nats removes a push consumer; gitea, mesh-conditions, mesh-issues and messenger drop what named the removed controller (issue 1126) #607
Assignees
agent-ace
agent-g14
agent-novox
agent-shanks
jschoubben
mesh-tools
miel.truyen
nox-agent
Clear assignees
agent-ace
agent-g14
agent-novox
agent-shanks
jschoubben
mesh-tools
miel.truyen
nox-agent
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: novox/mesh-catalog#607
Reference in New Issue
Block a user
Group fix/consumer-remove-any with mesh-delivery. Issue 1126 (notes 146-150); the operator asked 2026-11-04 to remove the old controller's consumers and every dead reference to the removed controller.
bff850f7nats: nats_consumer_remove reads a push consumer as well as a pull one, before and after removing it (live failure: 'nats: consumer is not a pull consumer' for EVENTS/controller, EVENTS/controller-merges, CONTROL/controller).052fb473,4f67143a,080aed68gitea: the verdict, plan and check-duration code bound to no event since ADR 0571 clause 13, and what only tests called; the merge refusal no longer promises how long a check takes.bf4ce7efmesh-conditions: the hold's controller special case.b0b7dfc5mesh-issues: mesh-controller is no registered or core owner; old issues keep their text.1572ab9bmessenger: the old controller's walk decoder and plan-moved names.Left on purpose: the lab's builds from mesh-controller (issue 1816 note 3: deleting them would make its suites skip and read as passed).
Quick fix: none without a hand act. Thorough, chosen: the repair tool reads any consumer kind, and the dead code goes instead of misleading the next reader.
Tests: nats removal table rows (pull and push); gitea's whole suite 216/216 against the mesh-sdk package; go vet and go test per Go module.
On delivery (novox): nats moves its tools bundle only (the server container is not restarted; the bus is backed up first); gitea, mesh-conditions, mesh-issues, messenger move.
Reviewed-by: review-consumer-remove@shanks at
080aed6855https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
js.Consumer refuses a push consumer ("consumer is not a pull consumer"), so the controller's leftovers on EVENTS and CONTROL could not be read and were not removed. The tool now reads a consumer of either kind (CONSUMER.INFO only, which is all its grant gives), before the delete and for the confirmation after it. Quick fix: this change, in the tool, one function, about an hour with its test rows (a pull and a push consumer). Thorough fix: the same tool also needs its grant to cover the KV_* streams' consumers (mesh-controller-take-in, mesh-controller-config-follow), and that grant is composed by mesh-sdk's BusRemovalGrants from a closed set (ControllerLeftoverConsumersOn), not by the manifest, so it is not in this commit: it needs a change in mesh-sdk and a group delivery. Chosen here: the quick fix for the read, since it is one place and blocks the operator's removal now; the grant is reported to the lead. Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCnDelivery
novox/mesh-catalog@080aed68556a— merged since 2026-11-04T10:36:43ZSource — source.main moved to novox/mesh-catalog@7f626147be. Whether it reached the machines is
mesh-delivery-view's, derived fromsource.main, the wants and the haves.Group
fix/consumer-remove-any: novox/mesh-catalog@080aed6855, novox/mesh-delivery@fc300216ecTransitions
The commit's note under
refs/notes/mesh-plankeeps every transition of this change up to its merge:git log --notes=mesh-plan.