Skip to content

NzbDAV to InfiniDysk migration

Back to InfiniDysk setup.

At a glance

Only eligible legacy installations need this migration. Keep the backup and review the full procedure before authorizing downtime.

flowchart TD
    A{"Legacy installation?"} -->|No| B["Keep the canonical installation"]
    A -->|Yes| C["Back up and run preflight"]
    C --> D["Review targets and downtime"]
    D --> E["Authorize the namespace migration"]
    E --> F["Verify paths, scans and playback"]
    F --> G["Keep recovery data until verified"]

Migration from NzbDAV

Fresh DUMB installations use the canonical infinidysk service key, InfiniDysk process name, INFINIDYSK_* configuration prefix, /infinidysk runtime root, /mnt/debrid/infinidysk mount, and /mnt/debrid/infinidysk-symlinks library root.

Because that namespace is already canonical, a fresh installation is not migration-eligible and the dashboard does not show Review migration. The migration job API returns null until an actual guarded namespace job exists.

Existing installations are not silently moved. DUMB accepts the legacy nzbdav key, dependency tokens, paths, and NZBDAV_* variables, normalizes them in memory, and continues writing the legacy identity back to disk until the user opts in. If both top-level service keys are present, DUMB refuses to guess which entry owns the existing data.

The automatic migration reminder appears only while InfiniDysk/NzbDAV is enabled. Disabled legacy configuration defaults alone do not trigger it. Disabling the service preserves migration eligibility, job history, and rollback information; re-enabling it restores any otherwise-due reminder.

The dashboard presents enabled existing installations with Review migration and Remind me later. The reminder is stored by the backend, so it applies across browsers and reappears after seven days.

Create and verify an independent backup first

Before accepting either migration path, back up the DUMB configuration, InfiniDysk/NzbDAV state, Arr and Prowlarr databases, affected media-server configuration, and the symlink library to storage outside the paths DUMB will migrate. Confirm that the backup is readable and restorable. DUMB creates a private rollback bundle during migration, but that bundle is a recovery aid—not a replacement for an independent backup. The dashboard requires an explicit acknowledgement of this backup before it enables Migrate now.

Move from NzbDAV to InfiniDysk

Read the guide

Review the choices for moving an older NzbDAV setup to InfiniDysk.

Move from NzbDAV to InfiniDysk in the actual DUMB frontend, with example data.

Frontend 1.91.0 · DUMB 2.23.0 · Example data.

The available compatibility cutover changes the persisted service identity and managed process name to InfiniDysk. It can also rename DUMB-generated attached instance/process labels such as Radarr NzbDAV to Radarr InfiniDysk. Before writing those changes, DUMB saves a private copy of the complete configuration under /config/migrations/infinidysk-backups. Saved dashboard ordering, keyboard shortcuts, and notification service filters follow any renamed process labels. It does not change:

  • /nzbdav or another existing runtime path
  • /mnt/debrid/nzbdav or /mnt/debrid/nzbdav-symlinks
  • Arr categories, root folders, item paths, tags, or tag IDs
  • Arr import-list or Radarr collection root paths
  • media-server library paths
  • existing symlink targets

The exact former DUMB repository defaults are updated to infinidysk/infinidysk so installs and updates continue to resolve; intentional custom forks remain untouched. Restart DUMB after accepting the compatibility cutover when the service is enabled.

Run namespace migration before selecting PostgreSQL

Both compatibility-only identity migration and the complete path require InfiniDysk's active main provider to be SQLite. DUMB blocks them once PostgreSQL is selected because the complete workflow mutates the SQLite main store and a compatibility rename would disconnect guarded database rollback lookup. Do not toggle providers merely to bypass this check.

The complete namespace migration is offered only after preflight can inspect the participating services and prove that paths, mounts, and destinations can be moved safely. Resolve its blockers in the dialog before accepting downtime. Keep the required independent backup outside the migration paths.

