Repository navigation
Tags
Tags
feat: add Cloudflare Images storage metrics (#34) Adds cloudflare_images_count_current and cloudflare_images_count_allowed gauges via the Images v1 stats REST endpoint. The images query is account-level (no zones required) so also fixes the MetricExporter zone guard to skip the "no zone context" early-return for pure account-level queries, allowing images and similar account-scoped metrics to refresh independently of zone discovery. Closes #27 Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
feat: 1-minute hostname metrics for alerting (#22) * feat: 1-minute hostname metrics for alerting, avg latencies, separate percentile families BREAKING: hostname metrics now use a fixed 1-minute window instead of 1h/2h. This is alpha/beta software — no backward compatibility required. Changes: - Hostname metrics now query a single 1-minute window (last completed minute) - Removed 'window' label from all hostname metrics - Primary latency metrics (_edge_ttfb_seconds, _origin_response_duration_seconds) now expose Cloudflare-reported averages instead of quantiles - Percentiles exported as separate metric families: _edge_ttfb_p50_seconds, _edge_ttfb_p95_seconds, _origin_response_duration_p50_seconds, _origin_response_duration_p95_seconds - Removed 'quantile' label (reserved in Prometheus for Summary types) - Added HOST_METRICS_DELAY_SECONDS (default: 120s) for hostname-specific ingestion delay, independent from global SCRAPE_DELAY_SECONDS - Added 'avg' block to HostnameHttpMetricsQuery GraphQL query - Updated config UI, runtime config, README Rationale (Prometheus best practices): - One metric should represent the same logical thing across all label dimensions - Avoid 'quantile' label on non-Summary gauges (Writing Exporters docs) - Prefer averages for alerting; percentiles are not aggregatable - Fixed window removes mixed-semantics multiplexing via label * fix: review fixes — stale docs, min delay validation, debug logging, remove unnecessary spread * chore: set HOST_METRICS_DELAY_SECONDS default to 60s
Add video metrics (#17) * Add video metrics * fix: review fixes for stream video metrics - Use base units per Prometheus conventions: minutes→seconds (*60), milliseconds→seconds (/1000) - Rename playback_starts_total to playback_count_total (count is adaptive group record count, not playback starts) - Fix JSDoc: mediaType is VOD/live source, not device type - Document Stream exclusion from FREE_TIER_QUERIES - Revert wrangler.jsonc KV id from local dev artifact - Update README: add stream metric tables, fix architecture counts (16→18 queries, 4→6 account-level) --------- Co-authored-by: Dillon Mulroy <dillon@cloudflare.com>
feat: exhaustive Magic Transit & Magic Firewall observability (#16) * feat: add Magic Transit tunnel SLO metrics and per-status failure breakdown Add new magic-transit-slo query using magicTransitTunnelHealthCheckSLOsAdaptiveGroups for authoritative tunnel health status, effective SLO, and target SLO thresholds. Enrich existing magic-transit handler with per-resultStatus failure counts. New metrics: - cloudflare_magic_transit_tunnel_failure_by_status (gauge) - cloudflare_magic_transit_tunnel_slo_status (gauge) - cloudflare_magic_transit_tunnel_effective_slo (gauge) - cloudflare_magic_transit_tunnel_target_slo (gauge) Closes request for error rates and healthy vs failed MT tunnel visibility. * feat: exhaustive Magic Transit/Firewall observability — tunnel traffic, per-rule MFW, enriched NA labels Add 2 new query scopes and enrich existing datasets for comprehensive Magic Transit and Magic Firewall monitoring: New queries: - magic-transit-traffic: per-tunnel bandwidth (bits/packets) with direction, on_ramp, off_ramp labels - magic-firewall-samples: per-rule sampled traffic (bits/packets) Enriched existing metrics: - MT health checks: add avg.tunnelState gauge (0=down/0.5=degraded/1=healthy) - MFW network analytics: add rule_id, ruleset_id, verdict labels - MT network analytics: add ingress_tunnel, egress_tunnel, on_ramp, off_ramp - DDoS network analytics: add mitigation_reason, mitigation_scope - TCP protection: add mitigation_reason, mitigation_scope, protocol_state - DNS protection: add dns_query_type * review: revert high-cardinality NA labels, tunnel_state → StateSet gauges - Revert enriched labels on combined NA query (MT tunnel names, MFW rule_id/verdict, DDoS/TCP/DNS extra dims) — breaking change to existing metrics, high cardinality in shared query - Replace tunnel_state float gauge (0/0.5/1) with StateSet-style boolean gauges (_healthy, _degraded, _down) per Prometheus/ OpenMetrics best practices - Extract classifyTunnelState helper with named threshold constants - Update README metric docs to match reverted labels * review: fix active tunnel filter, colon delimiter, GQL indentation, README docs - Fix active tunnel filter: uint8 field compared as String === "true" was always false; use numeric === 1 comparison - Replace colon-delimited key splitting with nested Map in both health check and SLO tunnel grouping (avoids tunnel name collision) - Normalize 3 new GQL queries from tabs to 2-space indentation - Fix README: remove stale dns_query_type label, update architecture diagram to 19 total / 7 account-level queries --------- Co-authored-by: Dillon Mulroy <dillon@cloudflare.com>
feat: hostname metrics, cache-miss metrics, lb-weight metrics; cleanu… …p docs/specs/.opencode; gitignore .opencode; strip KV namespace id; bump v0.0.1-beta.8
PreviousNext