Cloudflare Workers insights

Elevated stale cache serving (cloudflare_cache_status_stale) indicates origin unreachability or errors. While stale serving maintains availability, it masks underlying origin issues that require attention.

Low cache hit rates for cacheable content increase load on origin servers and Worker processing time. Cache hit rate below 70% for content that should be cached indicates cache configuration issues or cache busting.

High origin 3xx redirect rates (cloudflare_origin_status_3xx) indicate Worker requests are hitting redirects at the origin, adding unnecessary latency and complexity. Workers should typically handle redirects or request the correct URL directly.

Worker errors (cloudflare_workers_errors, faas_errors) represent failed request handling that impacts users. Sustained error rates above 1% indicate application instability requiring immediate attention.

Workers making many subrequests (fetch calls to external APIs or origins) are network-bound rather than CPU-bound. High subrequest counts combined with elevated latency suggest optimization opportunities through batching, caching, or parallelization.

Worker request duration measures end-to-end latency from request start to response completion. Elevated p95/p99 latencies degrade user experience and may indicate slow external API calls, inefficient code paths, or resource contention.

High revalidation counts (cloudflare_cache_status_revalidated) indicate cache entries expiring frequently, causing unnecessary origin checks. This increases latency and origin load even when content hasn't changed.

Sudden drops in request volume (cloudflare_workers_requested, faas_invocations) may indicate route misconfiguration, DNS issues, or traffic shifts to other Workers. Gradual declines may indicate legitimate traffic changes or user migration.

Origin 4xx errors (cloudflare_origin_status_4xx) represent client errors returned by the backend, often indicating incorrect request formatting from Worker, authentication failures, or API contract mismatches.

Origin errors tracked by cloudflare_origin_status_5xx represent failures from backend servers, not Worker code. Elevated origin error rates indicate backend capacity issues, misconfigurations, or service degradation.

High Non-2xx Response Ratio Signals Application Issues

Elevated ratios of 4xx/5xx responses indicate application errors, misconfigurations, or failing backends. A sustained ratio above 30% suggests widespread issues requiring immediate investigation.

Workers showing exceededResources or scriptThrewException invocation statuses indicate CPU time limits, memory constraints, or unhandled exceptions. These failures prevent request completion and appear as 1101 or 1102 error codes to clients.

Workers have CPU time limits (10ms for free tier, 50ms for paid). Tracking upper quantiles (p95, p99) of cloudflare_workers_cpu_time reveals requests approaching limits, which may cause invocation errors.

Workers can exceed CPU time limits in some cases without generating invocation errors due to rollover CPU time mechanism, masking performance degradation until burst limits are hit.

Requests with 'Internal Error' invocation status indicate Workers runtime failures, not user code issues. While rare, elevated rates signal infrastructure problems, but these requests are not counted toward billing, making cost-based alerting ineffective.

Workers throw explicit 'script will never generate a response' errors when Promises are never resolved or rejected, but these are only visible in production logs, not during local development where equivalent code hangs indefinitely.

Sharing I/O objects (streams, requests, responses) across Worker invocations by caching them in global scope causes 'Cannot perform I/O on behalf of a different request' errors, which are not caught until runtime in production.