Technical details: full namespace preflight inventory

The optional Migrate the complete InfiniDysk namespace path performs the remaining cutover as one guarded operation. It is not enabled until its preflight passes. The preflight makes no changes and checks:

  • InfiniDysk active reads can be measured;
  • InfiniDysk's effective repair.enable scheduler state can be inspected and changed through its authenticated API when necessary. An enabled value owned by an NZBDAV_CONFIG__... environment variable is a blocker: set that value to false, restart InfiniDysk, and rerun preflight so DUMB is not fighting an authoritative environment override;
  • every enabled Arr API can be inventoried. DUMB includes an Arr when its core_service metadata links it to InfiniDysk or live API inventory finds a legacy root/item path, import-list root, Radarr collection root, download-client/category reference, or nzbdav tag. This catches Prowlarr-managed Arrs whose metadata intentionally omits the InfiniDysk linkage. The preflight lists included and excluded Arr instances with the reason for each decision. Large Arr catalogs use a migration-specific 120-second API timeout; current queue entries are reported as pending quiescence rather than forcing the operator to empty every queue at the same instant;
  • every enabled InfiniDysk-linked rclone, Radarr, Sonarr, Lidarr, Whisparr, NeutArr, Profilarr, and Seerr instance is inventoried, including singular, list, and combined core_service/core_services values;
  • enabled Prowlarr instances are reachable so their Arr application connections and the shared legacy nzbdav tag can be inventoried. While an installation still uses the legacy namespace, normal DUMB setup reuses the existing nzbdav tag ID for InfiniDysk-linked applications instead of creating a competing infinidysk tag. Preflight blocks when both labels (or both legacy and canonical application names) already exist and would collide during rename. This can happen after an older failed migration attempt; reconcile the duplicate in Prowlarr, preserving the tag assignments you need, and run preflight again;
  • affected Plex, Jellyfin, and Emby activity and library APIs can be inspected. If no DUMB-managed media server is enabled and both dumb.plex_address and dumb.plex_token are configured, DUMB infers that Plex is external, verifies its server identity, and inventories its libraries through the Plex API;
  • destination paths do not contain conflicting data;
  • active mounts can be detached and every filesystem move is an atomic, rollback-safe rename; and
  • every detected legacy path is under a supported DUMB-managed root. This includes generated attached-service config and log paths such as /radarr/nzbdav and /log/rclone_w_nzbdav.log; paths outside managed roots remain blocked for manual review.

Before downtime begins, DUMB stores the complete DUMB configuration, relevant application configuration/database files, Arr, Prowlarr, and media-library API snapshots, scan-guard state, and a symlink manifest in a private mode-0700 bundle under /config/migrations/infinidysk-backups. The symlink snapshot catalogs link paths and target strings without checking target existence, so backup cannot trigger reads or metadata waits through the active rclone/FUSE mount.

These internal snapshots are designed to reverse this specific cutover. They cannot protect against pre-existing corruption, failure of the storage holding both the live data and rollback bundle, or unrelated application writes made during the migration.

When applied, DUMB first enters an automatic quiescence stage. It stops linked NeutArr, Seerr, Profilarr, and Prowlarr processes so they cannot add new Arr work. It also captures InfiniDysk's exact effective repair.enable value in the private rollback bundle, temporarily disables the internal scheduled health checks through InfiniDysk's API, and reads the value back before waiting on active reads. Health probes already in progress are allowed to finish, but the scheduler cannot continuously replace them. Each Arr remains available while its current queue drains; DUMB reports the remaining count in job progress and stops that Arr immediately after its queue reaches zero. If a failed, held, or otherwise stuck download prevents the queue from draining, use the still-running Arr UI to remove, retry, or resolve that entry while its producers remain stopped. DUMB never deletes or ignores queue items automatically. Affected media servers are handled independently. DUMB waits for playback to become idle and guards automatic scans. DUMB-managed servers are stopped as soon as they are safe. An inferred external Plex remains running because DUMB does not own its process; DUMB cancels/guards Plex scans through the API and requires it to remain idle. Pause Autoscan and any other external scan or request producer before cutover because the Plex token cannot stop those separate processes. DUMB then waits for InfiniDysk reads to drain and holds every managed stopped service through mutation and validation.

