Skip to content

chore(deps): update container images (major) - #27

Closed
renovate[bot] wants to merge 1 commit into
mainfrom
renovate/major-container-images
Closed

renovate[bot] wants to merge 1 commit into
mainfrom
renovate/major-container-images

Conversation

@renovate

@renovate renovate Bot commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

This PR contains the following updates:

Package Update Change
glitchtip/glitchtip major 6.2 → 7.0
gotenberg/gotenberg major 8.40.0 → 9.5.0

Release Notes

gotenberg/gotenberg

v9.5.0: 9.5.0

Compare Source

New Features

  • signed in /forms/pdf/info (#​1728). Each PDF now says whether it carries a digital signature. Gotenberg doesn't check the signature, and an empty signature field doesn't count. Thanks @​MaikNeubert.
  • A0, A1 and A2 paper sizes (#​1733). paperSize takes A0, A1 and A2 on the Chromium and LibreOffice routes and in the compose stage. Thanks @​SerhiyBalan.

Chore

  • Updated litho to v0.0.11.
  • Updated Go dependencies.

v9.4.1: 9.4.1

Compare Source

Behavior Changes

  • A WebSocket the outbound policy refuses fails the conversion with --chromium-proxy-server. That proxy replaces Gotenberg's pinning proxy, the only layer that cuts off a WebSocket handshake, so a page could open a WebSocket to an address --chromium-deny-private-ips, --chromium-deny-public-ips or the deny-list refuses, and Gotenberg only logged it. The conversion now fails with a generic 403, as it already did for a deny rule that names a path. A Web Worker's WebSockets still reach your proxy unchecked: enforce the policy there too. Thanks @​manus-pi.

v9.4.0: 9.4.0

Compare Source

New Features

  • position on the stamp routes and stages (#​1725). /forms/pdf/stamp-text, /forms/pdf/stamp-image and /forms/pdf/stamp-pdf, and their stages, take center, top-left, top, top-right, left, right, bottom-left, bottom or bottom-right. Each page places the stamp as a viewer shows it, rotation included, so a document mixing page sizes and orientations gets the stamp in the same place on each. x and y then move it from there, right and up. A turned stamp stays on the page in a corner, and text is placed by the ink of its glyphs: DRAFT centers on its capitals. Thanks @​spoltix.

Bug Fixes

  • LibreOffice converted documents without a page size to Letter (#​1723). A spreadsheet, CSV or text file that defines no page size came out Letter in 9.x, since LibreOfficeKit sets the locale to en-US when it starts. It comes out A4 again, as in 8.x. A document that defines its page size keeps it. Thanks @​TomokiOdakaSecureNavi.

  • A cookie Chromium refuses answered 500, such as one whose path does not start with /. It now answers 400, naming the cookie by its index and name.

  • A hung or dead Chromium or LibreOffice took longer to recover (#​1726). A browser that stopped answering cost two requests their whole --api-timeout before its restart: the request that finds it hung now restarts it after one health probe, and succeeds. A LibreOffice process killed after a timeout now restarts right away rather than on the next request, so /health no longer reports it down meanwhile. Thanks @​justinba1010

Chore

  • Updated litho to v0.0.10.

v9.3.0: 9.3.0

Compare Source

New Features

  • --pdf-max-render-pixels (#​1722). Caps the pixels of a page the render stage draws, which holds 16 bytes of memory per pixel. 0 keeps litho's default, 50 million pixels: a Letter or A4 page at 600 dpi, about 560 MB. A larger page answers 400, which says to lower dpi or raise the flag. flatten-transparency composites at a lower resolution to fit instead. Thanks @​DeJean100.
  • --pdf-disable-stages (#​1722). The Chromium and LibreOffice conversion routes answer 400 to a stages field, and only convert. With --pdf-disable-routes, no request runs PDF processing. Thanks @​DeJean100.

