Exportarr

AIO Prometheus Exporter for Sonarr, Radarr, Lidarr, Prowlarr, Bazarr and Sabnzbd

Note: This exporter will not gather metrics from all apps at once. You will need an exportarr instance for each app. Be sure to see the examples below for more information.

Usage

Docker Compose

See examples in the examples/compose directory.

Kubernetes

See examples in the examples/kubernetes directory.

Docker CLI

Replace $app, $port and $apikey with one of the supported apps, port and api key

# PORT must be unique across all Exportarr instances
docker run --name exportarr_$app \
  -e PORT=9707 \
  -e URL="http://x.x.x.x:$port" \
  -e API_KEY="$apikey" \
  --restart unless-stopped \
  -p 9707:9707 \
  -d ghcr.io/onedr0p/exportarr:latest $app

Visit http://127.0.0.1:9707/metrics to see the app metrics

CLI

Replace $app, $port and $apikey with one of the supported apps, port and api key

./exportarr $app --help

# --port must be unique across all Exportarr instances
./exportarr $app \
  --port 9707 \
  --url "http://x.x.x.x:$port" \
  --api-key "$apikey"

Visit http://127.0.0.1:9707/metrics to see the app metrics

Configuration

Environment Variable	CLI Flag	Description	Default	Required
PORT	--port or -p	The port Exportarr will listen on	8081	❌
URL	--url or -u	The full URL to the app being exported		✅
API_KEY	--api-key or -a	API Key for the app being exported		✅
API_KEY_FILE	—	Path to a file containing the API key (Docker/Kubernetes secrets); overrides API_KEY		❌
INTERFACE	--interface or -i	The interface IP Exportarr will listen on	0.0.0.0	❌
LOG_LEVEL	--log-level or -l	Log level (debug, info, warn, error)	info	❌
LOG_FORMAT	--log-format	Log format (console, json)	console	❌
DISABLE_SSL_VERIFY	--disable-ssl-verify	Set to true to disable SSL verification	false	❌
REQUEST_TIMEOUT	--request-timeout	HTTP timeout per request to the target app	60s	❌
AUTH_PASSWORD	--auth-password	Password for form auth		❌
AUTH_USERNAME	--auth-username	Username for form auth		❌
FORM_AUTH	--form-auth	Use form-based authentication	false	❌
ENABLE_UNKNOWN_QUEUE_ITEMS	--enable-unknown-queue-items	Set to true to enable gathering unknown queue items	false	❌
DISABLE_QUALITY_METRICS	--disable-quality-metrics	Skip per-item quality breakdowns (episodefile/trackfile lookups; ~1 API call per series/artist each scrape)	false	❌
DISABLE_EPISODE_METRICS	--disable-episode-metrics	Skip per-episode metrics (sonarr episode monitoring, bazarr episode-subtitle walk; load scales with library size)	false	❌
DISABLE_ALBUM_METRICS	--disable-album-metrics	Skip per-album metrics (lidarr album lookups; ~1 API call per artist each scrape)	false	❌
DISABLE_HISTORY_METRICS	--disable-history-metrics	Skip the history endpoint — its total forces a full count over the unprunable history table, slow on multi-year instances	false	❌
DISABLE_WANTED_METRICS	--disable-wanted-metrics	Skip the wanted/missing and wanted/cutoff endpoints — their totals force full counts, slow on very large libraries	false	❌
PROWLARR__BACKFILL	--backfill	Set to true to enable backfill of historical metrics	false	❌
PROWLARR__BACKFILL_SINCE_DATE	--backfill-since-date	Set a date (YYYY-MM-DD) from which to start the backfill	1970-01-01 (epoch)	❌
BAZARR__SERIES_BATCH_SIZE	--series-batch-size	Number of series per Bazarr episodes API call	300	❌
BAZARR__SERIES_BATCH_CONCURRENCY	--series-batch-concurrency	Concurrent Bazarr episodes API calls	10	❌

Prowlarr Backfill

The Prowlarr collector is a little different than other collectors as it's hitting an actual "stats" endpoint, collecting counters of events that happened in a small time window, rather than getting all-time statistics like the other collectors. This means that by default, when you start the Prowlarr collector, collected stats will start from that moment (all counters will start from zero).

