Prometheus Metrics

PRISM exports metrics in Prometheus text exposition format on the admin API at GET /metrics.

Request Metrics

Metric	Type	Description
`prism_requests_total{status="hit\|miss\|bypass\|stale\|error"}`	counter	Total requests by cache status
`prism_inflight_requests`	gauge	Currently processing requests
`prism_rate_limit_rejections_total`	counter	Requests rejected by rate limiter (429)

Render Metrics

Metric	Type	Description
`prism_render_active`	gauge	Currently active renders
`prism_render_queue_depth`	gauge	Requests waiting for a Chrome tab
`prism_render_total`	counter	Total renders completed
`prism_render_duration_ms`	histogram	Render duration in milliseconds
`prism_shadow_renders_total`	counter	Shadow renders completed (shadow mode)
`prism_content_validation_failures_total`	counter	Renders that failed content validation

prism_render_duration_ms histogram buckets: 100, 250, 500, 1000, 2500, 5000, 10000, 30000

Cache Metrics

Metric	Type	Description
`prism_cache_entries`	gauge	Number of entries in the render cache
`prism_cache_memory_bytes`	gauge	Memory used by cached entries
`prism_cache_evictions_total`	counter	Cache entries evicted (LRU or TTL)

Chrome Tab Pool Metrics

Metric	Type	Description
`prism_tab_pool_size`	gauge	Total tabs in the pool
`prism_tab_pool_available`	gauge	Idle tabs available for rendering
`prism_tab_crashes_total`	counter	Tab crashes detected
`prism_tab_creates_total`	counter	New tabs created
`prism_tab_recycles_total`	counter	Tabs recycled after max renders
`prism_chrome_restarts_total`	counter	Full Chrome process restarts

Origin Metrics

Metric	Type	Description
`prism_origin_requests_total`	counter	Requests forwarded to the origin server
`prism_origin_duration_sum_ms`	counter	Cumulative origin request duration (ms)

Warmup Metrics

Metric	Type	Description
`prism_warmup_urls_processed`	counter	URLs successfully warmed up
`prism_warmup_urls_failed`	counter	URLs that failed during warmup

Circuit Breaker

Metric	Type	Description
`prism_circuit_breaker_state`	gauge	0 = closed (healthy), 1 = open (tripped), 2 = half-open (probing)

System Metrics

Metric	Type	Description
`prism_uptime_seconds`	gauge	Seconds since PRISM started
`process_resident_memory_bytes`	gauge	Resident memory (RSS) of the PRISM process
`process_virtual_memory_bytes`	gauge	Virtual memory of the PRISM process
`process_cpu_seconds_total`	counter	Total CPU time consumed
`process_open_fds`	gauge	Open file descriptors
`process_max_fds`	gauge	Maximum file descriptors (ulimit)
`process_threads`	gauge	Number of OS threads

Renderer Internals

Metric	Type	Description
`prism_render_queue_wait_ms`	histogram	Time a render waited for a free tab. Distinct from render duration: this is the queue, that is the work.
`prism_render_retries_total`	counter	Renders retried after a stalled navigation. One request costing two tabs, so a climb here means the stall is getting more frequent.
`prism_render_failures_total{class}`	counter	Render failures by what they say about the shared browser. Only `class="systemic"` feeds the circuit breaker.
`prism_navigation_duration_sum_ms`	counter	Sum of Chrome navigation time — the origin plus its JavaScript.
`prism_html_extract_duration_sum_ms`	counter	Sum of DOM serialisation time. Rising against flat navigation means pages got larger, not slower.
`prism_oversize_rejections_total{check}`	counter	Documents refused for size, labelled by which check caught them.
`prism_passthrough_not_html_total`	counter	Requests proxied unchanged because the origin served non-HTML. Not a failure, but each one cost a Chrome navigation first.
`prism_requests_rejected_total{reason}`	counter	Requests shed, by reason: `queue_full`, `circuit_open`, `rate_limited`, `render_failed`.
`prism_rate_limited_total{limiter}`	counter	Rate-limit rejections attributed to the `ip` or `domain` budget. Which one is shedding is usually the whole question.

Chrome Process

Metric	Type	Description
`prism_chrome_rss_bytes`	gauge	Resident memory of the browser process and all its descendants. `process_resident_memory_bytes` covers PRISM only, and Chrome is where growth happens.
`prism_chrome_rss_age_seconds`	gauge	Age of the RSS sample above. Exported so a stale gauge reads as stale rather than as a current measurement.
`prism_chrome_recycles_total`	counter	Proactive recycles on age or render budget — routine hygiene, unlike `prism_chrome_restarts_total`, which counts failures.

Licence

Metric	Type	Description
`prism_license_valid`	gauge	1 when the licence allows rendering. Stays 1 throughout the grace window.
`prism_license_grace_active`	gauge	1 exactly while the grace window is open — the interval where renewing still helps. The only metric that shows it.
`prism_license_info{status}`	gauge	Licence state as a label: `valid`, `trial`, `expired`, `no_license`, `invalid`.

