Commit Graph
16 Commits
Author SHA1 Message Date
Marc-André Bombeck 518ce2fd37 fix(documents): read and render PDFs in a bounded child process
pdf.js decodes an image at the size the file declares, whatever size the
page is drawn at, and the server opened documents without an image limit.
A small upload declaring a very large image made the web process allocate
gigabytes until the container was killed, for every user at once, and the
thumbnail job ran the same document again after each restart. An image
whose soft mask declares the size does the same past pdf.js' own limit, and
nothing bounded how long a text read could take.

pdf.js is now opened with maxImageSize (20 million pixels, which keeps one
image inside the reader's ceiling and admits a 300 dpi A3 scan), and every
read and render runs in a child process, one at a time: its V8 heap is
capped, its resident memory is sampled and the child killed past 640 MB, on
Linux it volunteers to the out-of-memory killer first, a time limit ends a
read that runs on, and it starts with an almost empty environment. The
loader and the rendering moved into pdf-engine.mjs, which the child runs
with plain Node and which is traced into the standalone image.

A document that crosses a bound answers resource-limit. The thumbnail job
records it as a final failure, so the backfill does not queue it again.
2026-10-25 13:52:04 +01:00
Marc-André Bombeck 540c6bfaeb refactor(documents): read the text layer through the one shared loader
The text read and the rasterizer each opened pdf.js on their own: the text
read with bare options, the rasterizer through src/lib/documents/pdfjs.ts
with the standard font, CMap and wasm directories. The loader now serves
both, so a CID font with a predefined CMap is mapped to characters on the
text path too, and there is one set of options to keep in step.

The loader is typed with pdf.js' own declarations instead of a hand-written
shape. That drops isEvalSupported, which pdf.js 6 no longer has and which did
nothing. Only the render slice the DOM-typed declarations cannot describe
(the native canvas factory) keeps a named local type in the rasterizer.

The run joiner with its named steps moves into the loader as joinTextRuns,
and readPdfText is its one page walk; the second layout copy that mimicked
the previous library's tabs and blank lines is gone. extractPdfText keeps
its outcomes and page cap on top of it, and a failed read still says what
pdf.js threw (documents.local_extract.failed). A test draws a word in two
fonts into a real PDF and reads it back whole.

Refs #1277
2026-10-24 20:16:01 +02:00
Marc-André Bombeck 0b1757ff6f fix(documents): read the text layer through the same pdf.js that renders
Under Node, pdf.js runs its worker in the server process and keeps it on
a global, and whichever copy of pdf.js opens a document first owns that
global until a restart. The local text read went through pdf-parse, which
pins its own pdfjs-dist 5.4.296; the rasterizer uses the app's 6.4.299. A
text read first made every later rasterization fail with "The API version
does not match the Worker version" (thumbnails raster-failed, the vision
read of a scan 422), and a rasterization first made every later local text
read fail, which left the document out of the search index with nothing in
the logs.

src/lib/documents/pdfjs.ts is now the one place pdf.js is loaded and
opened, with one set of options (asset directories, no eval, no system
fonts). The text read walks the pages with getTextContent and lays the
text out the way pdf-parse did with the options it was called with: line
breaks, tabs between far-apart items, a blank line after each page, the
200-page cap. Text extracted from the PDF fixtures and from generated lab
tables, encrypted files and doctor reports is identical before and after.

A failed text read now emits documents.local_extract.failed with the
error pdf.js threw, and annotates the index event with it, instead of
returning a bare error.

The report tests read PDF text through the same reader. Two tests, one per
order, run a text read and a rasterization in one process; both failed
before this change.

Refs #1277
2026-10-24 19:32:13 +02:00
Marc-André Bombeck 4c50f48b26 fix(labs): name a failed re-check as an error and close the last budget gaps
A re-check or reservation that failed itself stopped the scan as
"unavailable", and the review screen said AI had been turned off. It now
stops as stoppedBy error with its own message in every locale, and the
failure is reported as one: an error signal and the worker error
reporter. A scan that failed that way before any row answers 500
aiRuns.failed.

