mesh/check checking 4ffe5f4d5243 on 85aca46348a0 | novox/mesh-delivery#25 in group feat/fastcheck
A new pull request waited 25 to 45 minutes behind every open pull request's re-judge after a
merge. The semaphore becomes a ranked admission (ADR 0536 §4); checker_status names the
waiting checks by kind.
mesh/check pass e76202876032 on b714eab995a5 | unit 422 run, 0 kept | integration 467 run, 0 kept | over bound: integration (slowest integration mesh-…
The checker's own places are released when a step ends or its wait passes the step bound; checker_status read every
'running' key in the build modules' steps buckets as a step in flight, so keys left by stopped, restarted or unassigned
build modules looked like full nodes. It now lists them under left_running with why (past the bound, module not on the
node, module not running), and leaves hq-only groups out of the unjudged candidates.
mesh/merge-gate head 9fad90ee890a merged into trunk b714eab995a5 — pass: builds mesh-delivery-builder → novox; no bus step; the change touches no module of…
The references come from the build modules' own pushes, so a new builder beside
an old build module writes none; saying it at start would send the controller
to keys that fail not stored.
The controller delivers from the builder's keys only while a running builder
says its done keys name references; a builder from before R4e says nothing,
so R4e's own delivery is asked of build-agent instead of waiting for ever.
The say is put once started and taken away when stopped.
mesh/merge-gate head efca168c5522 merged into trunk 0e372c87f1df — pass: group feat/r4e composed: every machine composes with the change as it did without …
mesh/delivery covered after its stop: a later walk that contains it was delivered: plan-1792707174627079192 (novox/mesh-delivery at b714eab9 into main, d…
mesh/check failed: unit: ..test: node:internal/modules/esm/resolve:989 | efca168c5522 on b714eab995a5 | unit 2092 run, 0 kept, 3 skipped | integration…
R4e round 2 finding R2-1, ADR 0521: nothing in the delivery plan orders the
forge module before the bridge, and the forge module's durable consumer keeps
its deliver policy when its filters change, so a check.status emitted before
it consumed the event was lost. The forge module now watches the checker's
owed state and writes mesh/check itself, once per revision.
The bridge emits no check.status, declares no emits, and neither reads the
checker's owed nor hears its stream; OwedAct still refuses mesh/check. The
integration test shows the bridge calls no tool and publishes nothing of its
own when the checker owes mesh/check.
R4e round 2 finding R2-1, ADR 0521: the forge module writes mesh/check from
the checker's owed state, read by a watch, because no event order between the
forge bridge and the forge module is kept by the delivery plan. A reader is
granted from the owner's announcement (ADR 0495 §2-§3), and the checker
announced nothing. It now says every state it keeps, with owed, previews and
verdicts readable: the forge module and the bridge read owed, the indexer
previews, the gate verdicts.
Review finding K1 of R4e, issue 1301 S4 (ADR 0503 §3, ADR 0513 §4: the forge
requires mesh/check alone). gitea_check_status rested on "only the bridge
invokes it", but node-tools invokes every tool and a tool is told no caller,
so any session could set mesh/check and merge past the checker.
The bridge now emits check.status on mesh.mod.mesh-delivery-forge-bridge.event.
check.status, which the bus grants to its own account alone (emits), with the
owed status, its key and its revision; the event id check.<key>.<rev> is the
bus's dedup id. A status the bus did not take is not kept as done. OwedAct
refuses mesh/check, and gitea.gitea_check_status leaves the invokes. The
integration test shows the tool runner's account refused on that subject and
gitea's durable consumer handed statuses emitted before it fetched.
The checker's go.mod and go.sum conflicted on the mesh-sdk pin (checker lane 9cb4d59, builder lane e76a5a3): every
module touched is pinned to the sdk lane's newest head d1af87a until the final SDK head.
A merged message or a preview without the objects gives the module none, naming the paths. Every other module keeps its
directory's tree. Module.ArtifactPaths is filled for the checker (the lead's interface).
The indexer holds no published list of predefined keys; each key a value names is taken as the value declares it.
No fingerprint input comes from the list: the gate and the controller still judge the value.
I1: a merge writes each module its manifests name, at its directory. I2: a pull request's changed directory, a
preview's touched module and a source revision's modules come from the modules bucket (a module a pull request adds is
found in the preview by its manifest). I4: the value of today's form (fp, none) is no longer written, since it was
computed from the build records; mesh-sdk keeps it readable until R4g. One commit: the three share applyMerged and its
tests. Pins mesh-sdk feat/r4e-sdk 9cb4d59.
mesh-catalog #293 failed with 'go.mod cannot be read to check its mesh-sdk pin: nats: no responders', put at the
build node, and stayed failed: the forge's tool runner was not answering for a moment, which is not the change's
failure. The forge's reads now mark such a call unheard, as the build steps' calls are (issue 1820).
The line said only that the composition passed with an SDK commit, and its time sat among the parts that are dropped
first, so a forge status never said how long the composition took or which machines it covered.
A TypeScript member of a group that changes mesh-sdk's package resolved the old version from the registry. Its Node
fetch now names the mesh-sdk member's head, which the Node build module packs and serves, publishing nothing. The paths
come from the indexer's modules value (lead decision L8), so a change of Go files or of the manifest alone resolves
the registry as before. A test holds that no step of a check carries publish.
mesh-controller's try asked build-agent, which leaves the checking at R4e. A try is now a key in the checker's state
tries: the checker judges the head merged onto its trunk as it judges a pull request's head, through the same
judge, places and holds, writes its end in the key and owes the forge nothing (no status, no preview, no verdict).
One try per head, at most four at once; hq and a repository whose trunk the gate does not name are refused with
"no level of <repo> is checked by the checker".
Moves the mesh-sdk pin to feat/r4e-sdk 9cb4d59, which accepts such a key and names delivery.WaitsFor; the builder's
package wait now uses that shared prefix (B7).
The forge's gitea_commit_status refuses the context mesh/check from R4e, so that only the checker's account, through
this bridge, can make a pull request pass the one required status.
The gate decides on mesh/check alone and matches an Approve's members by the heads hash over lowercase repositories
and heads. A success judged on an old trunk stood on the forge while the next check waited for a place or a delivery,
and a candidate whose repository the forge named in capitals carried a hash the gate never finds. A check that ends
after its trunk moved now owes pending, not its verdict.
Pins mesh-sdk feat/r4e-sdk a90f523 (R4e's group member). A trunk key's builds are asked to publish; a build answering
no reference fails the key 'not stored:', and a step's own 'not stored:', 'not pinned:', 'version taken:' or
'refused kind:' opens the key's error.
The checker took and kept nothing for a workspace given siblings, and every core repository has siblings, so the
kept store stayed empty. The digest step now folds the siblings' commits in and echoes them; a verdict is kept or
taken only when the echo equals the siblings the fetch was given, so a digest that misses one never poisons another
check (ADR 0444, issue 482).
mesh/merge-gate head 4a0c77105943 merged into trunk 2b635e2138ae — pass: builds mesh-delivery-checker → novox; no bus step; every machine composes with the…
jschoubben