Important Fixes, New mechanics, and many more

supabase/migrations/0005_failed_round_retry.sql

@@ -0,0 +1,83 @@
+-- Owners can safely retry a round after all worker attempts have failed.
+-- The queue has one durable outbox row per round. A retry records the failed
+-- state in the audit log, then safely re-queues that same row.
+create or replace function public.stage_two_retry_failed_round(
+  p_round_id uuid,
+  p_owner_id uuid
+) returns uuid language plpgsql security definer set search_path = '' as $$
+declare
+  v_round public.rounds%rowtype;
+  v_job public.ai_jobs%rowtype;
+begin
+  select * into v_round
+  from public.rounds
+  where id = p_round_id
+  for update;
+
+  if not found then raise exception 'round not found'; end if;
+  if v_round.status <> 'failed'::public.round_status then
+    raise exception 'round is not failed';
+  end if;
+  if not exists (
+    select 1 from public.campaigns
+    where id = v_round.campaign_id
+      and owner_id = p_owner_id
+      and status = 'active'
+  ) then
+    raise exception 'only the campaign owner can retry a failed round';
+  end if;
+  if exists (
+    select 1 from public.rounds
+    where campaign_id = v_round.campaign_id
+      and id <> p_round_id
+      and status in ('open'::public.round_status, 'queued'::public.round_status, 'resolving'::public.round_status)
+  ) then
+    raise exception 'campaign already has an active round';
+  end if;
+
+  select * into v_job
+  from public.ai_jobs
+  where job_type = 'resolve-round' and entity_id = p_round_id
+  for update;
+  if not found or v_job.status <> 'failed' then
+    raise exception 'failed round job not found';
+  end if;
+
+  insert into public.audit_entries(
+    campaign_id, actor_id, action, entity_type, entity_id, before_state, after_state
+  ) values (
+    v_round.campaign_id,
+    p_owner_id,
+    'retry_failed_round',
+    'round',
+    p_round_id,
+    jsonb_build_object('roundError', v_round.error, 'jobError', v_job.error, 'attempts', v_job.attempts),
+    jsonb_build_object('status', 'queued')
+  );
+
+  update public.ai_jobs
+  set status = 'queued',
+      attempts = 0,
+      claimed_by = null,
+      lease_expires_at = null,
+      available_at = now(),
+      error = null,
+      updated_at = now()
+  where id = v_job.id;
+
+  update public.rounds
+  set status = 'queued'::public.round_status,
+      error = null,
+      queued_at = now(),
+      resolved_at = null,
+      forced_by = p_owner_id
+  where id = p_round_id;
+
+  return v_job.id;
+end;
+$$;
+
+revoke all on function public.stage_two_retry_failed_round(uuid, uuid) from public, anon, authenticated;
+grant execute on function public.stage_two_retry_failed_round(uuid, uuid) to service_role;
+
+notify pgrst, 'reload schema';