To backfill all Prowlarr Data, either use PROWLARR__BACKFILL or --backfill.

Note that the first request can be extremely slow, depending on how long your Prowlarr instance has been running. You can also specify a start date to limit the backfill if the backfill is timing out:

PROWLARR__BACKFILL_SINCE_DATE=2023-03-01 or --backfill-since-date=2023-03-01

Scrape performance and sizing

Measured against real instances with every metric enabled — use these scaling rules to pick scrape intervals and container limits:

App	Typical scrape	What it scales with
SABnzbd	tens of milliseconds	flat — two small API calls
Radarr	a second or two	one large movie-list payload; mostly JSON transfer + decode
Sonarr	a few seconds	per-series fan-out (two API calls per series, 10 concurrent)
Lidarr	a few seconds	per-artist fan-out (two API calls per artist, 10 concurrent)
Prowlarr	fast	stats endpoint (see the backfill note above for the first scrape)
Bazarr	tens of seconds	full episode-subtitle walk; the time is spent inside bazarr generating the batched responses, so batch/concurrency tuning barely moves it

• Set scrape_interval longer than your worst scrape. Bazarr with episode metrics enabled commonly needs 60s or more; if a scrape arrives while the previous one is still running, exportarr skips it and raises the collector's error gauge (see "Changed scrape behavior" below). The other apps are comfortable at 15–30s.
• The first scrape after startup is the slowest (TLS handshakes); connections are pooled and reused afterwards.
• Memory scales with the largest API payload decoded: expect roughly 25–100 MB RSS, with the high end during bazarr's episode walk or a large radarr movie list. In Kubernetes, set GOMEMLIMIT to the container memory limit so GC stays ahead of the decode spike, and watch the exporter's own go_*/process_* metrics.
• If a scrape is too slow, reach for the DISABLE_* flags above rather than a shorter REQUEST_TIMEOUT — they remove the expensive endpoints entirely instead of cutting requests off mid-flight.

Upgrading from v2 to v3

v3 is a breaking release. Review each section before upgrading.

Removed

• Readarr support — the readarr command, its metrics, and its dashboard panels are gone (Readarr was retired upstream).
• Basic auth — HTTP basic auth and the --basic-auth-username/--basic-auth-password flags are removed. AUTH_USERNAME/AUTH_PASSWORD now apply to form auth only and require FORM_AUTH=true; setting credentials without form auth is a startup error.
• config.xml parsing — CONFIG/--config is removed; exportarr no longer reads the *arr's config file. Provide the key via API_KEY/--api-key, or API_KEY_FILE (environment-only) for Docker and Kubernetes secrets mounted as files.
• Legacy variable aliases — APIKEY, APIKEY_FILE, BASIC_AUTH_USERNAME and BASIC_AUTH_PASSWORD no longer work.
• Log levels — fatal, panic and dpanic are gone; valid levels are debug, info, warn, error.
• ENABLE_ADDITIONAL_METRICS — removed. The metrics it bundled are now collected by default (the per-item fan-out is parallelized and ~10× faster), with granular opt-outs instead: DISABLE_QUALITY_METRICS, DISABLE_EPISODE_METRICS, DISABLE_ALBUM_METRICS. Set all that apply to restore v2's default-off behavior.

