Symptom

For every bookmark you save, Linkwarden's worker creates four offline snapshots in the background and stores their paths on the link record: image (PNG screenshot), pdf, readable (extracted plain-text/JSON), and monolith (full self-contained HTML archive — the actual offline copy of the page, with images/CSS/JS inlined as data: URIs). When a snapshot fails the corresponding field is set to the literal string "unavailable".

On app.linkwarden.cloudronapp@1.21.1 the first three always succeed, but every link ends up with monolith: "unavailable":

GET /api/v1/links/14
{
  "id": 14,
  "url": "https://example.com/",
  "image":    "archives/1/14.jpeg",
  "pdf":      "archives/1/14.pdf",
  "readable": "archives/1/14_readability.json",
  "monolith": "unavailable",
  "lastPreserved": "2026-04-30T12:41:57.614Z"
}

In effect, the most useful preservation format — the offline HTML archive — is missing from every saved bookmark, while the UI just shows "unavailable" without surfacing why.

Worker logs show the underlying error:

Error: spawn monolith EACCES

Source: apps/worker/lib/preservationScheme/handleMonolith.ts calls spawn("monolith", …), which relies on the binary being resolvable through PATH for the worker process.

Root cause

The monolith binary is installed via cargo install --locked monolith in the Dockerfile. Cargo defaults to $CARGO_HOME/bin, which during the build is /root/.cargo/bin. The base image keeps /root at mode 700, and the runtime worker runs as cloudron (UID 1000) — so even though the binary itself is -rwxr-xr-x, cloudron can't traverse /root to reach it.

Reproduce inside a 1.21.1 container:

$ ls -ld /root
drwx------ 1 root root 4096 Apr 23 02:20 /root

$ ls -la /root/.cargo/bin/monolith
-rwxr-xr-x 1 root root 12384776 Apr 23 02:22 /root/.cargo/bin/monolith

$ runuser -u cloudron -- /root/.cargo/bin/monolith --version
runuser: failed to execute /root/.cargo/bin/monolith: Permission denied

/root is on the read-only image layer, so chmod o+x /root at runtime is also rejected. This is purely a packaging issue — the binary is fine, just unreachable.

Fix

One-line change in the Dockerfile: pass --root /usr/local to cargo install so the binary lands in /usr/local/bin/monolith, which is on the default PATH and traversable for everyone.

-# install rust and monolith (/root/.cargo/bin/monolith)
+# install rust and monolith
+# install monolith into /usr/local/bin so it is reachable for the runtime
+# `cloudron` user (UID 1000); /root is mode 700 in the base image and would
+# otherwise hide /root/.cargo/bin/monolith → spawn EACCES in the worker.
 RUN curl https://sh.rustup.rs -sSf | bash -s -- -y
 ENV PATH="/root/.cargo/bin:${PATH}"
-RUN cargo install --locked monolith
+RUN cargo install --locked --root /usr/local monolith

Full patch attached: linkwarden-monolith-eacces.patch (3 files, +9/−3 — Dockerfile + CHANGELOG entry + CloudronManifest version bump 1.21.1 → 1.21.2).

I'd happily open a merge request directly, but I don't have a git.cloudron.io account — feel free to apply the patch with git am linkwarden-monolith-eacces.patch (or just take the diff).

Verification

Verified the fix on a running 1.21.1 container by approximating the post-patch state at runtime:

1. cp /root/.cargo/bin/monolith /app/data/bin/monolith (writable persistent volume)
2. Prepend /app/data/bin to the worker's PATH via a tiny NODE_OPTIONS=--require … shim (necessary because dotenv-cli won't overwrite an already-set PATH)
3. Restart the app

After that, a freshly created link preserves correctly:

{
  "id": 15,
  "url": "https://en.wikipedia.org/wiki/Monolith_(file_archive)",
  "monolith": "archives/1/15.html",
  "image":    "archives/1/15.jpeg",
  "pdf":      "archives/1/15.pdf",
  "readable": "archives/1/15_readability.json"
}

The resulting HTML is 600 KB and embeds data URIs as expected from monolith. Pre-existing "unavailable" links can be brought back with the Re-preserve all broken admin action.

Possibly related

Topic #13640 ("Missing HTML saved version and cookie popup in screenshots") describes the same observable symptom and was tentatively diagnosed as a resource issue. Based on the reproduction above I believe this EACCES is the actual underlying cause for the missing HTML snapshots.