Bots and Fleet

Metric	Type	Description
`prism_bot_verification_total{outcome}`	counter	FCrDNS checks of claimed crawlers: `verified`, `spoofed`, `unverifiable`. Rising `spoofed` is an attack becoming visible; rising `unverifiable` is your resolver degrading.
`prism_peer_purge_failures_total`	counter	Purge fan-outs that failed to reach a replica. Until durable purge lands, the only sign a purge did not converge.
`prism_no_transform_ignored_total`	counter	Responses rendered although the origin sent `Cache-Control: no-transform`. Counted whether or not `cache.honor_no_transform` is set, so ignoring the directive is a visible decision rather than an omission. Non-zero on a deployment that cares about the letter of RFC 9111 §5.2.2.6 is the signal to turn that setting on.
`prism_background_task_failures_total`	counter	Supervisor tasks that exited unexpectedly. Any non-zero value means a capability (crash recovery, orphan cleanup, limiter sweeps) is silently gone.

Scrape Configuration

Add PRISM to your Prometheus scrape_configs:

scrape_configs:
  - job_name: prism
    scrape_interval: 15s
    static_configs:
      - targets: ['127.0.0.1:4001']
    # If bearer_token is configured:
    authorization:
      type: Bearer
      credentials: your-secret-token

Grafana Dashboard

Import this JSON snippet as a Grafana dashboard to get started with key panels:

{
  "title": "Trident PRISM",
  "panels": [
    {
      "title": "Cache Hit Rate",
      "type": "stat",
      "targets": [
        {
          "expr": "rate(prism_requests_total{status=\"hit\"}[5m]) / (rate(prism_requests_total{status=\"hit\"}[5m]) + rate(prism_requests_total{status=\"miss\"}[5m]))",
          "legendFormat": "Hit Rate"
        }
      ],
      "fieldConfig": {
        "defaults": { "unit": "percentunit", "thresholds": { "steps": [
          { "value": 0, "color": "red" },
          { "value": 0.7, "color": "yellow" },
          { "value": 0.9, "color": "green" }
        ]}}
      }
    },
    {
      "title": "Render Duration (p50 / p95 / p99)",
      "type": "timeseries",
      "targets": [
        {
          "expr": "histogram_quantile(0.50, rate(prism_render_duration_ms_bucket[5m]))",
          "legendFormat": "p50"
        },
        {
          "expr": "histogram_quantile(0.95, rate(prism_render_duration_ms_bucket[5m]))",
          "legendFormat": "p95"
        },
        {
          "expr": "histogram_quantile(0.99, rate(prism_render_duration_ms_bucket[5m]))",
          "legendFormat": "p99"
        }
      ],
      "fieldConfig": { "defaults": { "unit": "ms" } }
    },
    {
      "title": "Active Renders & Queue Depth",
      "type": "timeseries",
      "targets": [
        { "expr": "prism_render_active", "legendFormat": "Active" },
        { "expr": "prism_render_queue_depth", "legendFormat": "Queued" }
      ]
    },
    {
      "title": "Tab Pool",
      "type": "timeseries",
      "targets": [
        { "expr": "prism_tab_pool_size", "legendFormat": "Total" },
        { "expr": "prism_tab_pool_available", "legendFormat": "Available" }
      ]
    },
    {
      "title": "Cache Entries & Memory",
      "type": "timeseries",
      "targets": [
        { "expr": "prism_cache_entries", "legendFormat": "Entries" },
        { "expr": "prism_cache_memory_bytes", "legendFormat": "Memory (bytes)" }
      ]
    },
    {
      "title": "Circuit Breaker State",
      "type": "stat",
      "targets": [
        { "expr": "prism_circuit_breaker_state", "legendFormat": "State" }
      ],
      "fieldConfig": {
        "defaults": {
          "mappings": [
            { "type": "value", "options": { "0": { "text": "CLOSED", "color": "green" } } },
            { "type": "value", "options": { "1": { "text": "OPEN", "color": "red" } } },
            { "type": "value", "options": { "2": { "text": "HALF-OPEN", "color": "yellow" } } }
          ]
        }
      }
    },
    {
      "title": "Origin Latency (avg)",
      "type": "timeseries",
      "targets": [
        {
          "expr": "rate(prism_origin_duration_sum_ms[5m]) / rate(prism_origin_requests_total[5m])",
          "legendFormat": "Avg Origin Latency (ms)"
        }
      ],
      "fieldConfig": { "defaults": { "unit": "ms" } }
    },
    {
      "title": "Process Memory",
      "type": "timeseries",
      "targets": [
        { "expr": "process_resident_memory_bytes", "legendFormat": "RSS" },
        { "expr": "process_virtual_memory_bytes", "legendFormat": "Virtual" }
      ],
      "fieldConfig": { "defaults": { "unit": "bytes" } }
    }
  ]
}

Alerts

The alerting rules ship with the product rather than living here: 12 of them in deploy/ops/prometheus-rules.yaml (plain Prometheus) and deploy/ops/prometheusrule.yaml (the Operator CRD), each with a runbook in deploy/ops/runbooks/. A test keeps every metric they reference in step with what the binary exports, keeps the two files identical, and fails if any alert loses its runbook.

This page used to carry its own examples with the same alert names and different thresholds, which is worse than having none: an operator who copied both got two rules called PrismCircuitBreakerOpen firing at different times. Use the shipped set.

Request Metrics​

Render Metrics​

Cache Metrics​

Chrome Tab Pool Metrics​

Origin Metrics​

Warmup Metrics​

Circuit Breaker​

System Metrics​

Renderer Internals​

Chrome Process​

Licence​

Bots and Fleet​

Scrape Configuration​

Grafana Dashboard​

Alerts​