While the persisted job is waiting on verified active playback from a DUMB-managed media server, Open progress offers Stop active playback and continue. The action requires typing STOP ACTIVE PLAYBACK, immediately stops the listed media server, and therefore terminates its active streams. DUMB then waits for InfiniDysk active reads to reach zero before continuing. This override applies only to verified media playback: it cannot bypass Arr queues, unknown media/API state, active InfiniDysk reads, path conflicts, or any other safety check. Operators may instead leave the job waiting for playback to finish normally.

The playback-stop action is never offered for inferred external Plex. Stop its playback from Plex or the owning container/platform, pause Autoscan, and let the persisted job continue polling. The progress message distinguishes Arr queues, managed playback, external Plex activity, and remaining InfiniDysk reads so an active read alone is not mislabeled as a failed Arr queue.

Automatic quiescence waits up to one hour. A transient Arr API or database failure remains a visible pending condition and is retried during that window; it does not immediately trigger rollback. If activity remains, an Arr API stays unavailable through the deadline, or another non-retryable quiescence step fails, the job aborts before moving namespace paths, restores scan guards, and restarts only the processes DUMB stopped. This removes the requirement for the operator to pause every queue, automation producer, and playback session at exactly the same moment.

After a successful cutover DUMB restores and verifies the captured repair.enable value before restarting request/search producers. If any later stage fails, rollback restores the captured InfiniDysk database and verifies the same scheduler value before the job becomes terminal. The scheduler guard is recorded as infinidysk-health-check-guard.json in the private migration bundle.

After quiescence, DUMB detaches the old rclone mount, moves the runtime/mount/symlink/log paths plus discovered DUMB-managed attached-service paths, rewrites symlink targets and InfiniDysk configuration records, saves the canonical DUMB paths, and restarts the stack in provider-first order. Parent namespace moves and nested generated roots are planned separately: for example, after moving nzbdav-symlinks, DUMB still renames children such as radarr-nzbdav and sonarr-nzbdav inside the new parent. It then verifies that every planned legacy source is absent, every canonical destination has the expected type, and no raw symlink target retains the legacy namespace. Every linked service that was running before the cutover is restarted, and DUMB reruns setup for rclone, Arr, NeutArr, Profilarr, and Seerr so their generated integration state sees the canonical linkage. Services that were stopped remain stopped. Any running service whose DUMB process or instance label is renamed is also included even when its linkage is otherwise unrelated, preventing an old process from being orphaned under its former name. Generated provider category directories beneath completed-symlinks are SQLite-backed virtual paths, so DUMB rewrites the history, queue, and DAV category/path records that materialize them while the provider is stopped. Do not try to rename those paths through the live FUSE mount. The database is part of the private rollback bundle. DUMB then updates and verifies configured or live-reference-discovered Arr root folders, existing movie/series/artist paths, import-list roots, Radarr collection roots, managed download-client names/categories, Arr tag labels (without changing tag IDs), Prowlarr Arr application names and core-service tag labels (also without changing IDs), and Plex/Jellyfin/Emby library paths. Compatible Arr builds receive bounded bulk-editor batches for root-prefix-only item changes; unsupported editor APIs fall back to exact per-record updates. When a supported bulk editor reports a conflict, DUMB bisects that batch to isolate the conflicting records instead of converting an otherwise large catalog into hundreds or thousands of serial API writes. DUMB validates every resulting API item path and requires the canonical filesystem path for every file-bearing Arr item and every changed list/collection root before continuing, so a partial or incompatible bulk result still fails into the normal rollback path. The job reports completed/total counts for the current Arr and across all affected Arrs instead of holding one percentage for an entire large catalog. Plex identity, scan-guard, inventory, and library-path operations use an extended API response window because large servers can respond slowly during cutover. If a Plex library update times out after submission, DUMB does not blindly submit it again: it reconnects and accepts the change only when the library exposes the exact requested path set. An unverifiable result still fails into rollback.

Do not manually change Arr roots or scan a media library while the job is still running. DUMB keeps media-server scan guards in place until canonical Arr item directories and media-library paths have passed these checks. After the job reports success, confirm the new roots contain the expected symlinks, then run the normal Arr and media-server scans. Keep automatic trash/deletion disabled until the refreshed libraries and sample playback are verified.

