lsm / neokai / 28058184411

82%

push

github

feat: flush pending event deliveries when a worker node activates (#2166)

* feat: flush pending event deliveries when a worker node activates

External events persisted as `pending` in `space_external_event_deliveries`
(while a subscribed node was idle or inactive) are now injected into the
node-agent session on its next activation and marked `delivered`.

Extends `flushPendingNodeQueue` to also drain DB-persisted pending
deliveries via a new `flushPersistedPendingDeliveriesForTarget` helper.
When the in-memory pending queue for a target is empty (the typical case
for events that arrived while the node was idle), the runtime reads
`pending` rows from `ExternalEventStore.listPendingDeliveries`, filters
to the activated target, and re-dispatches each through the existing
`enqueueDeliverableExternalEvent` pipeline so delivery mode, rate-limit
digesting, retry handling, and state transitions stay consistent.

Closes #654.

* fix: always flush DB-persisted pending deliveries on activation

Address review feedback on PR #2166: the previous guard
(`if (!queued || queued.length === 0)`) could skip DB-persisted pending
deliveries when the in-memory queue was non-empty for the same target.

Now always runs the DB flush, passing the set of delivery keys just
drained from the in-memory queue so they're not double-dispatched.
Also filters to only deliveries with a non-null failureReason — rows
with null failureReason are still being processed by the original
dispatch path (in-flight or digest-pending) and must not be
re-dispatched.

Adds a test verifying DB-persisted items are flushed even when the
in-memory queue is non-empty.

* fix: merge/sort both pending sources and preserve TTL on retry

Address P2 review feedback on PR #2166:

1. Chronological ordering: items from in-memory queue and DB-persisted
   pending deliveries are now collected into a single array and sorted
   by `createdAt` before dispatch, so a newer in-memory event is never
   injected before a... (continued)

11214 of 16228 branches covered (69.1%)

Branch coverage included in aggregate %.

86 of 86 new or added lines in 1 file covered. (100.0%)

94715 of 113024 relevant lines covered (83.8%)

299.95 hits per line