fix(journeys): warm active-journey cache at boot + read-through findActive (EVO-1927)

[daniloleonecarneiro]

Summary

Regression: after an evo-flow restart, JourneysService.findActive() read from an empty/stale Redis active-journey index (no warm-up on boot). The JourneyTriggerProcessor then matched every event against ZERO journeys, so event-based triggers died silently on each deploy (Found 0 active journeys while Postgres had 56 active). Manual triggers were unaffected (they bypass the cache).

Two layers of defense:

1. Boot warm-up — JourneysService.warmActiveJourneysCache() loads active journeys from Postgres and populates the cache. The JourneyTriggerProcessor calls it in onModuleInit BEFORE subscribing to journey-triggers, so the first event matches the real journey set. Best-effort: a warm-up failure does not block the consumer.
2. Read-through in findActive() — on a cache miss (cache returns 0), fall through to the DB as source of truth and repopulate the cache so subsequent reads hit Redis again.
3. Observability — warn when the cache returns 0 while the DB has active journeys (the regression signature).

Test plan

• npx tsc -b clean.
• npx jest on journeys.service.spec.ts + journey-trigger-processor.service.spec.ts: 24/24 green.
  • findActive: cache-warm hit (no DB), empty-cache read-through + repopulate, observability warn, both-empty no-warn, best-effort on cache-write failure.
  • warmActiveJourneysCache: loads + populates, swallows per-journey write failures.
  • processor boot: warm-up runs in order before init/consume, consumer still starts when warm-up throws, no warm-up in non-journey-worker modes.

Notas

• Read-through and warm-up are independent; either alone fixes the regression, together they cover both the cold first event and any later cache eviction/TTL expiry.
• No schema/migration changes.

[sourcery-ai]

Sorry @daniloleonecarneiro, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery

[dpaes]

Approved — deep adversarial review (card + authoritative diff + real code): acceptance criteria met, no blocking findings. Merge intentionally held to land in the coordinated dependency-order pass. Branch is clean vs develop.