view_reach asked for mesh-delivery-view.reach, a name the mesh never issued, so it failed on the machines;
the same line was not gofmt'd, which failed the try of main. Quick fix: the name. Thorough for this module: the
manifest test now refuses any stdio.State literal in cmd/ that is neither an own state nor listed in reads.
mesh/check checking f007ca42279a on 4b36cc29a6ae | novox/mesh-delivery#81 in group fix/1496-start-health
Review of 50c4674: the bar on new wants naming F while a node fails it alone
also caught a node that runs F with no live want, the issue-1945 recovery path:
it went on neither list and its node-engine gave the module up. The bar now
holds only a node that neither wants nor runs F.
ADR 0557 version 9: a failure that is one node's holds nothing and sets nothing
back, but the version reaches no node that does not want it yet until it is
over. A node whose want does not name F gets no new want naming it (rule 2, in
words naming the failing nodes); wants already written stay.
Review round 2 of 16e4e61: gating the first-pass clear on a failed have left a
condition an earlier process raised standing for good once its node recovered
while no assigner watched. The first pass now clears it for every module and
every node the presence bucket knows, a quiet clear at the router for one that
is not open.
Review of 2ee7589: a node that sleeps after running F healthy kept its stale
have as a witness, so F stayed unheld while it failed on the online nodes; and a
failed-alone condition was never cleared once its node left the module.
A witness must now be online by the watchman's presence. F failing on a second
node is no longer one node's failure: the hold and the set-back stand (ADR 0557
version 6, §2). A remembered failed-alone key whose node is no longer one of the
module's nodes is cleared once; the first pass after a start clears it for every
node the presence bucket knows, for a module with a failed have.
On ace, 2026-10-26 21:19, a newly assigned mesh-build-node failed its first
health check at 8a98386f (later 31be2f24) while shanks, g14 and novox had run it
healthy for an hour. Being first by name, ace was the canary, and the assigner
held the version and moved the three healthy nodes back.
A failed health check of the newest fingerprint now holds it only when no other
node's have names it healthy (ADR 0557 §2 keeps ADR 0429 §6's hold and set-back
for that case). Otherwise the failure is that node's (ADR 0557 §1): nothing is
held or set back, the node keeps its want, and the assigner raises the node's
condition once (failed-alone.<node>.<module>), naming the node, the module, the
fingerprint and the have's words, and clears it once when it ends; the first
pass after a start clears what a previous process may have raised.
mesh/check checking c47afd8f4b88 on 3dd94f714d6e | novox/mesh-delivery#78 in group fix/capacity-follows-cpu-share
A build module's capacity.<node> now carries its node's build window (the operator, 2026-10-26: g14
builds only from 18:00 to 06:00); the seat shows it, outside it says it takes nothing new, and a
value the module cannot read shows as any hour with the value and why. The checker's and the view's
SDK pin moves to mesh-sdk d00dac1.
Only tries a build node holds count toward the 4 at once; tries waiting for a node are bounded
apart at 8, and the refusal names both counts (issue 1890: at 19:53 four tries queued behind 44
checks for up to 47 minutes refused every further try).
The checker wrote a try's asks only at its end, so a 31-minute try could not be told queued from
running (issue 1890). The try's record now names each ask as it is put, then the node that took
it, its state and the phase its step says, written on the step key's own changes. build_usage
names a running try's asks and takes a build module's capacity.<node> (configured bounded by its
node's build CPU, issue 1806's follow-up) when it is current with mesh-config.
mesh/check failed: ace: novox/mesh-catalog#505 does not merge into the trunk: files in conflict: .gitignore | 2e0013ec2500 on df8b9ac3b1b5 | unit 1245…
mesh/check failed: ace: novox/mesh-catalog#505 does not merge into the trunk: files in conflict: .gitignore | 2493aa65299e on df8b9ac3b1b5 | unit 1245…
One commit for the branch feat/1126-nodes-compose of mesh-delivery: its 16 commits squashed so its history does not carry the 17.8 MB compiled mesh-delivery-assigner a commit added by mistake and a later one removed (issue 1577). The tree equals the former head ff6c038d. The operator allowed this rewrite of the feature branch on 2026-10-26.
A delivery restarted the checker and the builder before the grants part created their new state hearing;
OptionalBucket answered nil and the hearing was never kept. They, and the builder's reasks and queues and the
indexer's takeins, now open their own states through mesh-sdk process.AwaitBucket (mesh-sdk
fix/1790-state-appears d515c71), which binds on the bus's STREAM.CREATED advisory and writes the latest kept
value. mesh-grants takes the same SDK, whose composer grants that advisory.
A capacity this tool cannot read is unknown to it, not 0: the seat says eligible_known false and the node not known.
A key mesh-config holds without a readable value is what the puller reads as 0, and was dropped silently.
Since mesh-sdk #90 a build module ends a left-running key as failed at the moment it notices; counted as finished,
such a step put hours into the medians and the time at capacity. The view takes mesh-sdk's step bound and restart
reason from the SDK it now pins.
The operator's question (how much are the build nodes used, is ace eligible) took six calls over four tools.
build_usage reads it from the bus's states: seats held (have), capacity (mesh-config resolved), pauses, the
build modules' steps by kind with left-running keys, the last hour and day with total and median time and time
at capacity, the running tries, and the backlog the view program reads from the seats' streams. A try now names
the asks it put, and the checker announces its tries readable, so the view tells a try's asks from a check's.
mesh-grants moves to the same mesh-sdk (3203358c), so its shadow user list grants the checker and the builder the
DEAD_LETTERS reads the controller's list will, and the take-over is not blocked by a difference. The checker and the
builder hear DEAD_LETTERS through buildrun.DeadLetterHearing: in the background, again at every reconnect and every ask
waited for while deaf, and keep whether they hear it in their new state hearing, which checker_status and builder_stuck
say as dead_letters. checker_try says that a try ended by the not-taken bound leaves its ask in the stream.
Not done as asked: a module health tool. The catalogue refuses a tool check on a module with no runtime of its own and no
other check beside it (ADR 0227 rule 8, ADR 0366 §1), so the deaf state is said in the status tools and the log.
jschoubben
mesh-tools