# Changelog

Changes in Readarr Redux since it forked from Readarr (upstream commit `0b79d3000`, 2025-06-27). Upstream's own history is in the git log.

Versions are `0.5.0.<CI run number>`. Only pushes to `develop` publish an image, so the numbers have gaps. Each entry is the image tag and the day it was built. The commit messages explain each fix in detail.

## Unreleased

Release readiness: a bug, security and dependency pass before the first Docker Hub and Unraid release.

- Fixed: A file skipped because another file in the same download had taken its part was reported as "Book has already been imported", so a stuck download looked safely imported. It now names the file that came first, and the queue shows it as "Part taken by another file" instead of a bare Warning.
- Changed: Queue status chips are icons in fixed columns that line up across rows, on the right of the row with its buttons. Compare, Import anyway and Blocklist and Search come after the chips, then the row's own buttons, so Remove stays at the right edge. The first column keeps only the status icon. Hovering a chip shows its name and the messages behind it.

- Fixed: Authors and books deleted without their files came back on the next scheduled Rescan Folders, monitored. Rescan Folders skips unchanged unmapped files again; an author's Refresh & Scan is what matches them again.
- Fixed: When part of a multi-part upgrade failed to import, the old files that went back in place were left out of the library, and the import told Connect scripts and webhooks they had been deleted.
- Fixed: Upgrading an ebook from a release with several formats (epub and azw3, say) counted the book as incomplete, so the old file was kept and could be left in a hidden `.readarr-upgrade` folder.
- Fixed: Tightening a metadata profile no longer removed the books it now leaves out, and each of them cost a metadata lookup on every refresh. They go the way of a book the source no longer has again: removed after a refresh 12 hours later, never with files or when edited by hand.
- Fixed: An author's "last book" was usually empty, and the next book could be wrong, because the query paired the lowest book id with the latest date.
- Fixed: Refresh logged a series you started by hand, or one whose books you all took out, as "not found in metadata and is being deleted". Series you started are left alone by refresh.
- Fixed: Deleting a folder holding a hidden file such as `.DS_Store` waited 9 seconds per folder.
- Fixed: Deluge's Web UI was disconnected from its daemon on every queue poll while it listed a torrent with no name.
- Fixed: A wrong qBittorrent password found while logging in again was reported as a connection error.
- Fixed: Losing the connection to Readarr briefly re-ran open interactive searches on every indexer and threw away the picks made in an open Manual Import.
- Fixed: The root folder health check linked to a page that doesn't exist, and a bad log file address showed an empty page.
- Security: The UI now uses react-router 7.18.4. The 7.18.2 bump never took effect, because react-router-dom kept its own 7.15.1 copy. lodash and @babel/runtime are on patched versions too.
- Security: .NET 10.0.13 runtime (October 2026 security release). ImageSharp 4.1.3, which fixes the five advisories 3.1.12 carried.
- Security: The container log (`docker logs`, the Unraid log viewer) has secrets such as API keys removed, as the log files already did.
- Security: Login throttling counts an IPv6 client by its /64, and a throttle table full of bans drops the oldest instead of refusing every new client.
- Security: A wrong username takes as long to reject as a wrong password, so login timing no longer reveals the username. Control characters in a username are escaped in the auth log, so they can't forge lines for fail2ban.
- Security: A malformed Basic auth header gets a 401 instead of a 500 with a stack trace.
- Security: Backups, log files and `initialize.json` are never sent as cacheable, even with a `?h=` query.
- Security: External links only open http and https addresses.
- Performance: The book list is no longer fetched again on every window focus, every cover costs one disk check instead of three, and the author and book lists build their lookup maps once instead of once per row.
- Docker: Images are also published to Docker Hub as `dervish/readarr-redux`. `latest` now means the newest release tag; follow `develop` for every merge. The runtime image is Debian 13 (trixie), and the frontend builds on Node 24.
- Unraid: A Community Applications template, published from [dervish666/unraid-templates](https://github.com/dervish666/unraid-templates).

The last redux state (app state, sidebar messages, UI settings, system status, the captcha and OAuth inputs) moved to zustand and react-query, and the router moved to React Router 7 (frontend migration stage 4). Nothing should look or work differently, and no saved setting resets.

- Changed: A page address with extra parts after a real page, such as `/calendar/foo`, now shows Not Found instead of the page it starts with.