Bug Fixes

  • extract-attachments answered 500 for an attachment whose data does not decode. It now answers 400, naming the attachment.
  • PDF errors repeated themselves. A render error read render page 1: page 1: page render exceeds its budget: render exceeds its budget: ..., and an extract-text error could name its page twice. Each now names its page and its failure once. So does a stamp-image error for an image that does not decode: invalid image data: png: invalid format.

Chore

  • Updated litho to v0.0.9.

v9.2.0: 9.2.0

Compare Source

New Features

  • compose (#​1663). /forms/pdf/compose draws pages of the uploaded PDFs on new pages of the size you choose: a label, a packing slip, a sheet of several pages. paperSize, or paperWidth and paperHeight, size the new pages; each entry of placements draws one page, with source, sourcePage, page, x, y, scale and rotation, x and y in points like a stamp. Every uploaded PDF must be placed. Annotations, form fields, bookmarks, tags and metadata of the sources are not carried. It is a stage too. Thanks @​twitf.
  • Exclusions and open ranges in page selections (#​1719). Every pages option takes ! to exclude a part, and ranges open at one end: !1 is every page but the first, 3- page 3 to the end, -3 pages 1 to 3, odd,!1 the odd pages after the first, !last every page but the last. A selection that leaves no page still answers 400. Thanks @​ericsu14.
  • downloadFrom names files without a Content-Disposition header. An entry takes filename, which wins over the server's. Without it, Gotenberg takes the filename of the Content-Disposition header, then the last segment of the URL path when it has an extension, as object stores such as S3 serve. A download nothing names still answers 400.

Bug Fixes

  • stamp-pdf drew rotated or cropped source pages shifted or turned. The stamp now draws the source page as a viewer shows it: its crop box, turned by its /Rotate.

Chore

  • Updated litho to v0.0.8.

v9.1.0: 9.1.0

Compare Source

New Features

  • extract-attachments (#​1718). /forms/pdf/extract-attachments returns the files a PDF embeds, every one or the one name gives, as <pdf>_<attachment>: invoice.pdf gives invoice_factur-x.xml. Several files come back in a ZIP. A PDF without attachments, or without the named one, answers 400. Thanks @​Betelgeuse1.
  • extract-text. /forms/pdf/extract-text returns the text of each PDF as a .txt file, pages joined by a form feed. pages selects pages; invisible, artifacts and annotations add invisible text, artifacts such as headers and stamps, and annotation text. Gotenberg does no OCR: a scanned page reads as empty.

Both are stages too: they come last in a stages list, after encrypt, sign or timestamp if any, and the response holds the PDFs and the extracted files.

Bug Fixes

  • backdrop-filter: blur() rendered blank on linux/ppc64le (#​1491). The ppc64le image pinned Chromium 146, which has the bug; it now installs the same Chromium as the other platforms. Thanks @​florianaiautomation.
  • Data race on concurrent stages. Stages that write many files at once, such as split or render on several PDFs, could register their output names concurrently.

Chore

  • linux/ppc64le moves from Chromium 146 to 156, from Debian's security updates.
  • Updated Go dependencies.

v9.0.0: 9.0.0

Compare Source

Gotenberg 9 is out 🎉

[!IMPORTANT]
latest now points to Gotenberg 9. Pin gotenberg/gotenberg:8 until you have upgraded. The upgrade guide maps every Gotenberg 8 route, field, flag, metric and span to its Gotenberg 9 counterpart.

One PDF engine

pdfcpu, QPDF, pdftk and ExifTool are replaced by litho, a PDF engine written in Go for Gotenberg, running in-process.

remediate repairs a PDF until it conforms to one of the eleven PDF/A levels, to PDF/UA-1 or PDF/UA-2, to both together, or to Factur-X, then checks its own result: Gotenberg-Remediation-Residual counts the violations left, Gotenberg-Remediation-Rolled-Back the fixes undone to keep pages, text and images. /forms/pdf/validate runs the same checks on any PDF.

litho also signs and timestamps, redacts from the file rather than from view, infers a structure tree, renders pages as images, fills and flattens forms in the scripts the image's fonts cover, and shrinks a PDF without touching image quality. A stage that would drop the signatures of a signed PDF answers 400, as does text no font can draw.

