[vpetersson-bot]

What's wrong today

Django's django.views.defaults.server_error loads 500.html off disk
and catches only TemplateDoesNotExist. On a device whose storage or
file-descriptor budget has already failed — which is the condition that
produced the original 500 — that load raises an OSError of its own.
Nothing catches it: it escapes handle_uncaught_exception, uvicorn drops
the connection, and the operator gets a browser error rather than a page
telling them their card needs attention.

Seen in the field on a Pi 5 that ran out of file descriptors: three
OperationalError: unable to open database file events — the real fault,
error-tracker issue ANTHIAS-63 — buried under twenty-four
OSError: [Errno 24] Too many open files: '.../app/templates/500.html'
raised while rendering the error page for them (ANTHIAS-62). The cascade
is worse than noise: the tracker ranks by event count, so the symptom
outranked its own cause 8:1.

What this does

Registers a handler500 that tries the branded template and, when
anything at all goes wrong reading or rendering it, serves a page that
has been in memory since import — no extends chain, no stylesheet link,
no static reference. 500.html extends _error.html, which pulls a
layout and a stylesheet, so the working path is several reads deep and
every one of them is covered.

Logged at warning rather than error: the exception that got us here has
already been reported, and re-reporting its fallout once per request is
what buried the cause. It still lands as a breadcrumb on the next event.

Only the 500 handler is replaced. The 4xx handlers read templates the
same way and would fail the same way, but they are reached by ordinary
requests on healthy devices; this one is reached because something is
already broken, which is what makes a disk read the wrong thing to
depend on here.

How I know it's right

A throwaway parametrized run against both handlers, with
loader.get_template raising EMFILE and a view raising
OperationalError:

django.views.defaults.server_error:    RAISED OSError(24, 'Too many open files') — client gets nothing
anthias_server.lib.error_pages.server_error:  500, 897 bytes

Committed coverage in tests/test_error_pages.py:

• the healthy path still serves the branded template;
• the fallback fires for EMFILE, EIO, ENOTCLEAN (117 — the errno
  behind the corrupted-ext4 crash-loop that
  ResilientWhiteNoiseMiddleware exists for) and TemplateSyntaxError,
  and the body it serves contains no <link and no /static/;
• the fallback logs exactly one record, at WARNING, with traceback;
• the root URLconf actually resolves handler500 to this view — a
  handler nobody routes to fixes nothing, and a typo there would only
  surface during a real outage;
• end-to-end through the exception middleware: view raises, template
  read fails, client still gets a 500 with a body.

uv run pytest -m "not integration" — 2376 passed, 3 skipped.
ruff check, ruff format --check and mypy . are clean (the two
tools/image_builder import-not-found errors are pre-existing, from the
optional docker-image-builder dependency group).

🤖 Generated with Claude Code

[Copilot]

Copilot encountered an error: Your billing is not configured or you have Copilot licenses from multiple standalone organizations or enterprises. To use premium requests, select a billing entity via the GitHub site, under Settings > Copilot > Features.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud