diff --git a/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md b/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md index 5a65f7d7..9ee6d1f6 100644 --- a/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md +++ b/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md @@ -582,3 +582,13 @@ The lead's one-line answers to the questions H2's develop and review raise (impl builder_health a tool beside it, as the runner's manifest has it: a tool check stands only beside a check of another kind the module does not run itself (ADR 0227 rule 8, ADR 0240 rule 2), and the validator refuses a tool-only health. "the health kind the runner uses" read so. ADR 0240. +- **H-A-440 (the publishes queue's consumers and what stays of the publish items, 01:51)** ยง793, B4. (1) The holders + are read at run time (the seat-holders view), so the builder makes the pull consumers at run time as it learns a + holder, one per holder and class, named `publishes--`, as DeclareAskWorkers makes a seat's workers; + the manifest declares the stream `publishes` and the consumers' form in its requires (the work-queue's options: ack + wait = the setting publish-pause-max, max ack pending 4 = "at most four sends in flight per holder", max deliveries + unbounded); "declared with the stream" reads so. (2) The bucket that goes is `publish-queue` (its items put once on the + stream at the first start, deduplicated by message id, then dropped); `reasks` stays, its ModePublish values keeping an + item's tries and next try with the three settings publish-pause, publish-pause-max and publish-tries; a send that + fails is handed back (nak) with the pause as its delay, so the stream redelivers and nothing is put again; after + publish-tries the builder terminates the item and says version-unreleased, loudly. P-NATS-native, P3, P-failure-loud.