# Restreaming Monitor upstream sources, handle recovery and rotate source credentials. After [registering a source](/guides/external-ingest), use its status and session history to manage recovery. Source state is independent of the broadcast outputs that consume its program. ## Follow the source lifecycle HLS sources progress through `created`, `probing`, `ready`, `connecting` and `live`. A lost source may enter `recovering`; `stopped` and `failed` end that session. Only observed canonical media establishes a live source. `restreamSources.get(sourceId)` returns the current session, recent health transitions, a safe `lastError` and observed usage. After a terminal failure, repair the input and start a new session. Avoid retry loops that ignore capacity or policy errors. ## Configure recovery and limits Sources default to a reconnect budget of three, four hours of duration and 2 GB of fetched bytes. The API permits zero to five reconnect retries and durations from one minute to 24 hours, bounded by the attached stream. Probes and retries count toward observed fetch usage. The deployment admits four source jobs globally by default. Probes also consume capacity. Increasing `STREAMING_MAX_SOURCE_JOBS` changes admission; it does not add source-worker resources. The fetcher bounds request time, redirects, manifest size, object size, discovered resources and concurrent requests. A source that exceeds a safety bound fails visibly. Do not bypass those checks by replacing the guarded worker binaries or disabling TLS verification. ## Change credentials or configuration Stop the source and wait for its active session to end before editing or probing it again. Updates require its current `version`. Its stream and protocol cannot be changed; register a replacement source instead. Upstream URLs and authorization are encrypted using the deployment's destination encryption key. Update `endpoint` or `authorization` to rotate them; an empty authorization value removes that header. Preserve the encryption key across restarts and backups. Replacing it without re-encryption makes stored credentials unreadable. Deleting a source stops ingest and dependent broadcasts, clears its saved authentication material, and preserves history and usage. It does not delete an already finalized recording. ## Diagnose failures | Code | Next step | | --- | --- | | `SOURCE_ADDRESS_REJECTED` | Check all hostnames, redirects, public DNS answers and permitted ports | | `SOURCE_BUSY` | Wait for the active probe/session to stop before changing configuration | | `SOURCE_CAPACITY` | Check active jobs and provisioned source-worker capacity | | `SOURCE_CONNECTION_TIMEOUT` | Check that the push encoder connects with a valid credential | | `SOURCE_PROFILE_UNSUPPORTED` | Check codecs and supported HLS playlist features | | `SOURCE_BYTE_LIMIT` | Review the source byte budget | | `SOURCE_FETCH_FAILED` | Check upstream availability, authentication and TLS | | `SOURCE_CANONICAL_FAILED` | Inspect source-worker health and the attached stream's readiness | Raw encoder diagnostics may contain upstream secrets and are intentionally excluded from public errors. Correlate source/session IDs with private operator logs instead of displaying those logs to viewers. ## Events and usage Subscribe to `restream.probing`, `restream.source_connected`, `restream.source_lost`, `restream.recovering`, `restream.stopped` and `restream.failed`. Verify signatures, deduplicate IDs and reload state after notifications. Fetched bytes include probes, manifests, initialization data and retries. Processing seconds are observed process elapsed time, not CPU-seconds. Push traffic belongs to normal stream usage; registry monitoring does not create another copy of that traffic. Keep these measurements separate from destination egress. See [Node methods](/reference/sdk/server/node/restream-sources), [React status](/reference/sdk/client/react/use-restream-source) and [streaming self-hosting](/guides/streaming-self-hosting).