Repository navigation
IEventSubscriptionAgentFamily.FindAgentUriAsync resolves only live agents — can't restart/rebuild a non-running projection #3124
Description
Activity
- added a commit that references this issue
on Jun 17, 2026 The core of this is delivered and merged in #3126.
IEventSubscriptionAgentFamily.FindAgentUriAsyncresolves from the registered subscription set (usage.Subscriptions) rather than only live agents, so a paused/stopped/crashed/not-yet-started projection still resolves — which is what makes restart/rebuild of a non-running projection possible. #3126 also:- corrected the interface + implementation XML docs (they previously said "live agent" — the exact contradiction reported here), and
- added regression tests proving a registered shard resolves with no running agent, and that an unregistered shard returns null.
Covered: the non-tenant case (the repro here,
FindAgentUriAsync("Trip:All", null)), the pre-start window, and single-database per-tenant partitioning (tenant in the shardRelativeUrl).One scoped gap remains, split out as a focused follow-up: #3128 — under database-per-tenant (sharded databases) a non-null
tenantIdcan't be resolved, because that needs atenantId → databaseIdmapping rather than matching the shardRelativeUrl. That isn't part of this issue's repro (which uses a null tenant), so it's tracked separately.Recommend closing this in favor of #3128 for the remaining database-per-tenant work.
- added a commit that references this issue
on Jun 17, 2026 - added a commit that references this issue
on Jun 18, 2026
Summary
IEventSubscriptionAgentFamily.FindAgentUriAsync(shardIdentity, tenantId, ct)returns null when there is no live (running) agent for the shard. That makes it impossible to resolve — and therefore pause/restart/rebuild — a projection whose agent is registered but not currently running (paused, stopped, crashed, or in the startup window before agents start).Why it matters
CritterWatch's admin commands (
PauseProjectionHandler,RestartProjectionHandler,RebuildProjectionHandler) all resolve the real agent URI viaFindAgentUriAsync(they must not hand-compose the URI). When an operator tries to recover a stopped/orphaned async projection — exactly the situation where Restart/Rebuild is most needed —FindAgentUriAsyncreturns null and the command no-ops (Restart/Rebuild return a "No live agent" result; Pause silently returns).This also appears to be the root of a flaky pause→restart round-trip: a paused agent may no longer count as "live", so the subsequent restart can't find it to clear the pause restriction.
Reported downstream as CritterWatch ProductSupport #18/#19/#20 (projection Pause/Restart/Rebuild "do nothing" on an Orphaned/Continuous projection) and tracked alongside CritterWatch's own flaky
can_pause_and_restart_a_projectionintegration test.Ask
Provide a way to resolve the agent URI for a registered subscription/projection shard regardless of its current run state — e.g.
FindAgentUriAsyncconsulting the registered subscription set (not just live agents), or a siblingFindRegisteredAgentUriAsync. That would let an operator restart/rebuild a projection that isn't currently running, which is the whole point of those operations.Repro sketch
FindAgentUriAsync(shardIdentity, null, ct)→ returns null though the projection is registered.