The author, book, bookshelf, add new, calendar, wanted, search, import and editor pages moved to react-query with the Sonarr v5 look (frontend migration stage 3).

- Changed: Saved columns, sort, filters and view choices on those pages are reset once, because they now live in new per-page stores.
- Fixed: A failed author list load no longer sends the author page back to the index as if the author were deleted, and no longer clears row selections on any list.
- Fixed: The calendar legend shows again, and the calendar's midnight "today" timer no longer spins.
- Fixed: Manual import stops when a checked row has no author, book or quality, instead of importing the rows before it.
- Fixed: The book edit modal shows the edition picker when opened from the book index too.

Fixes from the 2026-10-17 audit's Medium and Low findings.

- Fixed: A book the metadata source can't find is no longer deleted on the first not-found. A refresh at least 12 hours later has to miss it too. The wait starts again after a restart, which only ever delays a delete.
- Fixed: Refresh no longer deletes a book whose title, date, pages, overview or series you edited by hand.
- Fixed: A refresh running while you saved a book's metadata could write its older copy back over your edit.
- Fixed: Clearing an author's overview could bring your old text back from the metadata cache.
- Fixed: A bad series row in a metadata edit, such as an id of -1 or a new series with no title, deleted the book's series links. Every row is now checked before anything is saved.
- Fixed: A new edition added during an import showed the metadata source's title, overview and pages instead of the ones you set.
- Fixed: When saving an import's new rows failed, Import and replace left the other book without its file. The whole import is now undone: new files go back to the download, the replaced files go back in place, and their library entries come back.
- Fixed: Refreshing a single book left the cached author with a works list of just that book, so the next author refresh looked up every other book one at a time.
- Fixed: Deleting a book file could delete its author folder outright, past the recycle bin, when the folder held only hidden files or dot-folders such as an upgrade's `.readarr-upgrade`.
- Fixed: Recycle bin cleanup, recycling a folder, cross-drive folder moves and empty-folder tidying went through symlinks and junctions, so cleanup could delete old files wherever a recycled link pointed, the library included.
- Fixed: Audio written into a download folder after the import's clean-up check was deleted with the folder. The clean-up now deletes only the files it listed, never audio or books, and removes a folder only when it is empty.
- Security: Covers, cover proxying and metadata extras images only connect to public addresses, checked at connect time and on every redirect. The metadata source's own host is the one exception, for self-hosted setups. A cover server on another LAN host no longer works.
- Security: Metadata responses stop at 64 MB and covers at 20 MB, counted after decompression, so a broken or hostile server can't fill memory or disk.
- Security: With no login set up, the web UI now also asks for a one-time setup code, which Readarr writes to its console and log at every start. Before, a reverse proxy on the same machine that sends no `X-Forwarded-For` made every visitor look local, and they got the API key. Entering the code once lets that browser in to set up a login. To keep running with no login on purpose, set `AuthenticationNoneConfirmed` to `True` in `config.xml`, which drops the code and the setup prompt and leaves a warning in System > Status. Requests with the API key are unchanged.
- Security: The Docker entrypoint no longer follows symlinks in `/config` when it fixes ownership, so it can't hand files outside `/config` to PUID:PGID.
- Fixed: A failed Plex sign-in poll left the sign-in button spinning for good. Polling also gives up after ten minutes.
- Fixed: Editing the metadata source address while a save was still running lost the edit.
- Fixed: A failed Import anyway or Import and replace request showed no error. The queue row shows it, and the compare dialog stays open with the error.
- Fixed: The author overview field disappeared without a word when it failed to load. It now says so, with a Reload button.
- Build: CI and the Docker build read a Six Labors licence key from the `SIXLABORS_LICENSE_KEY` secret, ready for ImageSharp 4. ImageSharp stays on 3.1.12 until the secret is set.

## 0.5.0.94 (2026-10-17)

Fixes from the 2026-10-17 audit's High findings.

- Fixed: A multi-part audiobook download was marked imported, and with Remove Completed Downloads on deleted, while some of its parts had not imported. A download now counts as imported only when every audio file under its folder has imported or was moved or deleted by the same import.
- Fixed: Moving an author to its canonical id could orphan books that housekeeping then deleted. The books and the author now move in one transaction, and Cleanup Orphaned Books checks again before it deletes.
- Security: With no login set up (authentication None), only requests from the local network get in, and System > Status warns about it. Anything else gets a 401 instead of the API key.
- Security: Covers must be JPEG, PNG, GIF, WebP or AVIF by name and by content. Anything else is discarded, and covers are served with `nosniff` and a Content Security Policy.
- Fixed: Deleting a folder no longer goes through directory links inside it.

