giles
8ba3584556
Some checks failed
Test, Build, and Deploy / test-build-deploy (push) Failing after 45s
New next/kernel/delivery_state.erl folds delivery events into a
per-peer worker-shaped snapshot so the outbound queue survives
kernel restart.
Event proplist shapes:
[{type, enqueued}, {peer, _}, {activity, _}]
[{type, delivered}, {peer, _}, {cid, _}]
[{type, failed}, {peer, _}, {cid, _}, {now, _}]
[{type, dead_lettered}, {peer, _}, {cid, _}]
Projection state shape:
[{PeerId, [{peer, _}, {pending, _}, {attempts, _},
{next_retry, _}, {dead_letter, _}]}, ...]
Mirrors delivery_worker:new/1 (minus :dispatch_fn — that's the
live worker's concern) so a fresh gen_server can be hydrated
from the projection on restart.
Public API:
new/0
fold/2, fold_fn/0
peer_state/2, peers/1
pending/2, attempts/2, next_retry/2, dead_letter/2
The failed branch calls delivery_worker:backoff_for/1 directly,
so the projection and the live worker compute identical retry
slots and dead-letter thresholds. 6th failure -> dead-letter,
matching the worker.
14/14 in next/tests/delivery_state.sh covering:
- new/0 -> []
- enqueued appends to pending (FIFO)
- two peers maintain independent queues
- delivered clears matching pending entry
- failed bumps :attempts and sets :next_retry
- 6th failed -> dead-lettered (activity out of pending)
- explicit dead_lettered event moves activity to dead_letter
- peers/1 lists touched peers
- peer_state {ok, _} | not_found
- fold_fn/0 is fun/2 for projection:start_link
- unknown event type passes through
- delivered after failed clears retry state
delivery_worker.sh 17/17 unchanged, delivery_retry.sh 11/11
unchanged. Conformance preserved at 761/761.
The restart hydration helper (delivery_worker:state_from_proj/2
or similar) lands once 8b-timer can wire the live retry loop
(Blockers #3 — erlang:send_after substrate gap still open).