Stages, in your order

Post-processing is no longer flat fields in a fixed order. A stages list runs as written, on every route that makes a PDF and on /forms/pdf/pipe for PDFs you already have:

[{"op": "merge"}, {"op": "stamp-text", "text": "DRAFT"}, {"op": "remediate", "profile": "PDF/A-2b"}]

26 stages, each also a /forms/pdf/<op> route with form fields, plus six JSON read routes: info, validation, metadata, bookmarks, attachments and form fields. A broken order answers 400 before anything runs.

LibreOffice through lok

unoconverter, Python and the soffice listener are replaced by lok, a long-running LibreOfficeKit process that Gotenberg supervises and restarts after a crash. Macros stay off; --libreoffice-enable-macros runs them.

Fewer dependencies, less CVE surface

Java, Python, unoconverter, pdfcpu, QPDF, pdftk and ExifTool leave the image: fewer packages to patch, and no more CVE reports about tools Gotenberg does not ship. Variants, architectures, user, port and command don't change.

Upgrading

Start from the upgrade guide. /forms/pdfengines routes become /forms/pdf routes, post-processing fields become stages, and a field with a stage twin now starts with native. A removed route answers 404 naming its replacement, a Gotenberg 8 field 400 saying where it went:

Invalid form data: form field 'merge' is unknown to this route; Gotenberg 9 moved it to a stage: add {"op":"merge"} to 'stages'

Custom builds import github.com/gotenberg/gotenberg/v9. The PdfEngine, PdfEngineProvider and MetricsProvider interfaces and the Prometheus module are gone: OTEL_METRICS_EXPORTER=prometheus serves the same.

