens, cookies, or sensitive query parameters.
   - Identify which logs capture browser requests and internal service calls.

2. PHP-FPM slow logs:
   - Inspect every relevant pool’s slowlog, request_slowlog_timeout, permissions, and existing slow-log entries.
   - Propose appropriate thresholds and log locations based on observed request duration.
   - Explain how stack traces help locate slow PHP execution and upstream waits, and that slow logs are not a complete profiler.
   - Keep slow-log thresholds separate from request_terminate_timeout; do not introduce request termination as a measurement change.
   - Explain overhead, retention, required reloads, and rollback.

3. Browser HAR captures:
   - Define a repeatable capture procedure for every action in section 3.
   - Record browser/version, user role, dataset size, filters, pagination, network conditions, and cache state.
   - Separate initial page load from subsequent Livewire actions.
   - Measure request count, transferred bytes, relevant request duration, and time until the UI becomes usable.
   - Capture cold and warm runs separately and use consistent recording boundaries.
   - Sanitize HAR files before sharing; remove authorization headers, cookies, tokens, personal data, and sensitive request/response bodies.

4. Before/after comparison:
   - Establish a baseline before each approved optimization and repeat the same actions afterward.
   - Change one improvement group at a time so results can be attributed.
   - Use repeated runs and report sample count, median, and variability; report p95 only when the sample size supports it.
   - Correlate HAR requests, Apache timing, FPM slow traces, and upstream timings where available.
   - Do not add parallel request durations together as if they were total user waiting time.
   - Record errors, timeouts, and functionality regressions alongside speed improvements.
   - Distinguish measured results from estimates and disclose missing evidence.

For every measurement gap, show:

- Affected service, page, and user action.
- Exact configuration or source file paths and current line numbers.
- Existing logging configuration or available evidence.
- What cannot currently be measured and why.
- Clearly labelled proposed before/after configuration or commands—do not apply them.
- Risks, overhead, validation, and rollback steps.

Finish with:

- A complete action-by-action measurement checklist.
- A results table template containing action, test conditions, sample count, before/after median time, request count, transferred bytes, errors, and percentage change.
- A prioritized observability setup plan.
- Clear criteria for accepting an improvement or investigating a regression.

Stop and wait for my approval. After approval, implement only the agreed measurement setup, capture the baseline, and repeat measurements after each separately approved performance change. Report changed files, executed commands, sanitized evidence locations, and actual results. Do not implement unrelated optimizations or fabricate measurements.