dangernoodle-io / breadboard / 27630520387

100%

push

github

feat(bb_pub): add opt-in always-on ring mode (uniform telemetry queue) (#455)

- add `CONFIG_BB_PUB_BUFFER_ALWAYS` kconfig bool (default n, depends on
  `BB_PUB_BUFFER_ENABLE`); when y, all sink[0] publishes route through
  the ring (enqueue→drain each tick) for a uniform ordered queue;
  sinks[1..] keep existing direct fan-out unchanged
- ring is allocated eagerly at init when always-on (`bb_pub_buffer_init_eager`,
  called from `bb_pub_start`); on-failure mode keeps lazy alloc (zero
  standing cost until an outage)
- `captured_ms` injection: on-failure mode stamps any entry with ts>0
  unconditionally (every buffered entry is delayed by definition);
  always-on mode injects only when `(now_epoch_ms - entry_ts) > 1.5 ×
  interval_ms` so fresh entries drained same-tick carry no `captured_ms`
- `last_publish_ok` in always-on mode reflects whether the ring drained
  fully (empty after replay), not a direct-publish result
- add `bb_pub_test_set_buffer_always(bool)` runtime seam (under
  `BB_PUB_TESTING`) so both modes are exercisable in a single binary
  without rebuilding; `buf_reset()` in buffer tests now defaults to
  on-failure mode explicitly
- 6 new tests: always healthy (no captured_ms on fresh points), delivery
  order, outage→recovery with captured_ms on delayed entries only,
  ring-full eviction, last_publish_ok semantics, and regression (always=n)
- update kconfig table and bb_pub section in CLAUDE.md

1615 of 1615 branches covered (100.0%)

Branch coverage included in aggregate %.

2702 of 2702 relevant lines covered (100.0%)

1206.25 hits per line