**More downloads now stay in the queue as Import Failed and need removing by hand.** Readarr no longer calls a download imported while anything in it could be lost. That covers any audio that didn't import, including audio in `extras/`, dot-folders and sample or bonus folders; "Not an upgrade" duplicates; downloads whose history predates this release; folders that are missing or can't be read; partial scans; and folders containing links. Check the queue after upgrading and remove what you have dealt with.

## 0.5.0.91 (2026-10-16)

Editing metadata by hand. Migration 044.

- New: Edit Book has a Metadata section for title, release date, pages and overview. Values set there are kept over the metadata source's on every refresh, and clearing a field brings the source's value back. A changed title is also what downloaded files are matched against. Files keep their tags until you retag them.
- New: Edit Book also edits the book's series: add it to any of the author's series at a position, start a new series, change its position, or take it out. A refresh keeps all of it. Series links from the metadata source are no longer deleted just because the source stops listing them, if you added or moved them yourself.
- New: Edit Author has an Overview field, kept over the metadata source's, for authors the source has no bio for.
- Fixed: Saving the Edit Book dialog without changing the edition failed. Saving it also wrote the browser's copy of every edition back over the server's.

## 0.5.0.87 (2026-10-16)

Queue triage: dealing with a failed import without hovering or leaving the queue.

- New: The queue shows why each item is stuck as chips next to its status icon ("Match 61.4%", "Destination exists", "Not an upgrade", "Waiting to import" and so on). Hover a chip for every message behind it. The status icon's popover is still there.
- New: Import anyway, on a download that failed the minimum book match. It imports into the book it was grabbed for, and every other import check still applies.
- New: Blocklist and Search, on the same rows. It removes the download from the client, blocklists the release and searches for another, after a confirmation.
- New: Compare, on a download that failed because a file is already at its destination. It shows the download and the file in its way side by side: name, folder, size, date, quality, tags, and which book the library has that file as. Import and replace imports the download over it. The file in its way is set aside like an upgrade's old file, put back if the import fails, and recycled once it succeeds. Its library entry, if it has one, is removed.

## 0.5.0.81 (2026-10-16)

- Fixed: Manual Import listing the same file several times when several files in a folder matched the same book. The copies ticked together and a rejected copy kept Import disabled, so a mis-tagged file couldn't be moved to its right book.

## 0.5.0.78 (2026-10-16)

- Fixed: Rescan Folders skipping unmapped files that hadn't changed on disk, so a file left unmapped stayed unmapped even after a matching fix. Unchanged unmapped files are now matched again on every rescan.

## 0.5.0.58 to 0.5.0.77 (2026-10-16)

- Dependency updates only: js-yaml, source-map-js, fast-uri, immutable, moment, qs, browserslist, nanoid, postcss, websocket-driver and `@babel/core` in the UI, and `actions/setup-node` in CI. Builds that a newer merge cancelled published no image, so their changes arrived in the next one.

## 0.5.0.47 (2026-10-16)

- Fixed: Rescans leaving a file unmapped when it names its book by series position ("Defiance of the Fall 1", "Mistborn #2", "The Expanse, Book 3"), common for book 1 of a long series titled after the series. Such a file could also come out closest to a later book in the series, which the rescan then rightly refused. Files already unmapped are only picked up again from 0.5.0.78.

## 0.5.0.34 (2026-10-16)

Download-client and integration parity with Sonarr. Migration 043.

