Metrics Reference

Use this page to look up exported metric families, labels, and meanings.

Quick Lookup

If you need to check	Start here
request volume, outcomes, and status classes	Core Request Metrics and Request Breakdown Metrics
latency families	Latency Metrics
overload, brownout, and circuit protection	Overload And Admission Metrics
quota decisions and backend health	Quota Metrics
retries and hedges	Retry And Hedging Metrics
downstream or upstream TLS	TLS Metrics
control-plane or runtime generation signals	Control Plane And Runtime Metrics

Endpoint

method: GET
path: configurable by observability.metrics.path
default path: /metrics

Canonical Label Rules

The most important low-cardinality labels are:

upstream
backend
route
status_class
outcome
reason
decision
selector_dimensions
backend_mode

Operator rule:

build dashboards and alerts from canonical label values, not from prose parsing
keep quota, overload, auth, and generic backend failure separate

Core Request Metrics

Metric	Type	Meaning
`impulse_requests_total`	counter	Total requests seen by the proxy
`impulse_requests_success`	counter	Successful upstream responses
`impulse_requests_failure`	counter	Failed requests
`impulse_request_validation_rejects`	counter	Requests rejected by protocol validation
`impulse_policy_denied`	counter	Requests denied by runtime method/path policy
`impulse_external_auth_allowed`	counter	Requests explicitly allowed by external auth
`impulse_external_auth_denied`	counter	Requests denied, challenged, or redirected by external auth
`impulse_external_auth_timeout`	counter	External auth decisions that timed out
`impulse_external_auth_error`	counter	External auth transport or execution failures
`impulse_request_rate_limited`	counter	Requests rejected by scoped request rate limits

Request Breakdown Metrics

These families are the primary source for production dashboards because they preserve request totals while adding low-cardinality dimensions.

Metric	Type	Meaning
`impulse_upstream_requests_total{upstream,status_class,outcome}`	counter	Completed requests grouped by upstream, response status class, and final outcome
`impulse_backend_requests_total{upstream,backend,status_class,outcome}`	counter	Completed requests grouped by upstream and selected backend

Expected label values:

status_class: 1xx, 2xx, 3xx, 4xx, 5xx, other, unknown
outcome: success, failure, timeout, backend_error, overload_shed, rate_limited

Use these for questions like:

which upstream is producing 5xx responses?
which backend is taking most of the failed traffic?
are failures mostly timeouts, backend errors, or overload shedding?

These are the primary traffic families behind the packaged dashboards and recording rules.

Latency Metrics

Metric	Type	Meaning
`impulse_upstream_request_latency_ms_bucket{upstream,outcome,le}`	histogram bucket	End-to-end request latency grouped by upstream and final outcome
`impulse_upstream_request_latency_ms_sum{upstream,outcome}`	histogram sum	Sum of request latency observations in milliseconds
`impulse_upstream_request_latency_ms_count{upstream,outcome}`	histogram count	Count of latency observations
`impulse_route_latency_ms_p50{route}`	gauge	Approximate p50 route latency
`impulse_route_latency_ms_p95{route}`	gauge	Approximate p95 route latency
`impulse_route_latency_ms_p99{route}`	gauge	Approximate p99 route latency

Practical note:

if you only grep impulse_requests_total and impulse_requests_success, you are looking at the coarse top-level counters rather than the richer labeled families above
for Grafana and Prometheus alerting, prefer the labeled upstream/backend metrics and the histogram family

Early Data Metrics

Metric	Type	Meaning
`impulse_early_data_accepted`	counter	Requests accepted in early data
`impulse_early_data_rejected`	counter	Requests rejected in early data

Health And Backend Metrics

Metric	Type	Meaning
`impulse_health_checks_total`	counter	Active health checks executed
`impulse_health_checks_success`	counter	Successful active health checks
`impulse_health_checks_failure`	counter	Failed active health checks
`impulse_backend_timeouts`	counter	Backend timeout events
`impulse_backend_errors`	counter	Backend error events
`impulse_health_failures_total{reason=...}`	counter	Passive health failures by reason such as `5xx`, `timeout`, `transport`, `tls`

