A patch release for the import, search and request bugs reported since v1.51.0, plus a size limit on the cover cache and one small feature: adding a book straight from its Hardcover page. There are no database migrations and no new settings. Three behaviour changes are listed below.
Upgrade notes
- The cover cache is now limited to 2 GiB (
BINDERY_IMAGE_CACHE_MAX_MB, default2048). The cache lives under the data directory, which is the same volume as the database, and used to grow until old covers aged out after 30 days. A cache already over the limit is trimmed on the first sweep after you upgrade, and when it is full the oldest covers go first. Most libraries are nowhere near it. If your config volume is small, set the variable lower. The Helm chart exposes it underenv. - A request for a book that is already in the library now waits for an admin to approve it, even for an account with auto-approve on (#3908). Accounts with auto-approve still get instant approval for books the library does not hold.
- Move imports of flattened audiobooks keep files they did not import. A booklet in a subfolder, the cover in a disc folder or an archive volume now stays in the download folder, where it used to be deleted with it. A usenet job folder can therefore keep a few leftover files until you clear it. Copy and Hardlink imports are unchanged.
- Helm users: the chart's readiness probe now uses
/api/v1/health/ready(shipped in v1.51.0, so the image has it). A chart from this release needs an image of v1.51.0 or newer, so do not pinimage.tagto an older version. The NetworkPolicy also names the container port instead of usingservice.port, soservice.portcan be any value.
Added
- Add a book from its Hardcover page (#3839, #3883). When the Add Book search does not return a book that Hardcover has, paste the book's hardcover.app address or type
hc:and its slug, and Bindery fetches that one book from Hardcover and offers it like any other result. Manual Import's metadata search and Fix Match accept the same input. It needs a saved Hardcover API token. API clients can useGET /api/v1/book/lookup?hardcover=<slug>. Thanks NopeCharon for the change.
Changed
- The docs now say that OPDS still takes account passwords when local auth is off (#3915). With
BINDERY_LOCAL_AUTH_ENABLED=falsethe web login form andPOST /auth/loginare closed, but/opdskeeps accepting an account's username and password over HTTP Basic, because reading apps cannot follow an OIDC redirect. The behaviour is unchanged.docs/auth-oidc.mdhad said otherwise.
Fixed
- Titles with an apostrophe now search indexers that choke on one (#3895, #3904). Some indexers return nothing for any query containing an apostrophe, so a book like Carl's Doomsday Scenario was never found there. When the normal search finds nothing relevant, Bindery now also tries the title with the apostrophe removed and with it read as a space. Thanks chrisnlarsen for the report and the detailed query table.
- Adopting an audiobook no longer registers the ebook beside it (#3896, #3906). When one folder held both an epub and an m4b, adopting the audiobook row could pick the epub as the audiobook file, which then blocked the ebook row from being adopted. The lookup for an existing file now only considers files of the format being added. Thanks chrisnlarsen for the report.
- Move imports no longer fail when the download folder can't be cleaned up (#3897, #3905). If Bindery copied a finished download into the library but couldn't delete the source (for example a SABnzbd job folder it can read but not write to), the import was marked blocked, the copy stayed in the library, and the retry made a second copy as "Title (2)". Now the import counts as done, the source folder is left alone, and a warning in the log says so. Thanks chrisnlarsen.
- Requesters can ask for a book Bindery knows but has no file for (#3898, #3908). A search result showed "In your library" with nothing to click whenever the book only existed as metadata. Now the book shows a Select button and the request goes through, and approving it adds the missing format to the existing book. The badge stays only when both formats are covered, with a file or already monitored, and when just one format is missing the request dialog offers only that one. With tenancy on, a book that belongs to another account shows Select like any book you do not have, and the request is then refused as already in the library. Thanks flaevers for the report.
- Releases named "Title by Author" are no longer filtered out when the library holds a book titled with the author's name (#3902, #3911). If an author had a book like "Octavia E. Butler: The Last Interview", the search filter read her own name as another book and hid every "Survivor by Octavia E Butler" release. The author's name, or an alias, is now never treated as another title. Thanks digitalbile for the report.
- Flattened audiobook moves no longer delete files they never imported (#3913). With Move as the import mode, a multi-disc flatten or an audiobook file naming template brings over the tracks and the sidecars at the top of the download, then used to delete the whole download folder. Anything else inside a subfolder went with it: a booklet PDF in
Extras, the cover in each disc folder, an archive volume. So did a top level cover that failed to copy. Those files now stay in the download folder, the log lists them, and everything that did reach the library is still cleaned up. - The cover cache now has a size limit (#3914). Cached covers were only ever removed by age (30 days), so nothing stopped the cache from growing until it filled the config volume. It now stops at 2 GiB by default, counted the way the disk sees it (whole 4 KiB blocks per file). When it fills up, the oldest covers are removed to make room, and until that sweep has run covers are still fetched and shown as usual, just not saved. The sweep also runs shortly after startup now, where it used to wait a full day. Set
BINDERY_IMAGE_CACHE_MAX_MBto raise or lower the limit.GET /api/v1/system/statusreports it asimageCacheMaxBytesnext toimageCacheBytes. - Indexer daily query caps no longer forget requests after a database hiccup (#3935). If Bindery couldn't read its stored query counts for a moment (for example "database is locked") and then could, the searches it had already counted in the meantime were thrown away, so an indexer could go over its daily limit by up to about 30 seconds of requests. Those requests now stay counted and still get saved.
- Helm NetworkPolicy with a custom
service.port(#3909). WithnetworkPolicy.enabled=trueandservice.portset to anything other than 8787 (say 80), the policy opened the Service port on the pod, where nothing listens, so every connection to Bindery was dropped, including the PostSync smoke check. The policy now names the container port. - Upcoming date on the Wanted page was a day early west of UTC (#3910). The "not searched until" note on an Upcoming book read the release date in the browser's time zone, so anyone in the Americas saw the day before the real release day. It now shows the release day itself, in the same format as the book page.
- Helm chart: readiness probe on the database check (#3894). The probe moved to
/api/v1/health/readynow that the image serves it, so a pod whose database connection is unusable leaves the Service while it keeps running.