- Fixed: qBittorrent 5.2 login with a username and password. qBittorrent API keys are supported, and a release qBittorrent already has is rejected instead of failing the grab. qBittorrent and NZBGet back off properly when they're unreachable.
- Fixed: Torrent clients moved files out of finished torrents even with Remove Completed Downloads off. Those imports now copy or hardlink, so the torrent keeps seeding and its files stay in the download folder.
- New: Transmission 4 labels, and a post-import label. Torrents added before an upgrade from Transmission 3 still match by category folder.
- Fixed: Deluge reconnects to its daemon when torrents come back without names, and detects the Label plugin reliably.
- Fixed: SABnzbd 4.1 sorters are checked when testing the client.
- Fixed: HTML-encoded release titles. Nzb.su indexers move to nzb.life (migration 043); the dead NZBCat and SimplyNZBs presets are gone.
- New: Settings > Media Management > Minimum Book Match (advanced), the match percentage needed to import automatically. Default 80, as before.
- Fixed: Plex sign-in in browsers that enforce Cross-Origin-Opener-Policy.
- Removed: Twitter notifications and the Goodreads Owned Books list and notification. Saved ones are dropped at startup with a warning.
- Fixed: Goodreads OAuth signing payloads, which carry the token secret, no longer reach the trace log. The README now explains that Goodreads Bookshelf signs requests through auth.servarr.com.
- Fixed: Discord timestamps on systems with a non-Gregorian calendar, Goodreads list failures being swallowed, and Calibre conversion checks that could loop forever.
- Security: `System.Security.Cryptography.Pkcs` pinned to 10.0.13 for GHSA-846f-f764-76mx and GHSA-2fmv-rv2c-59x7. Without it the build fails NuGet's audit, which is why there is no 0.5.0.31.

## 0.5.0.22 (2026-10-12)

Migration 042.

- Fixed: Rescans linking untracked files to the wrong book and then writing that book's tags into them. One untracked file was outvoted by any book with more files of its own. A rescan now needs the edition's ISBN, ASIN or Goodreads id, or a title close to the book's own with the same numbers, and no other book fitting as well; otherwise the file stays unmapped. Files mis-tagged before this keep the wrong tags until they are imported again for the right book.
- Fixed: Retagged books importing again on every rescan.
- Fixed: A re-imported download leaving its earlier copy on the wrong book. The earlier copy now goes to the recycle bin.
- Fixed: Audiobook releases smaller than the quality's minimum size being grabbed, so single chapters and partial uploads stop coming in. Audio qualities default to a 10 MiB minimum (migration 042, only where none was set).
- New: Quality definitions show the audiobook minimum release size in MB. Ebook qualities read "Not size checked".

## 0.5.0.21 (2026-10-12)

- Fixed: Deleting a book with its files left the files and their database rows behind. The files now go to the recycle bin with their rows.
- Fixed: Rescan failing on a UNIQUE constraint on BookFiles.Path, which also stopped the scan before it added its unmapped files.

## 0.5.0.20 (2026-10-12)

- The vendored FluentMigrator no longer pulls in every database runner, so Microsoft SQL Server Compact, the Visual C++ 2008 runtime and `Microsoft.Data.SqlClient` are gone from the build and the image. Only SQLite and Postgres were ever used.
- New README, NOTICE.md, SECURITY.md and CONTRIBUTING.md for the fork.
- The header donate button and the Readarr donation tile are gone, since that money went to a retired project. The Sonarr, Radarr, Lidarr and Prowlarr tiles stay.
- System > Status says this is an unofficial fork and links to its source, issues and discussions, with the upstream wiki and repository marked as upstream. The metadata search help, update and API licence links point at the fork.

## 0.5.0.14 (2026-10-12)

- Runs on .NET 10, with the UI on React 18 and the SignalR 10 client.
- macOS: .NET 10 keeps app data in `~/Library/Application Support` instead of `~/.config`. On first start the old `~/.config/Readarr` folder is moved across, so installs no longer start with empty settings.
- Fixed: Retagging an audiobook whose metadata has no original release date crashing.
- Fixed: A failed tag write leaving an imported book file unmapped.
- Fixed: Importing a download with an unknown author crashing.

## 0.5.0.10 (2026-10-12)

- Fixed: Book covers stopping at the first book with no monitored edition.
- Fixed: Works the metadata source always fails on blocking the rest of an author's refresh, and being logged as errors on every refresh.
- Fixed: A metadata outage answering 500 getting past the limit on missing-book lookups.
- Fixed: Indexer connection resets logged as errors on every RSS sync.
- Fixed: Tidying empty folders failing, or skipping without a word, when a folder was already gone.

## 0.5.0.9 (2026-10-09)

- Fixed: Audiobook parts whose track tags clash being skipped without a word. Manual import now warns about them instead of rejecting them.
- Fixed: Filename part numbering hiding a missing part or overriding chapter numbers.
- Fixed: Manual import not flagging mixed editions.