Overload And Admission Metrics

Metric	Type	Meaning
`impulse_overload_shed`	counter	Total requests shed due to overload controls
`impulse_overload_shed_by_reason_total{reason=...}`	counter	Shed decisions by reason
`impulse_inflight_wait_admit_total{scope=...}`	counter	Successful admissions after micro-wait
`impulse_brownout_active`	gauge	Brownout mode active state
`impulse_circuit_breaker_rejected_total`	counter	Requests rejected by open circuits

Interpretation rules:

impulse_overload_shed_by_reason_total is for overload self-protection
impulse_request_rate_limited is for scoped rate-limit enforcement
neither of those is the quota contract signal

Quota Metrics

Metric	Type	Meaning
`impulse_quota_policy_outcomes_total{policy,decision,reason,selector_dimensions,backend_mode}`	counter	Quota outcomes grouped by matched policy, decision, canonical reason, selector dimensions, and backend mode
`impulse_quota_backend_health_total{backend_mode,reason}`	counter	Quota backend health and error observations

Interpretation rules:

decision answers whether quota allowed, denied, failed open, or otherwise changed request handling
reason answers why the quota result happened
selector_dimensions tells you which identity dimensions were part of the matched policy
backend_mode tells you whether the decision came from Redis, local fallback, or another configured enforcement mode

Use these to separate:

contract exhaustion
quota backend degradation
fail-open versus fail-closed behavior

Connection And Ingress Metrics

Metric	Type	Meaning
`impulse_active_connections`	gauge	Current active QUIC connections
`impulse_connection_cap_rejects`	counter	New connections rejected by active-connection cap
`impulse_ingress_packets_total`	counter	Total UDP packets processed
`impulse_ingress_queue_drops`	counter	Packets dropped due to full shard queues
`impulse_ingress_queue_drop_bytes`	counter	Bytes dropped due to full shard queues
`impulse_ingress_queue_bytes`	gauge	Bytes currently buffered in ingress shard queues
`impulse_ingress_bad_header_total`	counter	Packets dropped due to invalid QUIC headers
`impulse_ingress_rate_limited_total`	counter	Initial packets rejected by rate limiting
`impulse_ingress_unroutable_total`	counter	Non-initial packets for unknown connections
`impulse_ingress_draining_drops_total`	counter	Packets dropped while draining
`impulse_ingress_connection_create_failed_total`	counter	Connection creation failures
`impulse_ingress_version_neg_failed_total`	counter	Version-negotiation construction failures
`impulse_scid_rotations`	counter	SCID rotations

Buffer And Body-Pressure Metrics

Metric	Type	Meaning
`impulse_request_buffered_bytes`	gauge	Bytes currently buffered in request backpressure queues
`impulse_request_buffered_high_watermark_bytes`	gauge	Peak buffered-request bytes since process start
`impulse_request_buffer_limit_rejects`	counter	Requests rejected by request-buffer caps
`impulse_response_prebuffer_limit_rejects`	counter	Unknown-length responses rejected by prebuffer cap

Retry And Hedging Metrics

Metric	Type	Meaning
`impulse_retries_total`	counter	Total retry attempts
`impulse_retry_denied_total{reason=...}`	counter	Retry attempts blocked by reason
`impulse_retry_attempts_total{reason=...}`	counter	Retries triggered by error reason
`impulse_hedge_triggered_total`	counter	Hedge attempts started
`impulse_hedge_won_total`	counter	Hedge won the race
`impulse_hedge_wasted_total`	counter	Hedge lost or became unnecessary
`impulse_hedge_primary_won_after_trigger_total`	counter	Primary still won after hedge start
`impulse_hedge_primary_late_ms_total`	counter	Aggregate lateness after hedge trigger
`impulse_hedge_primary_late_samples_total`	counter	Late-primary observations

TLS Metrics

Metric	Type	Meaning
`impulse_downstream_tls_handshake_success_total`	counter	Successful downstream TLS handshakes
`impulse_downstream_tls_handshake_failure_total{listener,reason}`	counter	Downstream TLS handshake failures
`impulse_downstream_tls_certificate_selection_total{listener,selection}`	counter	Certificate-selection outcomes
`impulse_downstream_tls_alpn_total{listener,protocol}`	counter	Negotiated ALPN protocols
`impulse_downstream_tls_certificate_not_after_seconds{listener,server_name}`	gauge	Certificate expiration timestamp
`impulse_downstream_tls_certificate_days_remaining{listener,server_name}`	gauge	Estimated remaining days to expiration
`impulse_upstream_tls_failure_total{backend,phase,reason}`	counter	Upstream TLS failures

