Reader side of the OpenSky shared-cache v1 contract (mirrored in GODSIGH
SPEC2.md §4). Path ~/.cache/godverse/opensky-states.json (env OPENSKY_CACHE_FILE),
content the raw global /states/all body, freshness = mtime within 120 s.
Each poll cycle now checks the cache first: on a fresh hit it reads the global
body, filters the states to our bbox client-side (indices 5=lon, 6=lat), and
skips the HTTP call entirely (logs "cache hit"). On 429 or a network failure it
falls back to a stale cache rather than nothing. This worker only ever polls a
bbox, so it never WRITES the cache (never partial data into a global-contract
file) — it is a pure reader, and degrades to its old ≥450 s bbox fetch when no
writer is around (i.e. prod). Result: godstrument + GODSIGH running together
spend ~one poller's quota instead of colliding and 429-ing each other out.
Verified: unit tests for read_cache freshness/stale/missing + bbox filtering,
and an integration run where a fresh cache is served (count filtered to bbox,
zero HTTP).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
6d408a8bdb