Prometheus Metrics:
- Add METRICS_LABELS fallback for defaultLabels env config
- Separate includeSystem from collectDefaultMetrics with own env var
- Parse URL in /metrics endpoint to support querystrings and trailing slashes
- Sort keys in labelsToKey for stable/deterministic lookups
- Update documentation with all environment variables
Retry Utils:
- Remove abort listener on sleep completion to prevent memory leak
- Apply maxDelay cap AFTER jitter to guarantee max delay limit
- Validate/normalize attempt parameter in calculateDelay
- Count retries only when attempt > 1 (actual retry, not first try)
- Use metrics.retryExhausted instead of generic errors for exhausted retries
Tests:
- Add consistency tests to verify retryConfigs.rsocket matches constants
- Re-export RETRY_BACKOFF_DELAYS and RETRY_JITTER_FACTOR from Defaults
for backwards compatibility with existing imports
- Add 'as const' to RETRY_JITTER_FACTOR for type consistency
- Improve documentation explaining why values are hardcoded in retryConfigs
(ESM initialization order issues with indirect circular imports)
Hardcode values in retryConfigs.rsocket instead of referencing
RETRY_BACKOFF_DELAYS and RETRY_JITTER_FACTOR to avoid
"Cannot access before initialization" error in ESM module loading.
- Remove duplicate RETRY_BACKOFF_DELAYS and RETRY_JITTER_FACTOR from Defaults/index.ts
- Export constants from retry-utils.ts as single source of truth
- Add 'as const' for better type inference
- Fixes "Cannot access 'RETRY_BACKOFF_DELAYS' before initialization" error
- Add registryConfigured flag to ensure single execution of configureRegistry
- Fix race condition by moving setMetricPrefix/configureRegistry before Promise
- Remove unused prefix parameter from configureRegistry function
- Consolidate prefix to single source of truth (configuredPrefix global)
- Allow clearing default labels by passing empty object
- Add resetMetricsConfiguration() and getConfiguredPrefix() for testing
- Apply configured prefix to all metric names via getFullMetricName()
- Configure customRegistry with defaultLabels at initialization
- Prevent memory leak: collectDefaultMetrics only called once via flag
- collectDefaultMetrics now uses customRegistry instead of global
- Initialize prefix/defaultLabels before creating baileysMetrics
All metrics now properly include prefix (e.g., baileys_connection_attempts)
and inherit defaultLabels configured via environment variables.
Major improvements to prometheus-metrics.ts:
- Use custom registry instead of global promClient.register (isolation)
- Remove duplicate localValues - prom-client is now single source of truth
- Fix reset(labels) to respect labels param using remove() instead of reset()
- Migrate Summary class to use prom-client internally
- getMetricsOutput() now uses customRegistry for proper isolation
Fix misleading comments in retry-utils.ts:
- Change "exponential backoff" to "custom progressive backoff"
- Fix comment claiming constants are "from Defaults" (now local)
Breaking changes to getValues()/get() - now async as they query prom-client
Define RETRY_BACKOFF_DELAYS and RETRY_JITTER_FACTOR locally in
retry-utils.ts to avoid circular dependency that caused
"Cannot access 'RETRY_BACKOFF_DELAYS' before initialization" error.
- Add prom-client ^15.1.3 as production dependency
- Migrate Counter class to use prom-client internally
- Migrate Gauge class to use prom-client internally
- Migrate Histogram class to use prom-client internally
- Update getMetricsOutput() to use prom-client's native metrics()
- Update contentType() to use prom-client's contentType
- Enable collectDefaultMetrics() for Node.js system metrics
- Maintain backward compatibility with existing API
Benefits:
- Native OpenMetrics format export
- Proper histogram bucket handling
- Default Node.js metrics (memory, CPU, event loop, GC)
- Better integration with Grafana/Prometheus ecosystem
- Integrate RETRY_BACKOFF_DELAYS and RETRY_JITTER_FACTOR into retry-utils.ts
- Add retryConfigs.rsocket using the default delays
- Add getRetryDelayWithJitter() helper function
- Add getAllRetryDelaysWithJitter() for planning
- Document DEFAULT_CACHE_MAX_KEYS with usage example
- Adjust INITIAL_PREKEY_COUNT from 30 to 200 (moderate value)
- Adjust MIN_UPLOAD_INTERVAL from 60s to 10s (balance rate limiting/responsiveness)
- Revert syncFullHistory to true (avoid breaking change)
Based on RSocket's battle-tested configuration:
- Add maxWebSocketListeners config option (default: 20)
- 8 base WS events + 10 dynamic listeners + 2 buffer slots
- Add maxSocketClientListeners config option (default: 50)
- Replace dangerous setMaxListeners(0) with configurable limits
- Add warning log if user explicitly sets limit to 0
BREAKING: Previous behavior used setMaxListeners(0) which removed
all limits. Now defaults to safe limits but can be overridden via config.
