548 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
mesh-tools	d48365aed0	Merge pull request 'frombus: a node composing itself takes its modules from their builds and reads what providers elsewhere serve (issue 1126, group fix/1126-node-creds)' (#109) from fix/1126-node-creds into main	2026-10-27 20:32:30 +00:00
jschoubben	7a8e9f9706	Wait for a done key whose artifact names no reference, as the controller's take-in does, and hold the fold guard with a row (issue 1126, review round 1)	2026-10-27 21:12:13 +01:00
jschoubben	d7a758894a	Leave a module the node runs inside its own composition to its manifest, as the fold composes it, so taking builds in moves no folded module (issue 1126)	2026-10-27 21:11:20 +01:00
jschoubben	04de3e5da2	Compose a node's own modules from their builds and read what providers elsewhere serve, so a self-composing node hands moved modules and settles served settings as the controller does (issue 1126) ... frombus composed every module from the indexer's raw manifest: no moved module was handed, so the node stood in for own secrets the vault leaves to the provisioner (the seed's refusals on all four nodes), and a provider elsewhere had no config values, so mailu's served domain refused shanks. - takein: the controller's take-in of a done key (artifacts, launchers, the templates archive's texts, Resolve) as one SDK rule; catalogue.Launchers holds the launcher rule buildstep now delegates to. - frombus: each module placed on the composing node, and each kept at another build (Scope.KeptAt, a fingerprint), composes from its built manifest; another node's modules give the keys their served values fill (Reader.Served). - buscompose: Declared.Served and Principal.ServedReads grant every node the read of exactly those keys of mesh-config_resolved. - compose.ArtifactStoreAddress: the store as the node's composition routes to it.	2026-10-27 21:11:20 +01:00
mesh-tools	9f92233a28	Merge pull request 'process: a module's own bus connection, subjects and durable consumer in one helper (issue 1242)' (#108) from fix/1242-bus-helper into main	2026-10-27 19:49:19 +00:00
jschoubben	8ad9372b19	Name another module's own-stream subject in the SDK too (issue 1242)	2026-10-27 20:38:20 +01:00
jschoubben	3236cbad43	Give a module's own credential its dial, subjects and durable consumer in mesh-sdk, so a part no longer hand-writes them (issue 1242)	2026-10-27 20:36:14 +01:00
mesh-tools	a262c7dbe0	Merge pull request 'A node folds a placed module with no new-form build into its own composition, as the controller does, so the tunnel and the bus are not dropped when nodes compose themselves (issue 1126)' (#107) from fix/1126-fold into main	2026-10-27 08:09:03 +00:00
jschoubben	05b6783103	Fold a module no new-form build names into the node's own composition, as the controller does, so a self-composing node keeps the tunnel and the bus (issue 1126) ... mesh-wireguard (built by nobody) and every module a node runs inside its own composition (Node.InOwn, from frombus.InOwn of what it runs) were cut out as skipped: every node's shadow answered differs to the placement seed, and a node composing itself after it would drop their resources. compositions now folds them in place, module taken off, filled paths with them, as mesh-controller's compositionsOf does; a want naming a new-form build still moves one out.	2026-10-27 08:50:26 +01:00
mesh-tools	56de5ba840	Merge pull request 'A placement change is answered by the nodes it places: the candidate contract, composing a given placement, and replicated seats in the placement check (issue 1126, ADR 0559)' (#106) from feat/1126-l2b into main	2026-10-27 07:21:03 +00:00
mesh-tools	43eef46c41	Merge pull request 'A build step names the process that runs it, so a restarted build module clears what a gone process left running (issue 1890)' (#105) from fix/1890-step-owner into main	2026-10-27 07:19:36 +00:00
jschoubben	cadd7b1b2f	Ask the vault in a candidate answer only for credentials of modules the node runs, and refuse a stand-in of a running module, so a placement that adds a module is answered as added (issue 1126, ADR 0559 §4)	2026-10-27 02:43:10 +01:00
jschoubben	062c29c3a2	Judge a node's own composition in its candidate answer only where it runs one, as its agent applies it only on a want (issue 1126)	2026-10-27 02:27:04 +01:00
jschoubben	a37d2622ed	Name the words of a composition failed only for an answer not heard, and what an answer's revision is, so the assigner and the checker judge a late answer alike (issue 1126)	2026-10-27 02:22:59 +01:00
