To-be 45 in progress; issue 269: the controller cannot write a cancelled set

Phase 0 is being built in mesh-controller and mesh-host. Granting the
controller its own buckets showed that its grant never covered the work
queues' cancelled sets, so every cancel times out.
This commit is contained in:
jochen
2026-10-06 03:04:36 +02:00
parent 213dcbafe9
commit a6c0f002a5
2 changed files with 51 additions and 2 deletions
@@ -1,7 +1,7 @@
---
layer: to-be
status: designed
code: []
status: in-progress
code: [mesh-controller, mesh-host, mesh-tools, mesh-catalog, mesh-lab]
updated: 2026-10-06
decisions:
- 02-DECISIONS/0227-the-core-holds-nine-rules-each-checked-and-is-built-to-them-in-six-phases.md
@@ -0,0 +1,49 @@
---
status: located
opened: 2026-10-06
located-in: [mesh-controller]
fixed-by:
amended-design:
---
# 269. The controller cannot write a work queue's cancelled set
## Symptom
Found while granting the controller the buckets of
[to-be 45](../../03-DESIGN/01-to-be/45-a-core-that-cannot-fail-silently.md) Phase 0, not seen on the
live mesh. A bus server was given exactly the user list the controller composes for itself, and the
controller, connected as itself, asked to mark a build cancelled:
```
asserting the cancelled set of the build seat → succeeded
putting an ask's id into it → context deadline exceeded (3 s)
```
The same put into the controller's own new buckets succeeded once their subjects were granted, and
timed out the same way before.
## Cause
[ADR 0219](../../02-DECISIONS/0219-the-build-queue-is-controlled-through-the-controller-and-the-build-seat.md)'s cancel writes the ask's
id into the seat's cancelled set before it deletes the ask, so a holder taking it in that moment ends
it. Writing a key into a bucket is a publish to the bucket's own subject. The controller's grant
reaches the bus's management interface, which creates the bucket, and does not reach the bucket's
subject, which writes it. The server refuses the publish; the client waits for an acknowledgement
that never comes.
So `cancel` and `clear` cannot complete. No test composes the controller's grant and writes the set
as the controller: the queue's tests write it with an unrestricted connection.
## Fix
The controller's publish grant names each cancelled set it writes, as it now names its own buckets,
and a test writes the set as the controller, against a server holding the controller's composed
grant. The installer's first user list says the same.
Pull requests: novox/mesh-controller#78 and novox/mesh-host#32 — open, not merged. Checked against a
server holding the controller's composed grant: the put that timed out succeeds.
## How it is checked
Live: `cancel` of a waiting ask answers that it was cancelled, and a holder taking that ask ends it.