The overspend booked before a further read is now net of the cached
share, as the settlement books it. A reservation that names no known
owner stops the scan instead of taking the user plan's larger cap.

A page's text layer is cut at the last whole line; only a page whose
first line alone runs past the cap keeps part of it, cut at a space or
hard, rather than collapsing to the bare cut mark, and an endless line is
no longer read past the cap. The unused per-call text limit is gone. The
run result's description names every outcome a stopped scan can have.
2026-10-21 12:28:32 +02:00
Marc-André Bombeck bfc7991935 fix(labs): treat a PDF's text layer as a fenced hint and keep the day's counter current
The text layer went to the model as the authority for digits. It is not
one: a scanned PDF often carries an invisible OCR layer with its own
mistakes, a subset font can map to the wrong characters, and a crafted
PDF can carry invisible text that differs from what the reviewer sees.
The prompt now calls it a hint only, makes the image authoritative, forbids
adding a value the image does not show, and asks for a low value
confidence where the two disagree. The layer is fenced with the document
text markers, which it cannot close, and so is text mode's OCR'd text.
Each page's layer is capped while it is collected, and a read shares its
text budget out per page, cut at a line and marked, so a long first page
no longer pushes the later ones out.

A further read now first books what the reads so far cost beyond their
reservation, then its own ceiling, both against the cap of the owner the
route judged the first read by, so the counter the next read is judged
against never lags. The wire is re-checked before every read, the first
included. A re-check or reservation that itself fails stops the scan with
the pages read rather than failing it, and a scan stopped before any row
answers with the stop's reason (budget, or the wire's own refusal) instead
of an unreadable page.
2026-10-21 11:33:37 +02:00
Marc-André Bombeck 79d905f089 fix(labs): reserve and re-check each page of a scan, and read digits from the text layer
The route reserved one read's budget and checked the wire once, and a
20-page report then ran up to 40 calls for many minutes on that single
admission. Every read after the first now reserves its own share of the
day's budget and asks the wire again (aiEgressRefusal) before it is sent;
when either refuses, the scan stops there, keeps the pages already read
and names the rest with stoppedBy budget or unavailable. The run settles
against everything it reserved, with the provider's reported usage net of
its cached share, and a scan that fails after several calls is charged
for them instead of refunded in full.

A rendered PDF page now goes out with its own text layer, read by pdfjs
beside the image, so a digitally produced report's digits, decimal
separators and units come from its characters rather than from a 1600 px
picture. Splitting the PDF into single-page documents would need a PDF
writer the project does not ship; the text layer needs none and helps
every provider. A scanned page has no text layer and reads as before.

Codex's cut reply keeps the text it streamed (on a non-enumerable
property, so no log or event serialises it), and its closed rows are
salvaged like every other provider's. The merge folds a reading only into
one an earlier page listed, never two rows of the same page, and the
printed range is part of the key. OCR'd text the deadline cannot split
small enough for one read is refused up front as labs.ocr.tooLarge.
2026-10-21 11:10:02 +02:00
Marc-André Bombeck 10cfdf8b66 fix(labs): read a lab report page by page and keep a cut-off read's rows
A scan sent the whole report to one call under an 8000-token ceiling and
validated the reply against a 100-row envelope. A long multi-panel report
lost everything either way: the reply stopped mid-row and failed to parse,
which the person saw as "try a clearer photo", or it held more than 100
rows and the schema refused all of them. One malformed row refused the lot
as well. Reasoning models made it worse, because the call sent no
reasoning setting and OpenAI, Codex and Claude's 5 generation reason by
default against the same output allowance.