## 0.5.0.8 (2026-10-09)

- Fixed: Merged books moving to another author or losing their monitoring.
- Fixed: Author refresh adding narrators and publishers as new authors.
- Fixed: Author refresh failing with "Can't insert model with existing ID != 0".
- Fixed: Folder rescan failing on a file with no quality.
- Fixed: Upgrades recycling the old book when a new part vanished mid-import.
- Fixed: Commands that give a quality as its id or name failing with a JsonException.
- Fixed: Re-processing a manual import row dropping the chosen edition.

## 0.5.0.7 (2026-10-09)

- Fixed: Books deleted when Goodreads merged them into a new work. Refresh now follows the merge.
- Fixed: Goodreads Series import-list books for existing authors never being added.
- Fixed: Book lookups rewriting cover URLs on cached metadata.

## 0.5.0.6 (2026-10-09)

- New: Settings > Metadata Source page with a Test button. Fresh installs no longer default to the dead `api.bookinfo.club`.
- New: Metadata source health check.
- Fixed: Search, ISBN and ASIN lookups and the Goodreads Series import list bypassing the configured metadata source.
- Fixed: A metadata outage failing every refresh even when a cached answer exists. During an outage search falls back to the last good answer, and refresh keeps the existing data.
- Fixed: Failed searches and ID lookups showing as "no results" or "not found" instead of an outage.
- Fixed: Rate-limited (429) or stalled metadata responses blocking a command thread, or holding an author refresh for hours.
- Fixed: Malformed or empty metadata answers becoming the cached copy, and daily housekeeping deleting the cached copies the fallback relies on.
- Fixed: Metadata IDs above 2,147,483,647 failing to deserialize.
- Fixed: Metadata links, Calibre identifiers and ebook identifiers always labelled as Goodreads, whatever the source.
- Fixed: The source check accepting sources the metadata client can't use, following redirects, reading responses of any size, and failing on URLs with credentials or non-ASCII hosts.
- Fixed: Release dates landing a day early on servers east of UTC.
- Fixed: One bad import-list item or changed-authors call aborting a whole run.

## 0.5.0.5 (2026-10-08)

Security:

- Fixed: SQL injection through the paging sortKey (CVE-2025-50983).
- Fixed: Archive entries written outside the extraction folder (zip-slip).
- Fixed: Forwarded headers bypassing "Disabled for Local Addresses".
- Fixed: Static file and cover requests resolving outside their folders.
- New: Salted PBKDF2 password hashing (220,000 iterations). Older hashes are rehashed on the next login, and host config no longer returns the hash.
- New: Repeated failed logins are throttled.
- New: Existing sessions are signed out when credentials change or the API key is reset.
- Fixed: Password hashes in SQL trace logs and failed-query diagnostics, and Plex client identifiers in logs.
- Fixed: Cover resizing reaching ImageSharp's vulnerable TIFF codec.

Imports and upgrades:

- Fixed: Downloads importing into a different book than the one grabbed, or into any book once the grabbed one was deleted.
- Fixed: Packs being imported as the single book they were grabbed for.
- Fixed: Upgrades deleting or recycling the old files before the new ones were in place and saved, or when a new part was rejected, unidentified or missing.
- Fixed: An upgrade removing a file another book had just imported, and concurrent imports deleting each other's backups.
- Fixed: Searching for a series' first book grabbing later books or omnibus releases from that series.

Other:

- Fixed: Author refreshes hammering a struggling metadata source.
- Fixed: Book refresh crashing when the metadata no longer has the book.
- Fixed: Author merges losing the duplicate author's history.
- Fixed: Failed database updates reporting success.

## 0.5.0.2 (2026-10-07)

The first fork build.

- Fixed: Authors deleted when the metadata source answers 404.
- Fixed: Books deleted because a truncated author works list left them out, or because their lookup returned them under a new id.
- Fixed: Editions with files deleted when the metadata source drops them.
- Fixed: Multi-part upgrades deleting the parts they just imported.
- Fixed: One book's editions breaking release parsing for a whole search.
- Fixed: Stop calling home to Servarr. Sentry, update checks, server-side notifications and the system time check are gone; updates come as new Docker images.
- New: GitHub Actions build and a Docker image at `ghcr.io/dervish666/readarr`.
- Packages that only existed on Servarr's feeds are vendored, so the build no longer needs those feeds. MailKit 4.16.0 and ImageSharp 3.1.12.