mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (4 of 4 compose)
At 03:30 the store's collector held the registry still while the
node-engine's reconcile on that machine was fetching bundle blobs from
it: every archive failed 'connection refused' and the machine was held
until the next pass. Other machines can meet the same window.
A scheduled step now opens its window only once no apply is in flight
here (the apply lock is taken just to write the record, so a push still
never queues behind the window). An apply whose fetch the store does
not answer waits for a window open on its own machine to close, and
elsewhere retries with backoff within one bounded budget per apply,
well past the window's length; an answer such as 404 still fails at
once.
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (4 of 4 compose)
The witness moved a running controller's build into a 0700 directory and deleted it
on proof, while the old process could still be serving. Its directories are now
0711, and a build without a reader is retired and swept once /proc shows nothing
runs from it.
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (4 of 4 compose)
A container that crash-looped after its compose applied passed every check the
gate had: nothing looked at what a module runs. The node-engine now judges every
long-running resource on every look — one read of the runtime, one per service
manager — keeps the restarts it counts across recreates and its own restarts,
and says the state in every report and as an event on change, again every minute
while not healthy. It reads only; nothing is restarted for being unhealthy.
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (4 of 4 compose)
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (0 of 4 compose)
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (0 of 4 compose)
The toolchain the build seat runs merge-check.sh in (Go 1.26) and a newer local Go format a return of
several multi-line composite literals differently; the seat's is the one that judges, and it failed every
pull request on this file (novox/hq issue 283).
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (0 of 4 compose)
mesh/repo-check fail: its merge-check.sh failed: internal/bootstrap/publish_test.go
The build seat now runs the gate itself, judging a node-engine change with the change's
validator in the running controller because the module graph builds mesh-host from here.
This script is the repository's own layer (mesh/repo-check): format, vet, and the suite
under the race detector when the toolchain can.
merge-check.sh runs the suite, then builds the controller the mesh runs with this change's
validator in place of the one it vendors and runs the merge gate: a change to the node-engine
that would refuse what the mesh sends today fails its pull request, naming the machine. The
installer's first user list lets the controller say a check's verdict and hear a pull request's
head, as the controller now composes it.
The controller says a rollback as rolled-back and asks the bus machine's backup holder
for a snapshot before the planned bus step; the installer's first user list must grant
both, or a fresh mesh's controller is refused the first time it uses them.
The launcher trusted a counter only a by-hand reconcile ever cleared and a
known-good nothing in the daemon wrote, so no machine could roll its host back;
the controller and the node tools were replaced in place with nothing kept.
- The launcher runs a delivered host that is not known-good on trial: one that
crashes, stops for nothing, or does not report within ten minutes goes back
to known-good, once per version, recorded in rolled-back. The host proves
itself when the mesh takes a report under its own build, says every standing
verdict on its reports, never stands aside for a rolled-back version, and
restarts its service once when its launcher was replaced on disk.
- The engine keeps the controller's and the node tools' previous build beside
the new one and judges the new one: the lease taken by the controller it
started (read-only direct get of mesh-controller_lease/holder), or this
machine's runtime answering $SRV.PING.node-tools.<node>, within sixty seconds
of time it could ask. Not healthy: the previous restored, once, said. Proved:
the previous deleted. A build declared not-reversible is never rolled back.
- Retire never removes a version newer than the running one.
The installer's first user list must match the controller's composed grant, which now reads every
machine's backup holder; and a kept directory's report pointed at the one act that loses data.
The controller now derives a subscription to provisioner.retirement; the first
user list the installer carries must match it, or the genesis controller is
refused the subject its composition expects.
The controller's healer H1 answers a send that went unreported by asking the
machine first: a report lost on its way (issue 264) needs no second send. The
node-engine now hears mesh.node.<self>.ask.report on core NATS and enqueues a
reconcile whose account is said whether or not it is news; a delivery waiting
meanwhile is applied and reported instead. The answer is an ordinary report on
its own subject, so the node publishes nothing new and answers nobody's inbox.
The genesis lock grants the controller the healer-acted seat event, which it
now composes; the genesis test in mesh-controller holds the two equal.
The controller now composes a publish grant for its lease bucket
($KV.mesh-controller_lease.>), which it must write before it acts, and for
the seat event secret-replaced (hq ADR 0228, mesh-controller #82); and it no
longer publishes mesh.control.>: a machine's report has one writer, its
node-engine, and the writers table refuses a second at composition. The
installer's first user list must say what the controller derives, or a new
mesh's first controller is refused its lease and serves without one.
A delivery and the five-minute reconcile were two paths that applied, ordered only by a lock, and
each order it allowed was met live (issues 257, 261, 267). Now both only enqueue: one worker takes
the newest declaration held when it starts, applies it once and makes one report, and reports leave
in the order they are made.
A declaration may carry the controller's lease epoch beside its sequence; one older than what this
node applied is refused before anything is touched, counted, logged and reported. A report carries
the declaration's epoch and sequence and the host's own report sequence, kept on disk so it goes on
increasing across restarts and self-updates. Without an epoch, today's behaviour stands.
The controller now composes a publish grant for
mesh.seat.node-intrusion-prevention.tool.banned.*, which D8 asks every
machine; the installer's first user list must say what the controller
derives, or a new mesh's first self-check is refused it.
The controller's watchdog of a machine's heartbeat (S1) is bound to three
of its intervals, and a bound the controller guessed would not move when
the interval does: the heartbeat now carries interval_seconds.
Its self-check (D1) must judge every composed declaration as the host
does, and a second validator written from the host's rules would drift
from them: the host's own parsing is exported, unchanged, as
github.com/novox/mesh-host/validate.
The installer's first user list grants what the controller now composes
for itself: its condition buckets, the condition and doctor-heartbeat
events, the bus's two consumer advisories and $SRV.INFO — or the first
controller would be refused them until the broker's machine is pushed.
The controller keeps its calls and the hand-act log in two key-value buckets
of its own, and composes a grant to write them. The installer's first user
list must say what the controller derives, or the first controller writes
nothing until the broker's machine is pushed.
A reconcile that held the machine to the kept declaration just before a
delivery arrived queued its report while the delivery's apply waited for
it; the link published the apply's report and then the reconcile's, so the
mesh's last word from the machine named the older declaration and the
release plan waited on a report it had already been given. The link now
sets aside an unasked report about a declaration other than the one it
has applied since.
A host that delivered its own successor stood aside before the report of
that apply was published, so it failed with 'context canceled' and the
release plan waited for a report that never came (novox/hq issue 264).
Reports are now published on a context the stand-aside does not cancel,
and the last apply's report is kept until the broker takes it and said
again on the next link, so a crash between apply and report is covered too.
The carried user list must equal what the controller derives (its test reads
this file): add the provisioner.failing/recovered subscriptions (hq ADR 0224)
and the build seats' tool publishes it already derived (hq ADR 0219).
The resolver file moves from resolv-conf to the uplink's holder in one apply.
Orphans go first, so the file was deleted or given back its pre-mesh original
until the new owner's turn came. Now the old record is forgotten once the new
one is recorded at the same path, and the kept original goes with it.
jochen
mesh-admin