sentry-reliability-patterns

What this skill does

# Sentry Reliability Patterns

## Overview

Build Sentry integrations that never take your application down via three pillars: safe initialization with graceful degradation, a circuit breaker that stops hammering Sentry when unreachable, and an offline event queue that buffers errors during outages. Every pattern prioritizes application uptime over telemetry completeness.

## Prerequisites

- `@sentry/node` v8+ (TypeScript) or `sentry-sdk` v2+ (Python)
- A valid Sentry DSN from project settings at `sentry.io`
- A fallback logging destination decided (console, file, or external logger)
- Understanding of your application shutdown lifecycle (signal handlers, container orchestration)

## Instructions

### Step 1 — Safe Initialization with Graceful Degradation

Wrap `Sentry.init()` in try/catch so an invalid DSN, network error, or SDK bug never crashes the app. Track initialization state with a boolean flag. Protect `beforeSend` callbacks with their own error boundary.

Create `lib/sentry-safe.ts` with `initSentrySafe()` and `captureError()`. See [graceful-degradation.md](references/graceful-degradation.md) for full implementation.

Key rules:

- Never let `Sentry.init()` crash the process — wrap in try/catch, set `sentryAvailable = false` on failure
- Verify client creation with `Sentry.getClient()` — invalid DSNs silently produce no client
- Always log errors locally as baseline before attempting Sentry capture
- Wrap user-supplied `beforeSend` hooks in nested try/catch — return raw event on hook failure

### Step 2 — Circuit Breaker for Sentry Outages

When Sentry is unreachable, continued attempts waste resources and add latency. Track consecutive failures and trip open after a threshold. After cooldown, enter half-open state and send a single probe.

Implement `SentryCircuitBreaker` class with closed/open/half-open states. See [circuit-breaker-pattern.md](references/circuit-breaker-pattern.md) for full implementation. Expose state via [health-checks.md](references/health-checks.md) endpoint.

Key rules:

- Default: 5 failures to trip open, 60-second cooldown before half-open probe
- In open state, skip Sentry calls entirely and log to fallback
- On half-open success, reset to closed with zero failure count
- Expose `getStatus()` for health check endpoints and monitoring dashboards

### Step 3 — Offline Queue, Custom Transport, and Graceful Shutdown

Buffer events when network is unavailable and replay on reconnect. Use bounded file-based queue to survive restarts. Pair with signal handlers that flush via `Sentry.close()` before process exit.

Implement three modules:

- `lib/sentry-offline-queue.ts` — `enqueueEvent()` and `drainQueue()`. See [network-failure-handling.md](references/network-failure-handling.md)
- `lib/sentry-transport.ts` — Custom transport with exponential backoff retry. See [timeout-handling.md](references/timeout-handling.md)
- `lib/sentry-shutdown.ts` — `SIGTERM`/`SIGINT` handlers calling `Sentry.close(2000)`. See [timeout-handling.md](references/timeout-handling.md)

Key rules:

- Cap offline queue at 1000 events, evict oldest when full
- Drain queue on startup and when connectivity restores
- Call `Sentry.close(timeout)` before `process.exit()` — without it, in-flight events are silently dropped
- For critical errors, use [dual-write-pattern.md](references/dual-write-pattern.md) to send to multiple destinations via `Promise.allSettled`

## Output

- Safe init wrapper catching SDK failures, starting app in degraded mode
- `captureError()` with automatic fallback to local logging
- Circuit breaker stopping sends after repeated failures, self-healing after cooldown
- Health check endpoint exposing SDK status and circuit breaker state
- File-based offline queue buffering events during outages, draining on reconnect
- Signal handlers flushing in-flight events before process exit
- Custom transport with exponential-backoff retry logic

## Error Handling

| Error | Cause | Solution |
|-------|-------|----------|
| App crashes on `Sentry.init()` | Invalid DSN or SDK bug | Wrap in try/catch via `initSentrySafe()` |
| Events lost on `SIGTERM` | No `Sentry.close()` before exit | Register signal handlers with `Sentry.close(2000)` |
| Sentry outage cascades latency | Every error path hits Sentry HTTP | Circuit breaker trips after 5 failures |
| Events lost during network blip | SDK drops events silently | Retry transport + offline queue |
| Silent event loss | SDK fails without throwing | Health check probes with `captureMessage` + `flush` |
| Queue grows unbounded | Never drained, Sentry permanently down | Cap at 1000 events, drain on startup |
| `beforeSend` crashes pipeline | User hook throws | Nested try/catch, return raw event |

See [errors.md](references/errors.md) for extended troubleshooting.

## Examples

See [examples.md](references/examples.md) for complete TypeScript and Python integration examples including full-stack wiring of all three patterns.

## Resources

- [Sentry JS Configuration](https://docs.sentry.io/platforms/javascript/configuration/) — `beforeSend`, `sampleRate`, init options
- [Custom Transports](https://docs.sentry.io/platforms/javascript/configuration/transports/) — retry and offline transports
- [Shutdown & Draining](https://docs.sentry.io/platforms/javascript/configuration/draining/) — `Sentry.close()` and `Sentry.flush()`
- [Sentry Python SDK](https://docs.sentry.io/platforms/python/) — `sentry_sdk.init()`, `flush()`, scope management
- [Sentry Status Page](https://status.sentry.io/) — monitor platform outages

## Next Steps

- Emit circuit breaker state changes to observability platform (Datadog, Prometheus) for outage alerting
- Set up periodic `drainQueue()` via `setInterval` (Node) or cron (Python) instead of startup-only
- Apply retry transport pattern to Python via `sentry_sdk.init(transport=...)` parameter
- Test failure modes in staging — simulate Sentry failures with `beforeSend` to verify circuit breaker behavior
- Add [dual-write](references/dual-write-pattern.md) for P0/fatal errors to secondary destinations (CloudWatch, PagerDuty)