mirror of
https://github.com/koala73/worldmonitor.git
synced 2026-04-25 17:14:57 +02:00
b8615dd106
* fix(intelligence): read regional-snapshot latestKey as raw string
Regional Intelligence panel rendered "No snapshot available yet" for every
region despite the 6h cron writing per-region snapshots successfully. Root
cause: writer/reader encoding mismatch.
Writer (scripts/regional-snapshot/persist-snapshot.mjs:60) stores the
snapshot_id pointer via `['SET', latestKey, snapshotId]` — a BARE string,
not JSON.stringify'd. The seeder's own reader (line 97) reads it as-is
and works.
Vercel RPC handler used `getCachedJson(latestKey, true)`, which internally
does `JSON.parse(data.result)`. `JSON.parse('mena-20260421T000000-steady')`
throws; the try/catch silently returns null; handler returns {}; panel
renders empty.
Fix: new `getCachedRawString()` helper in server/_shared/redis.ts that
reads a Redis key as-is with no JSON.parse. Handler uses it for the
latestKey read (while still using getCachedJson for the snapshot-by-id
payload, which IS JSON.stringify'd). No writer or backfill change needed.
Regression guard: new structural test asserts the handler reads latestKey
specifically via getCachedRawString so a future refactor can't silently
revert to getCachedJson and re-break every region.
Health.js monitors the summary key (intelligence:regional-snapshots:
summary:v1), which stays green because summary writes succeed. Per-region
health probes would be worth adding as a follow-up.
* fix(redis): detect AbortSignal.timeout() as TimeoutError too
Greptile P2 on PR #3302: AbortSignal.timeout() throws a DOMException with
name='TimeoutError' on V8 runtimes (Vercel Edge included). The existing
isTimeout check only matched name==='AbortError' — what you'd get from a
manual controller.abort() — so the [REDIS-TIMEOUT] structured log never
fired. Every redis fetch timeout silently fell through to the generic
console.warn branch, eroding the Sentry drain we added specifically to
catch these per docs/plans/chokepoint-rpc-payload-split.md.
Fix both getCachedJson (pre-existing) and the new getCachedRawString by
matching TimeoutError OR AbortError — covers both the current
AbortSignal.timeout() path and any future switch to manual AbortController.
Pre-existing copy in getCachedJson fixed in the same edit since it's the
same file and the same observability hole.
* test(redis): update isTimeout regex to match new TimeoutError|AbortError check
Pre-push hook caught the brittle static-analysis test in
tests/get-chokepoint-history.test.mjs:83 that asserted the exact old
single-name pattern. Update the regex (and description) to cover both
TimeoutError and AbortError, matching the observability fix in the
previous commit.