Files
cgalo5758 4054159c83 Characterize the relay's walk of a group: the statements in order and the call's counts
TestRelayStatements runs the relay once or twice for each of five
scenarios on a counted database handle: a pool row that completes, a
connection row that pages and then resumes after the last recipient it
reached, a wide row whose failing recipient is requeued and whose own row
fails again, an expansion failure, and a spent record budget that releases
the second row untouched. Each golden holds, per call, the statements in
order, the result's counts, how many records the call touched for the push
dispatch, and the outbox rows of the scenario's subjects after the call
(kind, status, attempts and last error, ids printed by name). The order of
the walk's writes (recompute, requeue, release with its cursor, completion)
was pinned by nothing before, and the rows' last error texts only as
non-empty.

The test retires the connections earlier tests left approved while it runs
(the real providers', once a registration test has run) and restores them
and drains their notices afterwards, so a recipient's row reaches only the
scenario's own connections whatever ran before.

No production file changes, so no output changes. The names and results of
the package equal the parent's apart from the new test and its 5 subtests.
15 breaks, each caught by the new test; the parent has no counterpart to run
them against. The first break (a connection row's page is the whole budget)
is run against the spent-budget scenario, because it needs two deliverable
connections. Unprobed: a requeue, a release or a completion that fails, as no
fixture fails one of those statements and nothing else; a cancellation inside
the page; the relay's Warn and Error lines, which go to the process-wide
logger; the retention delete's count, which is 0 in every file because no
completed row is old enough after the reset.
2026-10-11 11:50:42 -05:00
..