diff --git a/03-DESIGN/01-to-be/47-milestones/H-deliver-the-delivery-system.md b/03-DESIGN/01-to-be/47-milestones/H-deliver-the-delivery-system.md index 83ae352c..95b6c060 100644 --- a/03-DESIGN/01-to-be/47-milestones/H-deliver-the-delivery-system.md +++ b/03-DESIGN/01-to-be/47-milestones/H-deliver-the-delivery-system.md @@ -720,4 +720,4 @@ Added by the lead at 21:30 CET, from the run of 19:50 to 21:30; a row met again | At 21:47 the operator said the lead can and should merge the pull requests, and at 21:48 that they cannot be bothered with them, so merges are the lead's from here, as the mesh's rule on merges says, and the lead will not put a merge to the operator again. By the lead's reading, the earlier refusals were the lead session's own permission check reading the handovers' words ("the operator decides", "a check refused the earlier dispatch") as routing around a review, not a rule of the mesh; a plain handover quoting the operator passed. An agent merges the six at 21:48, in this order: the node-engine's #244 first (it ends the restart loop; its pipeline delivers the node-engines, and then the held puts are released); the delivery repository's #317, #318 and #319 as one call; then hq's #797 and #798, through the forge's interface (rows 174, 187 and 200); the runbook. | H1 (closed); other | | The merges landed: the node-engine's fix (#244) merged at 21:48:30, and its pipeline passed at 21:50:53, two and a half minutes from merge to pass, the node-engine's want moving on the home server and the control-node at 21:51 (the other two and the node-engines' restarts are being read; the held puts of row 183 are released once the restart loop has stopped); the three delivery fixes (#317, #318 and #319) merged as one call at 21:49, and their pipeline of 23 movers runs. hq's two pull requests (#797, this record, and #798, the packet filter's designs) have no merge path in the mesh: the forge tool never merges (ADR 0266), and the merge verb takes only a watched branch, which hq, registering no module, is not; the lead session's permission check refused dispatching hq's registration as a source, so the operator merges them at the forge for now. The fix of the journal lines (H-A-320 (1)) is built and in review (rows 176, 200 and 201); the node-engine, the pipeline runner (`mesh-delivery-pipeline`), the indexer (`mesh-delivery-indexer`), the gate and the checker. Owed, by H-A-332: a repository with no module registers as a source only; its own check runs as the mesh's; its pull requests merge like any other. | H1 (closed); H2 (hq as a source, the journal lines) | | The node-engine's fix is delivered: the new node-engine ran on the home server and the operator's desktop from 21:52, each restarting the tool runner once more for a real change (a secret filled) and then no more; the held puts of row 183 were released at 21:54 (the home server) and 21:58 (the operator's desktop and the control-node), and the loop of rows 179 to 181 is over. One smaller flap of the same kind remains: the reads runtime restarts for the node-files holder's words derived from held modules' stubs, and H-A-335 extends the held-module rule to every runtime process the node-engine composes (in build). The fix of the journal lines (H-A-320 (1)) went to a second round under H-A-333 (every said end writes its line in the one place the event is said; the builder's answered line at its one finish; the asked line kept). The next milestone's design part, H2 "the delivery system's owed work", was written (forty-three lines, §775 to §817, with the two proofs: a change delivered in under ten minutes, read from the gate's new delivery time, with no hand act and no restart by hand) and pushed, with hq as a watched source as a line (H-A-334) and the bend of the destroy channel kept; its four reviews by architectural domain run, and a progressive insight on ADR 0668 adds the fault class "away" (rows 181, 200 and 202); the node-engine, the pipeline runner (`mesh-delivery-pipeline`) and the builder. | H1 (closed); H2 (begun) | -| The pipeline of the three merged delivery fixes (rows 198 to 200: the cut placing, the stranded want of room, the versioned end) ended failed at 22:00: the control-node's node-engine restarted its tool runner for a changed fingerprint in one module's words at 21:59, while the forge's bundle took a package's first chunk, so the publish died and the contracts module's build failed, and every mover downstream was not asked; the change itself was sound (every test run passed, and the same build passed on its next try at 22:05). The next merge's pipeline (the journal-lines fix, row 202, merged at 22:01) carried the same movers and re-pinned them to the published packages at 22:06, so nothing was run again by hand. A design's part is reviewed before its prepare: the four reviews by architectural domain of H2's part reported thirty-seven musts and notes in twelve minutes, answered in one batch (H-A-338 to H-A-341) and applied by the writer; and a TypeScript bundle's mesh packages get no catalog edge before its first build (H-A-336, owed in H2b) (rows 179, 198 to 200, 202 and 203); the node-engine, the tool runner (`node-tools`) and the Node build module. Owed, by H-A-342: a publish refused by a holder that is away ends the run again like a fetch; and the tool runner drains its calls in flight at a stop, within a bound, before the node-engine starts the new one. | H1 (closed); H2 (begun) | +| The pipeline of the three merged delivery fixes (rows 146, 170, 184, 187 and 190: the cut placing, the stranded want of room, the versioned end) ended failed at 22:00: the control-node's node-engine restarted its tool runner for a changed fingerprint in one module's words at 21:59, while the forge's bundle took a package's first chunk, so the publish died and the contracts module's build failed, and every mover downstream was not asked; the change itself was sound (every test run passed, and the same build passed on its next try at 22:05). The next merge's pipeline (the journal-lines fix, row 202, merged at 22:01) carried the same movers and re-pinned them to the published packages at 22:06, so nothing was run again by hand. A design's part is reviewed before its prepare: the four reviews by architectural domain of H2's part reported thirty-seven musts and notes in twelve minutes, answered in one batch (H-A-338 to H-A-341) and applied by the writer; and a TypeScript bundle's mesh packages get no catalog edge before its first build (H-A-336, owed in H2b) (rows 146, 170, 179, 184, 187, 190, 202 and 203); the node-engine, the tool runner (`node-tools`) and the Node build module. Owed, by H-A-342: a publish refused by a holder that is away ends the run again like a fetch; and the tool runner drains its calls in flight at a stop, within a bound, before the node-engine starts the new one. | H1 (closed); H2 (begun) |