v2	v3
APIKEY	API_KEY
API_KEY_FILE=/path	unchanged — still supported (Docker/Kubernetes secrets)
CONFIG=/path/config.xml	URL + API_KEY
BASIC_AUTH_USERNAME/..._PASSWORD	removed — form auth uses AUTH_USERNAME/AUTH_PASSWORD + FORM_AUTH=true
LOG_LEVEL=fatal	LOG_LEVEL=error
ENABLE_ADDITIONAL_METRICS=true	(default behavior — remove it)
ENABLE_ADDITIONAL_METRICS unset	DISABLE_QUALITY_METRICS + DISABLE_EPISODE_METRICS + DISABLE_ALBUM_METRICS all true

Changed scrape behavior — update your alerts

• A failing collector no longer fails the whole scrape with HTTP 500. /metrics now returns 200 with everything that succeeded, plus a per-collector error gauge (e.g. radarr_collector_error, radarr_queue_collector_error) set to 1 for whatever failed. Alerts that relied on the target reporting up == 0 when the app was down should alert on *_collector_error > 0 instead.
• sabnzbd_collector_error renamed its target label to url, matching every other metric.
• Overlapping collections are skipped: if a scrape arrives while the previous sonarr/lidarr/bazarr collection is still running against a slow instance, exportarr skips it (raising the collector's error gauge) instead of stacking concurrent walks onto the app — the failure mode behind bazarr CPU drainage (#380).

Changed metrics

• bazarr_subtitles_score_total{score="93.45%"} (one series per distinct score — unbounded cardinality) is replaced by a histogram, bazarr_subtitles_score, with percentage buckets 10..90, 95, 100. Use histogram_quantile() or the bucket series directly (#239).
• <app>_queue_total now emits one series per (status, download_status, download_state) combination with accurate counts. An empty queue emits a single zero series (empty label values) instead of no series at all, so dashboards can tell "zero items" from "scrape failed". v2 emitted a single series carrying the total queue size under whichever labels the last queue item happened to have — sums still work, per-label panels will show corrected values.
• The self-instrumentation duration gauges (<app>_scrape_duration_seconds, sabnzbd_queue_query_duration_seconds, sabnzbd_server_stats_query_duration_seconds) are now histograms, so histogram_quantile() works across scrapes instead of only seeing the last value. They additionally expose sparse native histograms to scrapers that negotiate them; classic buckets remain for everyone else.
• Log output is structured slog (time=… level=… msg=…, or JSON with --log-format json); update anything parsing exporter logs.

New in v3

No action needed, but worth knowing:

• <app>_diskspace_free_bytes / <app>_diskspace_total_bytes (per disk, with totals — disk usage is finally computable).
• sabnzbd_speed_limit_bps / sabnzbd_speed_limit_percent.
• bazarr_throttled_providers and bazarr_signalr_connected{app="sonarr"|"radarr"}.
• bazarr_episode_subtitles_missing_total stays exported even with DISABLE_EPISODE_METRICS=true (sourced from bazarr's cheap badges endpoint), and bazarr_subtitles_missing_total always includes the episode count instead of misleadingly reporting movies-only (#407).
• REQUEST_TIMEOUT / --request-timeout (default 60s) caps each request to the target app.
• Huge-library relief: DISABLE_HISTORY_METRICS and DISABLE_WANTED_METRICS skip the endpoints whose totals force full table counts (the queries that can hang a multi-year instance's UI during scrapes). Combined with v3's pageSize=1, no-sort requests and partial scrapes, large sonarr/radarr instances scrape reliably again.
• Resiliency: a panic inside any collector — including concurrent fan-out workers — now degrades to that collector's error gauge instead of crashing the exporter, and scraping an empty instance (e.g. bazarr with no series, #244) is clean.
• Retries of failed requests now back off with jitter (~250ms, then ~500ms) instead of re-sending immediately, giving a struggling instance breathing room.
• The exporter exposes its own runtime metrics (go_*, process_*), so its CPU, memory, and GC behavior are visible alongside the app metrics.
• Large-library scrapes are dramatically faster: per-series/artist lookups now run concurrently, and the per-item metrics on a ~500-series library drop from ~30s to a few seconds.