- Add DEFAULT_CACHE_MAX_KEYS with limits per store type (prevents memory leaks)
- Add RETRY_BACKOFF_DELAYS [1s, 2s, 5s, 10s, 20s] for exponential backoff
- Add RETRY_JITTER_FACTOR (0.15) to prevent thundering herd
- Change INITIAL_PREKEY_COUNT from 812 to 30 (safer for rate limiting)
- Change MIN_UPLOAD_INTERVAL from 5s to 60s (avoids rate limiting)
- Change syncFullHistory default to false (conservative approach)
Based on RSocket fork's production-tested configuration.
- Integrate RETRY_BACKOFF_DELAYS and RETRY_JITTER_FACTOR into retry-utils.ts
- Add retryConfigs.rsocket using the default delays
- Add getRetryDelayWithJitter() helper function
- Add getAllRetryDelaysWithJitter() for planning
- Document DEFAULT_CACHE_MAX_KEYS with usage example
- Adjust INITIAL_PREKEY_COUNT from 30 to 200 (moderate value)
- Adjust MIN_UPLOAD_INTERVAL from 60s to 10s (balance rate limiting/responsiveness)
- Revert syncFullHistory to true (avoid breaking change)
Based on RSocket's battle-tested configuration:
- Add maxWebSocketListeners config option (default: 20)
- 8 base WS events + 10 dynamic listeners + 2 buffer slots
- Add maxSocketClientListeners config option (default: 50)
- Replace dangerous setMaxListeners(0) with configurable limits
- Add warning log if user explicitly sets limit to 0
BREAKING: Previous behavior used setMaxListeners(0) which removed
all limits. Now defaults to safe limits but can be overridden via config.
- Add DEFAULT_CACHE_MAX_KEYS with limits per store type (prevents memory leaks)
- Add RETRY_BACKOFF_DELAYS [1s, 2s, 5s, 10s, 20s] for exponential backoff
- Add RETRY_JITTER_FACTOR (0.15) to prevent thundering herd
- Change INITIAL_PREKEY_COUNT from 812 to 30 (safer for rate limiting)
- Change MIN_UPLOAD_INTERVAL from 5s to 60s (avoids rate limiting)
- Change syncFullHistory default to false (conservative approach)
Based on RSocket fork's production-tested configuration.
- Add bounds validation for config values (prevent DoS via env vars)
- Replace ?? false with || false in isValidMapping (correct operator)
- Rename deleteMapping to deleteMappingFromCache (clarify behavior)
- Keep deleteMapping as deprecated alias for backward compatibility
- Document destroyed state handling in read-only methods
- Document async metrics module loading race condition
- Document storeLIDPNMappings return type change
- Add failure tracking and warning in getLIDsForPNs
- Document exponential backoff retry pattern in config and method
FIX: startPromise was never cleared after stop(), causing start()
to return the old resolved promise instead of creating a new server.
This broke any flow that stops and restarts metrics (config reloads, tests).
Now stop() clears startPromise, allowing proper server restart.
Fixes based on Copilot AI and Codex code review comments:
1. Race Condition Fix (lines 1065-1069):
- BEFORE: setTimeout(100ms) was incomplete - callers could resolve
before server was actually ready
- FIX: Cache the Promise from first start() call using startPromise
- All concurrent calls now return the same Promise
- Properly handles both success and error cases
2. CPU Percentage Calculation (lines 993-996):
- BEFORE: Division by cpuCount gave artificially low values
(e.g., 25% instead of 100% on 4-core for single-threaded process)
- FIX: Removed cpuCount division
- Now follows Unix/Prometheus standard: 100% = 1 core fully used
- Values can exceed 100% on multi-core (200% = 2 cores at 100%)
- Updated metric description to clarify this behavior
Fixes based on Copilot AI and Codex code review comments:
1. CPU Usage Metric (line ~930-933):
- FIX: process.cpuUsage() now calculates delta between measurements
- Returns actual percentage (0-100) instead of cumulative seconds
- Normalized across CPU cores
2. Duplicate SystemMetricsCollector (line ~1353-1355):
- FIX: Removed duplicate instantiation from PrometheusMetricsManager
- Now uses MetricsServer's collector via getSystemCollector()
3. Collection Interval (line ~1006-1008):
- FIX: Added collectIntervalMs to MetricsConfig
- Configurable via BAILEYS_PROMETHEUS_COLLECT_INTERVAL_MS
4. Error Handling (line ~1021-1022):
- FIX: Enhanced error messages with structured JSON response
- Includes error message, timestamp, and logs to console
5. Server Binding Security (line ~1034-1038):
- FIX: Default host changed from 0.0.0.0 to 127.0.0.1
- Configurable via BAILEYS_PROMETHEUS_HOST
- Warning shown if exposed on all interfaces
6. Race Condition (line ~998-1001):
- FIX: Added isStarting flag to prevent concurrent initialization
- Multiple concurrent start() calls now handled safely
- Support BAILEYS_PROMETHEUS_ENABLED, BAILEYS_PROMETHEUS_PORT, etc.
- Add support for BAILEYS_PROMETHEUS_LABELS JSON configuration
- Maintain backward compatibility with METRICS_* prefix
- Update default port to 9092 (matching user's .env)
- Enable by opt-in (enabled=false by default)
- Improved startup log with labels info
Claude