jschoubben	ba398e2460	Answer a placement candidate as ADR 0559 has it: added and removed modules, per-module refusals, both digests of a difference, and a node's read of every candidate key (issue 1126)	2026-10-27 02:19:27 +01:00
jschoubben	82f10c3bc9	Let frombus and the composer compose with a given placement and compare a node's compositions with what it runs, so a placement candidate can be answered and checked (issue 1126)	2026-10-27 02:12:11 +01:00
jschoubben	7e99c354ac	Give a catalogue pull request's placement a candidate state and each node an answer in its own have, so a placement change is checked by the nodes it places before it merges (issue 1126)	2026-10-27 02:07:47 +01:00
jschoubben	8e81d87610	Let a replicated mesh seat be held on several nodes in CheckPlacements, so the seed of today's placement is not refused (issue 1126)	2026-10-27 02:04:01 +01:00
jochen	db1007ffdd	Delete a gone process's ask key instead of ending it, so its redelivery still answers the asker (issue 1890) ... ADR 0525 §6 has the stream redeliver an ask whose node stopped, and the asker takes the first end it reads: ending the key at a restart failed checks a redelivery would still run. The sweep now deletes such a key, guarded by the revision read, so the asker keeps waiting and the stale key leaves the counts of what runs; a hand-started step of a gone owner is still ended at once. A missing boot id is said once at start.	2026-10-27 01:29:48 +01:00
jochen	981a35b598	End a step left by a previous process at once: every step names its owner (issue 1890) ... A delivery that replaced mesh-build-go left a check ask's key running, and the new process's start sweep spared it as an ask younger than the step bound, so it held a build slot until a hand end (issue 1890, note 21). Each step now records its owner (boot id, pid, process start); the sweep ends a running step of another owner at once and says its owner, while a key with no owner keeps the earlier rule. The conformance kit gains the left-running case every build module runs through its own start sweep.	2026-10-27 01:19:29 +01:00
mesh-tools	b0cbc4e6ac	Merge pull request 'A build step's failure contract: go -json and TAP parsers and a conformance kit every build module runs, so a verdict names the failing test (issue 1901)' (#104) from fix/1901-failure-contract into main	2026-10-26 23:44:13 +00:00
jschoubben	a2759ec92e	Merge main into fix/1901-failure-contract: the capacity tests stay one function, so the SDK's test count does not grow (issue 1901)	2026-10-27 00:39:10 +01:00
mesh-tools	f178323b57	Merge pull request 'The mesh-vault seat promises own, the verb L2 gave the vault (issue 1126)' (#103) from fix/1126-vault-seat-own into main	2026-10-26 22:00:44 +00:00
mesh-tools	684acb922d	Merge pull request 'Delivery records are true at the source: the layouts for what a repository's merges reached (issue 1280)' (#102) from fix/delivery-records-true into main	2026-10-26 21:48:43 +00:00
jschoubben	e78e881ecb	The vault's seat promises own, which L2's vault serves and its claim names, so the controller's catalogue check passes (issue 1126) ... mesh-catalog's mesh-vault claims its seat serving own (a machine's node-engine asks its own machine's secrets and pair credentials sealed to its stated key), and buscompose already grants that call to node principals (AsksItsOwnSecrets, VaultOwn); only the seat's verb list lacked it, so a claim naming it was refused.	2026-10-26 22:38:58 +01:00
jschoubben	68d65e6bbf	Merge remote-tracking branch 'origin/main' into fix/delivery-records-true	2026-10-26 22:19:13 +01:00
mesh-tools	823d52051f	Merge pull request 'Nodes compose: placement, node records and one composer input for three callers (issue 1126)' (#98) from feat/1126-nodes-compose into main	2026-10-26 21:11:05 +00:00
jschoubben	b4310dfb8e	Merge remote-tracking branch 'origin/main' into fix/delivery-records-true	2026-10-26 22:04:42 +01:00
jschoubben	9796444201	Merge remote-tracking branch 'origin/main' into feat/1126-nodes-compose ... # Conflicts: # test/standing.test.ts	2026-10-26 22:03:26 +01:00
mesh-tools	7187beaf04	Merge pull request 'The provider loop's tests wait on what they mean, not on the clock (issue 1901 note 27)' (#101) from fix/standing-test-races into main	2026-10-26 20:58:44 +00:00
jschoubben	5f45175ed0	Merge remote-tracking branch 'origin/main' into feat/1126-nodes-compose	2026-10-26 21:56:47 +01:00
jschoubben	18e541c1ba	The standing test allows 10 ms between the loop's clock and the test's stamp of an announcement	2026-10-26 21:55:46 +01:00
jschoubben	cf930388f9	The standing test's runs last 0.6 of failingAfterMs of clock as well as two attempts, so a success that does not reset a run is caught; the addition test holds a thousand stable passes, so 'first pass' is proved (review of the standing test races)	2026-10-26 21:49:03 +01:00
jschoubben	36ae0c691b	The standing and retirement tests wait on what they mean, not on a slept fraction of a bound: announcements are timed from the clock at the moment they are said, short runs are counted in attempts, an addition is proved by a hold it must not wait for (the second timing race of one kind in a day; issue 1126)	2026-10-26 21:44:06 +01:00
mesh-tools	2cbe219371	Merge pull request 'A build seat's capacity follows its CPU share and its build hours; asks say their phases (issues 1890, 1806)' (#100) from fix/capacity-follows-cpu-share into main	2026-10-26 20:33:28 +00:00
jschoubben	3a095a66be	The standing test keeps c failing alone after its ask is dropped, so its only recovery is its retirement: clearing every failure at once raced the loop's read of the new list under load (issue 1126; the hidden failure of try-1793042418286545971, issue 1901)	2026-10-26 21:33:20 +01:00
jschoubben	805b4d9a35	Merge main into fix/capacity-follows-cpu-share: the build modules' binary refusal (sdk #99) beside the capacity work	2026-10-26 21:26:58 +01:00
jschoubben	d00dac140b	A build module takes asks only within its node's build hours ... The operator, 2026-10-26 21:18: g14 builds only from 18:00 to 06:00. A build-hours setting names a window per node in its local time (crossing midnight allowed, absent = any hour); outside it the capacity judged by is 0, running steps finish, and the puller wakes on one timer at the next edge. The change is said in the log and in capacity.<node> with the window and when it opens.	2026-10-26 21:21:37 +01:00
jschoubben	e44cbaf57f	A module with no floor holds no head back; a merge among its changes reads not yet on the machines (review of issue 1280, ADR 0556 clause 2)	2026-10-26 21:12:48 +01:00
jschoubben	1f70c4c0aa	A settle ask's group begins with a colon, which no branch name holds, so a branch named settle/... still merges on its Approve (review of issue 1280)	2026-10-26 21:12:31 +01:00
jschoubben	5ecda46124	A settle ask's group and the act its options bind, shared by the gate and the teller (ADR 0556 clause 5)	2026-10-26 21:12:31 +01:00
jschoubben	aa8e9afc7c	A module's floor counts every running have, healthy or not; health stays in its words (ADR 0556 clause 2, insight)	2026-10-26 21:12:31 +01:00
jschoubben	54a8afac73	The view's reach carries the reached head and each change's source revision; nothing records a merge as delivered (ADR 0556 v8)	2026-10-26 21:12:31 +01:00
jschoubben	43de3c937d	Layouts of the indexer's taken and the view's reach, and the delivery seat's settle verb, so whether a merge is on the machines can be derived (issue 1280, ADR 0556)	2026-10-26 21:12:31 +01:00
jschoubben	03dce12378	The standing test keeps c failing alone after its ask is dropped, so its only recovery is its retirement: clearing every failure at once raced the loop's read of the new list under load (issue 1126; the hidden failure of try-1793042418286545971, issue 1901)	2026-10-26 20:33:23 +01:00
mesh-tools	dae8e288d2	Merge pull request 'A change that adds a compiled binary is refused, in any repository (issue 1577)' (#99) from fix/1577-checker-refuses-binaries into main	2026-10-26 19:32:25 +00:00
jschoubben	dd9d9d5b5f	Compare haves with SameHave in the round-trip tests, now that a have carries a list (issue 1126)	2026-10-26 20:20:12 +01:00
jschoubben	2c0ab3a592	A have carries the values a self-composing node's running composition holds sealed to its key, so its composer reuses them after a restart (review of L2, round 2; issue 1126, ADR 0547 §3)	2026-10-26 20:19:54 +01:00
jschoubben	47a66e3c0d	Take a crashed file's cause only from what the run wrote after the last test, give a subtest's panic to that subtest, and keep the SDK's test count (issue 1901, review round 1)	2026-10-26 20:17:42 +01:00
jschoubben	3968849b1b	Review of L2: the export is a placement file at its version with its modules (S1); a node asks the vault only when a module placed on it declares an own secret, so one with none composes whether or not the vault answers (H8's failure) (issue 1126)	2026-10-26 20:08:32 +01:00