Evaluate WAL journal mode for improved tracker performance

[josecelano]

Summary

This PR tracks the ongoing evaluation of SQLite's WAL (Write-Ahead Logging) journal mode for the production tracker database. The goal is to measure performance impact under high-traffic conditions (~8,000-17,000 UDP requests/sec).

Closes #86

Current Status

Phase 1: ✅ Baseline metrics collected (Jan 13-28, 2026)
Phase 2: ✅ Switched to WAL mode (Jan 28, 2026, 13:16 UTC)
Phase 3: ⏳ Awaiting comparison data (Feb 4-19, 2026)
Phase 4: ⏳ Analysis and decision (Feb 19, 2026)

Changes

• Added Phase 1 baseline performance data (15 days):
  • UDP4 announces: ~2,400-3,200 req/sec
  • Average announce time: 23.6 µs (consistent)
• Added comprehensive experiment documentation in docs/issues/86-*.md
• Successfully switched production tracker from DELETE to WAL journal mode
• Documented all execution steps and results

Technical Details

WAL Mode Switch (Phase 2)

• Date: January 28, 2026, 13:16 UTC
• Downtime: ~60 seconds
• Command: PRAGMA journal_mode=WAL;
• Files created: sqlite3.db-wal (0 bytes), sqlite3.db-shm (32K)
• Status: Tracker operational, processing normally

Database Profile

• Size: 17 GB
• Traffic: ~2,500 req/sec average, peaks at 3,200 req/sec
• No low-traffic periods (24/7 constant load)

Next Steps

1. Wait 7 days for meaningful data collection in WAL mode
2. Export metrics from Grafana (Feb 4-19, 2026):
   • UDP4 announces per second
   • UDP average announce time
3. Compare Phase 3 data with Phase 1 baseline
4. Decide whether to keep WAL mode based on results
5. Update Torrust Tracker Deployer recommendations

Testing

• ✅ Health check passed: {"status":"Ok"}
• ✅ Tracker logs show normal operation
• ✅ WAL files created successfully
• ✅ Processing ~2,500 requests/sec

Notes

• This PR should remain as draft until Phase 4 (analysis) is complete
• The production tracker is currently running in WAL mode
• Related research: Research database backup strategies torrust-tracker-deployer#310