jschoubben
79993fb498
The bundles refactor took ADR 0188 on main, so this work's record is 0201 and every comment citing it moves with it. Main also took migration 0055 (an older build never replaces a newer), so the store's collected-artifacts table is 0056 — a number two migrations share is a schema nobody can trust. make check passes except TestTheResolverIsToldEveryMachineOnTheNetworkAndToldAgainWhenOneLeaves, which fails on main too and now for two stacked reasons (hq issues 203 and 202).
24 lines
1.4 KiB
SQL
24 lines
1.4 KiB
SQL
-- What the artifact store no longer keeps (novox/hq ADR 0189, issue 108).
|
|
--
|
|
-- The mesh removes from its store only what it put there and can account for: every digest it
|
|
-- could remove is already in a build record, so the sweep reads its own records rather than
|
|
-- enumerating the store. What it does not get from those records is whether it has already
|
|
-- removed something -- `build.made` says what that build published, for ever, which is history
|
|
-- and not an index of what is on disk.
|
|
--
|
|
-- Without this the sweep would reissue a delete for every artifact it has ever collected, every
|
|
-- time it runs, and each one would answer 404 -- a number of requests that grows with the mesh's
|
|
-- whole history and never shrinks.
|
|
--
|
|
-- Keyed by the reference as the mesh records it (`artifact-store://<module>/<artifact>@sha256:…`),
|
|
-- because that is the identity the record uses everywhere else. Not a foreign key to build: two
|
|
-- builds can publish the same digest (the same source built twice produces the same bytes), and
|
|
-- what is collected is the artifact, not the attempt that made it.
|
|
create table artifact_collected (
|
|
reference text primary key,
|
|
|
|
-- When the store answered. Kept so a reader of an old build record can tell "this artifact is
|
|
-- gone" from "this artifact was never there", which are different kinds of surprise.
|
|
at timestamptz not null default now()
|
|
);
|