DNS And Backend Refresh Metrics

Metric	Type	Meaning
`impulse_backend_dns_refresh_success_total`	counter	Successful backend DNS refreshes
`impulse_backend_dns_refresh_failure_total`	counter	Failed backend DNS refreshes
`impulse_backend_dns_address_set_changes_total`	counter	Refreshes that changed address set
`impulse_backend_client_rotations_total`	counter	Backend client rotations caused by DNS changes
`impulse_backend_dns_last_refresh_success_seconds`	gauge	Unix timestamp of last successful refresh

JWT And JWKS Metrics

Metric	Type	Meaning
`impulse_jwt_validation_failures_total{reason}`	counter	JWT validation failures by stable rejection reason
`impulse_jwt_algorithm_rejections_total{algorithm}`	counter	Tokens rejected because their `alg` is not in the allowlist
`impulse_jwks_unknown_kid_total{jwks_source_id}`	counter	Tokens presenting a `kid` absent from the cached key set
`impulse_jwks_refresh_success_total{jwks_source_id}`	counter	Successful JWKS refreshes
`impulse_jwks_refresh_failure_total{jwks_source_id}`	counter	Failed JWKS refreshes
`impulse_jwks_age_seconds{jwks_source_id}`	gauge	Age of the active key set since last successful refresh
`impulse_jwks_state{jwks_source_id,state}`	gauge	Current cache state, `1` on the active state series
`impulse_jwks_active_keys{jwks_source_id}`	gauge	Usable verification keys currently loaded
`impulse_jwks_last_refresh_attempt_seconds{jwks_source_id}`	gauge	Unix timestamp of the last refresh attempt
`impulse_jwks_last_refresh_success_seconds{jwks_source_id}`	gauge	Unix timestamp of the last successful refresh

JWKS series are labelled by jwks_source_id, an opaque per-source identifier — the configured endpoint URL is never used as a label value, so query strings and embedded credentials cannot leak into the metrics endpoint. Map a source id back to its endpoint through jwks.sources[] in the /admin/runtime snapshot, which reports both jwks_source_id and a sanitized jwks_endpoint.

impulse_jwks_state reports one of never_fetched, fresh, stale, refresh_failed_retained, quarantined_retained, or empty_unusable. Only the last is unable to validate tokens; the two *_retained states are still serving last-known-good keys.

Control Plane And Runtime Metrics

Metric	Type	Meaning
`impulse_runtime_validation_attempts_total`	counter	Staged validation requests accepted by the control plane
`impulse_runtime_preview_attempts_total`	counter	Staged preview requests accepted by the control plane
`impulse_runtime_activation_total{result,reason}`	counter	Runtime activation outcomes grouped by result and canonical reason
`impulse_runtime_rollback_total{result,reason}`	counter	Runtime rollback outcomes grouped by result and canonical reason
`impulse_runtime_rejections_total{reason}`	counter	Runtime activation or rollback rejections grouped by canonical operator reason
`impulse_runtime_active_generation`	gauge	Current active runtime generation identifier
`impulse_runtime_history_depth`	gauge	Number of retained runtime history entries visible to the active generation
`impulse_control_api_connection_limit_drops`	counter	Control API connections dropped by limiter
`impulse_watchdog_restart_requests`	counter	Watchdog restart requests
`impulse_watchdog_restart_hooks`	counter	Restart hooks executed
`impulse_watchdog_degraded_windows`	counter	Degraded watchdog windows
`impulse_runtime_panics`	counter	Observed runtime panics

Use these when:

activation, rollback, or restart workflows are misbehaving
the active generation may differ across nodes
watchdog activity might be driving drain or restart behavior

Metrics To Control API Workflow

Use this fast path when the metrics tell you the class of problem but not the live state:

Metric family	Next control-plane read
`impulse_quota_policy_outcomes_total`	`GET /admin/runtime`
`impulse_quota_backend_health_total`	`GET /admin/runtime`
`impulse_runtime_activation_total`	`GET /admin/runtime/history`
`impulse_runtime_rollback_total`	`GET /admin/runtime/history`
`impulse_runtime_rejections_total`	`GET /admin/runtime/history`
`impulse_watchdog_*` or `impulse_runtime_panics`	`GET /admin/runtime`
`impulse_downstream_tls_*`	`GET /admin/runtime`