Add recorder: raw CAN and GNSS sessions with failsafes and a storage bound

Reads SocketCAN and gpsd directly. Sessions start on the first frame or on
'ctl start', end after 15 s of bus idle or on 'ctl stop'; blips under 3 s are
discarded. candump -L chunks fdatasync'd every second and zstd'd once closed;
oldest sessions deleted to stay under 50 GB / above 20 GB free. Kernel and
queue drops, link changes and clock steps are recorded as events.

Bench-tested on vcan with the first car capture (byte-identical) and installed
on the car Orin as yukart-recorder.service.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DDwfd6Ki8UfG4hK6W5YpJz
This commit is contained in:
2026-10-09 20:15:13 +09:00
commit 1f678fe87f
19 changed files with 1784 additions and 0 deletions
+33
View File
@@ -0,0 +1,33 @@
---
type: guidelines
---
# Core services guidelines
## 1. Listen-only until `obd-poller` exists
- Nothing here opens a CAN socket for writing. The recorder only reads, and the
host brings `can_pcan` up listen-only.
- `obd-poller` will be the single writer. Until it lands, any transmit is a bug.
## 2. Capture never waits
- A frame missed is gone. Producers hand frames to the writer without blocking;
what doesn't fit is **counted**, recorded as an event, and never silently lost.
- Capture reads SocketCAN and gpsd directly. It does not depend on the broker, a
display, or any other service being up.
## 3. Every failure leaves complete files
- Append, flush and `fdatasync` about once a second. Power loss costs at most the
last second.
- Replace files by write-temp, fsync, rename. Never rewrite in place.
- Anything found half-done at startup is repaired or marked, not deleted.
## 4. Raw stays raw
- Recordings are `candump -L` text plus the gpsd JSON as received, so `can-utils`,
`canplayer` and anything already written against the first captures keep working.
- Decoding happens downstream of the recording, never instead of it.
## 5. Time is evidence, not truth
- The car unit's clock can boot stale and be stepped mid-drive. Record what is
needed to correct it later (sync state, monotonic time, observed steps, GNSS
time beside local receive time) rather than trusting it.
## 6. To be added