A PDF of more than one page is now rendered and read a page per call on
every provider (up to 20 pages), as many reads as the run's deadline
allows, and the pages are merged in order with a reading printed twice
listed once. A photo, a one-page PDF on a provider that reads PDFs, and
OCR'd text up to 4000 characters stay one read; longer text is split at
line boundaries. Each read carries an explicit reasoning off, so a model
that can stop reasoning does and one that cannot gets its minimum budget
on top of the answer's. A reply cut at the output limit is detected from
the finish reason (Codex's incomplete stream included), keeps the rows
that closed before the cut and is named in the result instead of retried;
when no read produced a row the run fails as labs.ocr.outputLimit. A
provider failure after the first page keeps the pages already read.
Rows are validated one by one, the row cap is 400 for the scan and the
commit, and a report past it keeps its first 400 and says how many it
had. The run settles what the provider reported it spent rather than the
fixed reservation, so a page-by-page read still counts against the day's
budget in full.
2026-10-21 10:40:45 +02:00
Marc-André Bombeck 4b73099e75 fix(documents): render PDFs with pdfjs font data and refuse blank pages
Refs #1124. The rasterizer opened pdfjs without its standard font, CMap
and wasm directories and told it to use system fonts. The runtime image
has none, so any text pdfjs could not draw from an embedded program (a
non-embedded standard font, or an embedded one it cannot parse) came out
as nothing, and an image-only provider was sent the page background. A
workstation hides this because its own fonts stand in.

pdfjs now gets the directories from its own package, resolved at runtime
from the working directory, and Liberation Sans from that package is
registered as the canvas' sans-serif, serif and monospace faces for the
text pdfjs draws with a substitute.

A document whose every rendered page is blank now reports render-failed,
so the read records raster-failed instead of indexing a transcript of an
empty page as a success.
2026-10-04 10:37:24 +02:00
Marc-André Bombeck 5f548d331f fix(labs): say when a scanned PDF was read from its first pages only
A PDF longer than the pages a scan renders was read from its first ten
pages, and the cut was recorded only in the wide event. The person checking
the extracted rows had no way to know that values on later pages were never
read.

The rasterizer now returns the PDF's page count, the extract response
carries pageCoverage { read, total } when the cap applied, and the review
screen says how many of the pages were read.
2026-10-01 10:05:35 +02:00
Marc-André Bombeck 589b35589b Prove the raster-failed chain and name the fax-codec cause
The residual scan from the report is a CCITT Group 4 single-image PDF,
the codec class pdfjs rasterization is known to fail on. Chain checks:
a pdfjs open failure or throw already lands as render-failed with the
error text annotated; the zero-pages shape now annotates too instead
of returning silently. An end-to-end test drives the REAL rasterizer
through prepareVisionInput and pins the rasterFailed mapping; the test
header documents why a tiny in-repo CCITT fixture is deliberately not
checked in (probed against this pdfjs version, it renders the page as
blank instead of throwing, so the fixture cannot deterministically
fail and would be flaky across upgrades). The detail view's
raster-failed line now names scanned fax compression formats as a
known cause, in all six locales.

Refs #776
2026-08-13 13:04:41 +02:00
Marc-André Bombeck 893db887e9 Record why a document's index attempt failed and say so
A content-index attempt that failed (raster failure, a scan with no
text layer, a provider error) recorded its reason only in the server's
observability stream. In the vault, hasContentIndex false looked
identical for a document nobody ever tried and one that failed three
times, and the detail sheet had nothing to say about either.

Apply the summaryState pattern to the index. Two additive columns on
InboundDocument (migration 0333) record when the last attempt finished
and why it produced no index, as a closed reason set: local-empty,
local-unsupported, decrypt-error, raster-failed, provider-error,
pdf-needs-anthropic, empty-transcription. The writer sits at the choke
point the auto job and the corpus backfill share, and in the manual
index route; a success clears the reason. The provider path's specific
failure now survives the local fallback, since "no text" is less
actionable than "the PDF would not render". rasterizePdf names its
failure class so a build that cannot rasterize at all is distinguished
from a PDF that would not render. New guard in both index paths: an
empty provider transcription never becomes a successful empty index.

The detail sheet renders the reason in plain words (muted, six
locales). The DTO and OpenAPI carry the two fields; the disaster
recovery backup and restore carry them verbatim like their summary
siblings.