Everything else

PDF
  • set-form selects several list items, refuses a value a field cannot hold, and lists the PDF's fields when you name a missing one. flatten-form draws a value without appearance over the field's own background and border.
  • Flattened values and text stamps beyond Western European text use an installed font, embedded and subset. Uncovered text answers 400 naming the characters.
  • optimize packs objects into compressed streams, stores identical streams once, recompresses images losslessly and subsets fonts. Every pass runs unless turned off. An encrypted input needs a decrypt or encrypt stage first.
  • merge keeps each file's bookmarks, remapped, and names the result after the first file in alphanumeric order.
  • render draws pages as PNG or JPEG at the resolution you ask.
  • --pdf-max-stream-bytes, --pdf-max-embedded-file-bytes and --pdf-max-image-pixels bound what litho reads. --pdf-allow-list, --pdf-deny-list and their siblings filter the timestamp authorities it calls.
LibreOffice
  • nativePdfa takes PDF/A-1b, PDF/A-2b, PDF/A-3b or PDF/A-4; nativePdfua takes PDF/UA-1. The PDF/A export is no longer redone after the other steps: end the stages with remediate when they may break conformance.
  • initialView, magnification and pageLayout take names such as outline or fitWidth, not numbers. firstPageOnLeft, which LibreOffice ignored, and nativePdfFormats are gone.
  • nativeWatermarkColor takes #rrggbb, nativeWatermarkRotation degrees counter-clockwise, 0 being horizontal. A plain and a tiled watermark together answer 400.
  • CSV files split into columns. .epub is no longer accepted.
  • PDFs are named after their documents, report.docx.pdf, even for a single one. Gotenberg-Output-Filename still overrides.
Chromium
  • paperSize names a format, for LibreOffice too: A3, A4, A5, B4, B5, Letter, Legal or Tabloid.
  • waitForWindowStatus, renamed from waitWindowStatus, now waits as documented. skipNetworkAlmostIdleEvent=false now really waits for networkAlmostIdle.
  • The page sent to the HTML and Markdown routes passes the allow and deny lists, which keep governing what it loads.
  • Print settings Chromium rejects answer 400: a scale out of range, a negative length, a waitDelay longer than the API timeout. A wait that outlasts the request answers 503 naming the field and its value.
  • failOnHttpStatusCodes judges the main page by its frame, not its URL. Screenshot quality goes from 1 to 100.
  • DevTools messages leave the debug logs unless --chromium-debug-protocol is set. URL queries and extra header values stay out too.
API
  • archives unpacks uploaded ZIPs, bounded by --api-archive-max-entries and --api-archive-max-ratio.
  • Errors say what to send: the value sent and the one wanted, a stage option's expected value instead of a Go type, the host that does not resolve. NaN and infinities are refused.
  • Webhook and downloadFrom URLs must be absolute http or https URLs with a host. Gotenberg-Webhook-Events-Url is required with Gotenberg-Webhook-Url and receives errors as webhook.error events: the error URL, method and lists are gone. A failed upload says what the webhook answered, or why it was unreachable.
  • A non-ASCII Gotenberg-Output-Filename is encoded per RFC 6266. Every answer carries the trace header, 404 and 405 included; the version header too, without authentication.
  • The banner is an aligned startup summary, one log record outside a terminal. --api-bind-ip takes IPv6, --api-start-timeout bounds module readiness, /debug reports the lok and litho versions.
  • Debug logs mask form values, extra header values and document passwords.
Telemetry
  • Names follow the OpenTelemetry rules: gotenberg.chromium.requests, gotenberg.<engine>.queue.size, chromium.pdf, webhook.send. A failed request carries error.type; the status attribute and the errors counter are gone.
  • Each webhook and download attempt is an HTTP client span under the whole call's span, with server.address and a url.full stripped of credentials, query and fragment. Outbound URLs lose their query in errors and logs too.
  • The server span stays unset on success and 4xx, and sets Error on 5xx with the status code as error.type. OTEL_PROPAGATORS selects the propagators. Each written document is traced, its editing stages as child spans.
Security
  • The 8.40.0 fixes are in: IANA non-globally-reachable ranges are private for the *_DENY_PRIVATE_IPS checks, Chromium distrusts the compromised eMudhra roots from the first request, and the pinning proxy stays on with --chromium-host-resolver-rules.
  • Pinning proxy tunnels are bound to the server that accepted them. Refused outbound requests are logged with their context, redacted URL and rule.
  • Engine temporary files are collected by exact pattern, Chromium's profile directory is removed after a failed start, and handler panics answer 500 at once.
Removed
  • The deprecated --log-format, --log-enable-gcp-fields, --api-trace-header, --api-disable-health-check-logging, --chromium-incognito, --webhook-error-allow-list and --webhook-error-deny-list flags, and the --pdfengines-* and --prometheus-* ones.
  • UNOCONVERTER_BIN_PATH, replaced by LOK_BIN_PATH, and the pdfcpu, QPDF, pdftk and ExifTool paths.
  • downloadFrom[].embedded, facturxVersion and the merge bookmark options: the guide names their replacements.

Chore

  • Debian 13, Chromium 156 (146 on ppc64le), LibreOffice 26.8 from Debian's backports, Go 1.27, lok v0.0.6, litho v0.0.7.
  • CI runs the integration suite on every push to main and tests the Chromium and LibreOffice variants.
  • The Bruno collection mirrors every route.

Configuration

📅 Schedule: (in timezone Asia/Jakarta)

  • Branch creation
    • "before 5am on monday"
  • Automerge
    • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

👻 Immortal: This PR will be recreated if closed unmerged. Get config help if that's undesired.


  • If you want to rebase/retry this PR, check this box

This PR was generated by Mend Renovate. View the repository job log.

@renovate renovate Bot added the dependencies Pull requests that update a dependency file label Nov 14, 2026
@renovate
renovate Bot force-pushed the renovate/major-container-images branch from d302920 to e39467a Compare November 14, 2026 12:19
@mrrizkin

Copy link
Copy Markdown
Owner

Major container image upgrades need a separate migration and are now gated behind dependency-dashboard approval in renovate.json.

@mrrizkin mrrizkin closed this Nov 14, 2026
@mrrizkin
mrrizkin deleted the renovate/major-container-images branch November 14, 2026 23:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

dependencies Pull requests that update a dependency file

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant