[Jochemp88]

What

Hardening for 1.2.1: MTC's certificate is verified again, every Tesla/MTC call has a timeout, a hanging run is stopped, and the Docker image gets a health check.

MTC certificate is verified again

MTC's server mtc.outsystemsenterprise.com presents MTC's own certificate for *.mtc.nl (also valid for mtc.nl), which fails the normal host name check. That is why verification was switched off completely (verify=False) and the warning was hidden, so the MTC password and IBAN went over an unverified connection.

Now (MTCAdapter in MTC.py):

• the certificate chain is verified against the trusted CAs (certifi), and the certificate must be valid for mtc.nl (urllib3 assert_hostname);
• if MTC ever installs a certificate for its real host name, the first request of a session, which carries no credentials, falls back to the normal check. The certificate is always verified; when neither check passes, the run stops.

Checked live with the credential-free calls tesla_mtc makes before logging in (module version, JavaScript files, app store URLs): all succeed with verification on. Login and submitting were not run live (no test credentials); they use the same session and adapter.

Timeouts

• HTTP_TIMEOUT = (10, 60) (connect, read) on every Tesla call and, through the adapter, on every MTC call.
• A claim is never sent twice: with timeouts, the existing retry loop would resend a claim when MTC does not answer after it was sent. Because claims cannot be withdrawn, a read timeout on the claim request now ends with "Claim submission failed … may have been submitted. Check MTC before running again." The next run's duplicate check finds the claim if it arrived. A connection that fails before sending is still retried.

Web interface and Docker

• RUNNER_RUN_TIMEOUT (default 900 s, 0 = off): a run that has not finished is stopped and shown as failed, so the page and Home Assistant never stay on Running. On POSIX the whole process group is ended, so a wrapper script in RUNNER_CMD (for example one that pipes into tee) cannot keep the run open.
• HEALTHCHECK with web/healthcheck.py (standard library, uses RUNNER_PORT and RUNNER_TOKEN): healthy/unhealthy in Docker and Synology Container Manager. The CI smoke test checks it.
• OCI labels (title, description, source, license).

Logs and dependencies

• The CSRF token and session cookie values are no longer logged (cookie names only, at DEBUG).
• The startup warning about disabled certificate verification is gone.
• tqdm pinned to 4.70.1 like the other dependencies.

Tests

19 new tests:

• certificate checks, including the actual urllib3 pool settings, the fallback and the stop when both checks fail;
• default and explicit timeouts, and the Tesla calls;
• no resend after a read timeout, while a connection failure is still retried;
• run time limit, including behind a wrapper script on Linux;
• health check, with and without a token;
• a guard that verify=False and disable_warnings never come back.

To check that the tests catch regressions, 11 deliberate bugs were introduced one at a time (for example "mtc.nl check removed", "claim resent after no answer", "hanging run not stopped"); every one made the tests fail.

Checked locally

• ruff check . and pytest: 75 passed. One test (process group) is POSIX-only and runs in CI.
• Not run locally: Docker job and Python 3.12. This pull request's own run checks them.

Docs: RUNNER_RUN_TIMEOUT and the health check in docs/web-runner.md; version 1.2.1 with a changelog entry (still Unreleased, dated at release).

🤖 Generated with Claude Code