Arr's Update All action refreshes existing records; it does not change root folders, import-list roots, or Radarr collection roots. If manual recovery is required after an older migration, use the Arr root-folder change workflow and choose No, I'll move the files myself because DUMB already moved the symlink library. Update import lists and Radarr collections separately before scanning, then run Update All only as the follow-up refresh.

After the full migration succeeds

Do not start this checklist unless the result is Namespace migration completed. A rolled-back, interrupted, or attention-required result needs its recovery details reviewed first.

  1. Confirm InfiniDysk, its rclone mount, every affected Arr, Prowlarr, and the affected media server report healthy.
  2. Open each affected Arr and confirm its canonical root is populated, sample existing items use the infinidysk-symlinks path, and System → Health has no missing-root or download-client path warning.
  3. Open the Arr's main library page and click Update All in its top toolbar: Movies → Update All in Radarr, Series → Update All in Sonarr, Artists → Update All in Lidarr, or the equivalent main-library Update All action in Whisparr. This is the circular-arrows refresh action beside the main library controls—not a System task or root-folder edit. Do not use Change Root Folder or ask Arr to move files after a successful migration—DUMB already updated the roots, existing item paths, import-list roots, and Radarr collection roots and moved the managed symlink namespace.
  4. In Prowlarr, confirm the affected Arr applications connect and the canonical infinidysk tag is present where the legacy nzbdav tag was used.
  5. Keep automatic trash/deletion disabled, then scan the affected Plex/Jellyfin/Emby libraries. DUMB updates and verifies their paths but does not start the scans.
  6. Compare expected library counts, inspect several items, and test playback plus seeking for at least one movie and one episode.
  7. Only after those checks pass, restore the operator's preferred automatic trash/deletion behavior. Retain the independent backup and DUMB rollback bundle for a suitable observation period.

The compatibility-only migration keeps all paths unchanged. After that path, verify InfiniDysk starts, attached services reconnect, and existing playback works; an Arr or media-library scan is not normally required.

Finalize the migration and remove recovery data

After the checklist passes and the migrated stack has remained healthy for a suitable observation period, the InfiniDysk service page can permanently purge DUMB-owned migration recovery material. This action is optional. It appears only when the backend advertises infinidysk_migration_cleanup, the latest recorded migration is a successful terminal result, and cleanup has not already been finalized. It is unavailable during active, failed, interrupted, rolled-back, or rollback-attention jobs.

Choose Review cleanup to request a short-lived preview. The preview reports counts, total bytes, user-facing deletion categories, retained categories, and its expiry without exposing private filesystem paths. Review it after every refresh because the token is bound to the current migration state. To proceed, acknowledge that:

  • the post-migration validation and playback checks passed; and
  • DUMB's rollback bundle and detailed migration history will no longer be available.

Then type the exact phrase REMOVE INFINIDYSK MIGRATION DATA and select the permanent cleanup action.

Cleanup deletes the detailed InfiniDysk namespace-migration state, preflight inventory, namespace job history, and DUMB-owned migration backups under /config/migrations/infinidysk-backups. It replaces the state with a small private mode-0600 tombstone containing only the selected mode, terminal status, and finalization metadata. DUMB retains:

  • current DUMB and InfiniDysk configuration;
  • the InfiniDysk runtime, application data, and databases;
  • mounts, symlink libraries, and normal symlink snapshots;
  • the minimal controller-owned PostgreSQL cutover authorization and database-migration job evidence under /config/arr-postgres-migration, which keeps restart and guarded database-rollback contracts verifiable; and
  • operator-managed backups outside the namespace rollback root.

Cleanup is irreversible

DUMB cannot reconstruct the deleted recovery bundle, detailed preflight, or job history. Keep independent operator backups somewhere outside the DUMB-owned migration backup directory. Cleanup does not delete service data or the separate PostgreSQL migration authorization/job store, but anything intentionally placed inside /config/migrations/infinidysk-backups is in the purge scope.

After finalization, the migration notice and manual migration action remain permanently hidden. Finalizing a compatibility-only migration records that the retained legacy paths are the accepted final namespace, so DUMB no longer prompts for the complete namespace workflow.

If migration or rollback fails

If a stage fails, DUMB stops the partially migrated stack, restores paths and saved files with their captured ownership and permissions, reapplies and validates the previous Arr/Prowlarr/media-library references, removes every Arr root path created only by the failed direction, and restores the exact captured symlink targets after the original symlink roots are back in place. Missing symlink roots, restore errors, target mismatches, or obsolete Arr roots make rollback report that attention is required rather than claiming recovery. Rollback restarts the original process names and publishes its own path, symlink-catalog, configuration, service, Arr, Prowlarr, media-library, and scan-guard progress stages. A terminal rolled-back job finishes at 100% and retains a redacted root-cause message; 100% means the recovery workflow finished, not that the requested cutover succeeded. When rollback needs attention, the dashboard lists the sanitized component errors and retained backup/config paths. Inspect those details and verify the legacy stack before performing a targeted repair; do not manually toggle service/provider settings, blindly restore the complete bundle, or rerun the migration.

The terminal namespace statuses have different meanings:

  • failed_rolled_back means DUMB's automatic rollback completed. Verify the legacy service, paths, integrations, and playback before generating a fresh preflight; do not reuse old preflight approval.
  • rollback_attention_required means one or more rollback surfaces could not be proven restored. DUMB freezes ordinary service/config lifecycle changes. Open the persisted result, inspect its component errors and retained paths, and repair only the named surface; this is not a signal to restore the whole bundle.
  • interrupted before any namespace filesystem mutation may allow DUMB to cold-start the known legacy topology so a new live preflight can be made. An interruption at an unknown or post-mutation stage remains frozen for recovery review. Never force a provider or path toggle to bypass either state.

Older DUMB builds did not retain file ownership metadata in this rollback manifest. If an older rollback reports that InfiniDysk/NzbDAV cannot read its configuration path, inspect the owner and mode of the named database and its SQLite -wal/-shm sidecars. Restore only the affected files to the configured PUID/PGID and their prior restrictive mode before restarting DUMB; do not recursively change ownership across the blob, mount, or symlink trees.

Combined relationships remain combined: for example, decypharr, nzbdav becomes decypharr,infinidysk; DUMB does not discard the Decypharr relationship. Zurg is deliberately not an InfiniDysk migration consumer. Its core_service field is used by other workflows and does not make Zurg part of this cutover.

Progress, interruption, and final paths

The complete cutover runs as a backend-persisted background job. The migration dialog shows the current stage, percentage, and recent stage history. Closing the dialog or navigating to another page does not cancel the job; a persistent InfiniDysk migration running banner shows the latest progress and reopens the dialog. Reloading the dashboard or signing in from another browser/device reattaches to the same backend-owned job. If a browser observes it finish while the dialog is closed, that browser shows a dismissible result banner once. A terminal job first discovered after a later reload, sign-in, or focus event is not announced again as a new completion; its persisted result remains manually reopenable from the InfiniDysk service page.

The live job record and full preflight inventory are stored as separate private mode-0600 sidecars under /config/migrations. This keeps job-status polling small and responsive even when the rollback inventory contains tens of thousands of Arr records; polling never rewrites the complete inventory.

Do not restart the DUMB container while the migration is active. A backend restart cannot safely resume an in-process filesystem/API cutover, so DUMB marks the retained job interrupted and directs the operator to inspect its private backup bundle and current service paths before retrying. Normal dialog closure and dashboard navigation do not interrupt it.

The final paths are /infinidysk, /log/infinidysk.log, /mnt/debrid/infinidysk, and /mnt/debrid/infinidysk-symlinks. Historical snapshot filenames are retained; future scheduled snapshots use infinidysk-{timestamp}.json. DUMB intentionally does not launch Arr or media-server scans after success. Review service health, then scan the affected libraries so their availability state is refreshed.

Preflight must be current

A passing preflight expires after 30 minutes. DUMB repeats structural/API checks before starting and then owns the live quiescence window. Queue, playback, and active-read activity shown as pending conditions does not invalidate the preflight; it must drain during the one-hour automatic quiescence stage before DUMB moves any path. The explicit playback-stop action interrupts viewers but does not waive the active-read drain.