33 Commits

Include renames

Author	SHA1	Message	Date
jschoubben	28ee22ef27	Give the integration test's EVENTS stream the artifacts-gone subject alone, so it does not overlap the build seats' streams (issue 1937)	2026-10-24 04:25:19 +02:00
jschoubben	80a0183567	Build a done key again when the artifact store no longer holds its blob, and on the controller's artifacts-gone (issue 1937) ... The store's nightly collector deleted the bare blobs of node-tools', mesh-host's and the forge bridge's done keys, which the controller had never taken in, and the builder kept them done: their next delivery would refuse. The builder now asks the controller's held before it leaves a done trunk key done (once per key per process, at its start and on each module value), claims a key whose blob is gone building as a new revision and says so in its log and the key's step, and hears artifacts-gone, said when a take-in found a gone blob. No timer: a start, a value or an event.	2026-10-24 04:17:01 +02:00
jschoubben	b303e32a41	Close the builder test's nodes from the bed's start, so no pull in hand takes the first ask (issue 1928)	2026-10-23 23:03:38 +02:00
jschoubben	b83af04f0c	Read an ask's stream at take-over by its name, as the broker grants it, and wait for the ask when that read fails rather than fail a key whose build may still finish (issue 1910)	2026-10-23 16:59:37 +02:00
jschoubben	d32ec6b7f8	Say what the builder, the indexer and the watchman do since R4g in their comments: no build-agent is asked, no build record moves a fingerprint, no checks-running bucket is mirrored (issue 1887, review round 2 item 6)	2026-10-23 16:47:36 +02:00
jschoubben	ce51457974	Prove on the real bus that a builder started again while its asks waited puts no second ask (issue 1887 AC3, review round 2 item 2)	2026-10-23 16:46:58 +02:00
jschoubben	bac82bb91e	Count the asks the builder puts and read their kind in its integration test, so 1887/AC1 is proven on the real bus (issue 1887, review round 2 item 3)	2026-10-23 16:46:57 +02:00
jschoubben	357e28f31d	Merge remote-tracking branch 'origin/feat/r4g' into feat/r4g ... # Conflicts: # modules/mesh-delivery-builder/internal/build/run_test.go	2026-10-23 16:46:05 +02:00
jschoubben	5ddc221833	Bound a build or check ask's step from its start, and drop the checker's whole-check bound, so asks waiting in the streams during a large rebuild are never failed and asked twice (issue 1887, review round 2 item 1; ADR 0525 §3, §6)	2026-10-23 16:45:35 +02:00
jschoubben	84b2c92ccf	Put again, under its own id, an ask a build key names that neither its stream holds nor any node ran, so a builder that stopped between naming and putting it leaves no key waiting for an ask nobody put (issue 1910)	2026-10-23 16:39:56 +02:00
jschoubben	9f27e2b1f8	Build each build module's part of a fingerprint as one ask on its seat's stream, named in the key before it is put, so the builder chooses no node, holds no capacity and never asks twice (issue 1887 A4a, A4b, A4c, ADR 0525 §1-§6)	2026-10-23 15:57:03 +02:00
jschoubben	9fad90ee89	Say that the builder writes references only once each build module answered one (issue 1886, ADR 0522 review) ... 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.	2026-10-23 00:52:50 +02:00
jschoubben	4b8b1c047c	Say in the builder's own bucket of instances that it writes references (issue 1886) ... 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.	2026-10-23 00:48:28 +02:00
jschoubben	12cf0cc1f5	Format the builder's integration test	2026-10-22 18:21:04 +02:00
jschoubben	7cc97cd6b6	Name a candidate's named paths in its fetch, and prove on a real bus that claude-code's manifest is built (issues 1758 and 1751)	2026-10-22 18:20:59 +02:00
jschoubben	b39090aab9	Read manifests with mesh-sdk's ParseManifestForInputs in the indexer and the builder, so claude-code, dunst and xterm are built too (issue 1751) ... The indexer's own copy goes. Pins mesh-sdk feat/r4e-sdk e76a5a3 in every module of the group.	2026-10-22 18:20:37 +02:00
jschoubben	4bb04a47f9	Build each fingerprint on one node and drop the second-node retake, as ADR 0478 §4 has it from R4e (issue 1587, B9) ... A key done on one node before R4e is not built again for a second; a key failed for differing digests stays as it is.	2026-10-22 18:18:59 +02:00
jschoubben	8b2e8895c2	Prove on a real bus that every done key names its references, a refused push is built again, and a pinned upstream is copied (issue 1587, B8, ADR 0498 checks)	2026-10-22 18:03:39 +02:00
jschoubben	12f61998c5	Publish a merged candidate's image once; a retaken key builds only what is missing (review r1)	2026-10-22 10:33:52 +02:00
jschoubben	bbc690539b	Build an image on one node through mesh-build-image; only a trunk key publishes (issue 1705)	2026-10-22 10:19:26 +02:00
jschoubben	7214bf4368	Fail a candidate's build when its fetch answers files in conflict or no merge, so the trunk's build is not stored under the candidate's fingerprint (review D4)	2026-10-21 21:50:30 +02:00
jschoubben	284f21c4e3	Key the builder's candidate test as the indexer keys a group	2026-10-21 21:35:53 +02:00
jschoubben	66908d9bbc	The builder builds the fingerprints a candidate carries, once each, from the head merged onto the trunk (issue 1684)	2026-10-21 21:28:38 +02:00
jschoubben	e554094b74	The builder uses mesh-sdk's buildrun package for the caller, the steps, the places, the presence, toolchain and have sets and the choice of nodes, so the checker shares them (issue 1671)	2026-10-21 21:25:00 +02:00
jschoubben	46702d1af3	Keep the builder off a node whose capacity setting is 0, and start the setting at shanks=3,ace=2,g14=0,novox=0 (issue 1661)	2026-10-21 13:46:14 +02:00
jschoubben	dcf62b0413	Make the builder's capacity a module setting, hold back a failed second-node build until the nodes change, and close at start a key left building for a fingerprint the module no longer has (issues 1661, 1664, 1660)	2026-10-21 13:34:39 +02:00
jschoubben	bfc64ac65e	Build an archive with no toolchain, mark digests that differ and a one-node key so each is handled, reserve a place when a node is chosen, choose only nodes whose build module is healthy, and take over any key found building (issue 1607, review round 4)	2026-10-21 13:15:28 +02:00
jschoubben	27e4c3891f	Run at most a node's build capacity of build steps on it at once, and hold a step back for a place with no timer (issue 1607, round 3)	2026-10-21 12:41:42 +02:00
jschoubben	f555a05cfb	Build a key only with the toolchain its fingerprint names, accept '.' as a root module's path, and choose only online nodes (issues 1615, 1607 review)	2026-10-21 11:36:32 +02:00
jschoubben	0b0cfb6e37	Test the builder whole over the real bus with fake build modules, and look again at a value that arrives while its key is built (issue 1607, AC6, AC7)	2026-10-21 11:25:13 +02:00
jschoubben	46c9f33d64	Build each new-form fingerprint through the build modules on two nodes and write its key done or failed (issue 1607, AC7, AC8)	2026-10-21 11:23:39 +02:00
jschoubben	71bb5a5fb3	Watch the build modules' toolchains and steps and keep the toolchains by node, with no timer (issue 1607, AC6)	2026-10-21 11:21:46 +02:00
jschoubben	c3804b012f	Add the builder part: its manifest, its credential and its bus connection (issue 1607, AC1)	2026-10-21 11:18:30 +02:00