change: retire the last log scraping #91

Merged

mysticalsoap merged 2 commits from change/retire-log-parsing into trunk

2026-08-20 14:14:02 -04:00

mysticalsoap commented

2026-08-20 13:03:46 -04:00

Owner

Problem

Follow-up to #90: three free-form text scrapes remained after the management-interface migration. parse_tun_device still read the log for raw custom configs (and for tap configs, whose name write_config saw and threw away); the stdout loop waited on a SIGTERM[hard,] sentinel that a real exit never prints; and monitor.py split ip route output positionally — the same failure class that broke the 2.7 log parse.

Fix

The device name is always known before launch: write_config returns a tap config's own name; a raw config's dev directive is read pre-launch, and a bare/missing one gets a name assigned via --dev on the command line (later options beat --config). parse_tun_device is deleted.
The stdout read loop ends on EOF alone; the dead sentinel and post-loop line emit are gone. Stdout is now only the GUI log and the double-hop iptables retry.
monitor.py reads the default route with ip -j + JSON, matching the bypass pin in tunnel.py.

Verification

130 passed. New TestConfigTunDevice covers concrete/bare/missing dev directives, per-role assignment, and the bare-tap case (a tap config must not get a tun_* name — the prefix carries the device type).
The EOF-only loop is backed by the captured DCO session fixture: a real disconnect prints only SIGTERM[soft,exit-with-notification].
ip -j route show default field names (gateway, dev) verified against the live host.
The captured-session device tests from #90 retire along with the parser they tested; the fixtures stay (management transcripts are active test inputs, the stdout logs back the EOF rationale and document real 2.7.6 formats).

**Problem** Follow-up to #90: three free-form text scrapes remained after the management-interface migration. `parse_tun_device` still read the log for raw custom configs (and for tap configs, whose name `write_config` saw and threw away); the stdout loop waited on a `SIGTERM[hard,]` sentinel that a real exit never prints; and `monitor.py` split `ip route` output positionally — the same failure class that broke the 2.7 log parse. **Fix** - The device name is always known before launch: `write_config` returns a tap config's own name; a raw config's `dev` directive is read pre-launch, and a bare/missing one gets a name assigned via `--dev` on the command line (later options beat `--config`). `parse_tun_device` is deleted. - The stdout read loop ends on EOF alone; the dead sentinel and post-loop line emit are gone. Stdout is now only the GUI log and the double-hop iptables retry. - `monitor.py` reads the default route with `ip -j` + JSON, matching the bypass pin in tunnel.py. **Verification** - 130 passed. New `TestConfigTunDevice` covers concrete/bare/missing `dev` directives, per-role assignment, and the bare-tap case (a tap config must not get a `tun_*` name — the prefix carries the device type). - The EOF-only loop is backed by the captured DCO session fixture: a real disconnect prints only `SIGTERM[soft,exit-with-notification]`. - `ip -j route show default` field names (`gateway`, `dev`) verified against the live host. - The captured-session device tests from #90 retire along with the parser they tested; the fixtures stay (management transcripts are active test inputs, the stdout logs back the EOF rationale and document real 2.7.6 formats).

mysticalsoap added 2 commits

2026-08-20 13:03:46 -04:00

change: read the default route as json in the monitor af70617f22

The positional split of 'ip route show default' output is the same
failure class that broke the 2.7 log parse -- any field the kernel adds
or reorders shifts the indices silently. ip -j mirrors how tunnel.py
already reads routes for the bypass pin.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

change: resolve the tunnel device from the config, not the log

ci / test (pull_request) Successful in 37s

Details

ci / test (push) Successful in 58s

Details

2b61ef9bc0

The last log-scraping unit is gone: write_config now reports a tap
config's own device name instead of discarding it, and a raw config
aqomui did not write either brings a concrete dev directive or gets a
name assigned on the command line (later options beat --config). With
nothing reading state from stdout anymore, the read loop ends on EOF
alone -- its SIGTERM[hard,] sentinel never matches a real exit, which
only prints SIGTERM[soft,...] (see the captured session fixture).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

mysticalsoap referenced this pull request

2026-08-20 13:04:30 -04:00

decision: stay on OpenVPN 2.x, no openvpn3-linux migration #92

mysticalsoap merged commit 2b61ef9bc0 into trunk

2026-08-20 14:14:02 -04:00

mysticalsoap deleted branch change/retire-log-parsing

2026-08-20 14:14:02 -04:00