Refs #776
2026-08-13 12:40:50 +02:00
Marc-André Bombeck 74a386fcfd chore(deps): update July dependency bundle 2026-07-20 14:18:48 +02:00
Marc-André Bombeck 45da6c3c7d chore(release): v1.28.27 — runs on CPUs without AVX2
The bundled Skia x64 build (document thumbnails, scanned-PDF rasterization)
uses AVX2 instructions; on an x86-64 host without them — Celeron/Atom-class
NAS CPUs, pre-2013 Xeons, VMs masking CPU flags — the first canvas call is an
uncatchable SIGILL that kills the process and crash-loops the container (live
incident: a Synology DS1520+, exit 132 ~18 s after every boot, reported with
kernel traps and the correct root cause in hand).

The renderer is now gated on a one-time CPU-feature check before the module
ever loads: linux/x64 requires the avx2 cpuinfo flag; arm64 and non-Linux are
unaffected; unreadable cpuinfo fails open. Behind a closed gate, thumbnail
generation and PDF rasterization take their existing fail-soft paths (vault
shows type icons; PDFs read as text) and the boot-time thumbnail backfill
skips enqueueing. NATIVE_CANVAS=off|on overrides detection, documented in the
env example and whitelisted in compose.
2026-07-11 10:02:29 +02:00
Marc-André Bombeck ee76672798 feat(documents): preview thumbnails in the vault overview
Render a small JPEG preview for each stored document and show it at the
card's leading edge, falling back to the kind icon when none exists.

- Generate previews in the background on upload (new document-thumbnail
  pg-boss queue) plus a boot-time corpus backfill for existing documents;
  upload never blocks on or fails because of it. Images are downscaled via
  @napi-rs/canvas, a PDF's first page is rasterised then downscaled, every
  other type is skipped. Generation never throws.
- Store the preview encrypted at rest in a 1:1 DocumentThumbnail side table
  (migration 0236) so the list/detail queries that omit the original blob
  never drag the preview either. Same AES-256-GCM string codec + fail-closed
  decrypt as the content index; wired into the key-rotation walk.
- Serve it through an owner-scoped GET /api/documents/inbound/[id]/thumbnail
  (auth + module gate + userId-scoped query + fail-closed decrypt), always
  image/jpeg + nosniff + private, no-store; 404 on miss. The card renders an
  authed lazy <img>, onError falls back to the icon.
- Add hasThumbnail to the document DTO so the card only fetches a preview
  that exists.

Also drop the "search looks inside your documents" caption from the overview
(the index-all prompt stays) and the per-card searchable marker; both were
noise. Remove the now-unused i18n keys across all six locales.
2026-07-10 01:15:32 +02:00
Marc-André Bombeck a89e511359 fix(documents): run pdfjs + canvas un-bundled so scanned PDFs actually rasterize
The bundled (Turbopack) copy of pdfjs-dist failed its render path in the
standalone image with a bare Error, so a scanned PDF on a non-Anthropic
provider fell back to 'PDF scanning needs a Claude vision provider' — while the
identical un-bundled module rasterizes the same document fine (verified in the
container end to end on the real document). Externalize pdfjs-dist and
@napi-rs/canvas so the server loads the real modules, and hoist pdfjs-dist to
the top level in the image so the runtime import resolves. Also log the render
error message (not just its name) and drop the redundant 'not saved' note from
the document summary panel.
2026-07-09 16:22:00 +02:00
Marc-André Bombeck d2451021bc feat(documents): rasterize PDFs so image-only-wire providers can read them
Only Anthropic accepts a native PDF block; codex and other image-only-wire
providers dropped it. Add a bounded, best-effort server-side rasterizer that
renders a PDF's pages to JPEG images those providers read as ordinary image
input, so a scanned/image-only PDF is legible on the auto-read path.

- @napi-rs/canvas promoted to a direct dependency; its musl prebuilts + the
  pdfjs loader pinned into the standalone trace (next.config.ts)
- rasterize-pdf.ts: lazy pdfjs import (keeps DOMMatrix out of the eager chunk),
  first 10 pages, 1600px long edge, JPEG q80, never throws
- prepareVisionInput rasterizes a PDF for a non-Anthropic vision provider and
  keeps the native PDF block for Anthropic; the document capability reports
  PDFs readable via raster so the UI offers the read
2026-07-09 10:49:17 +02:00