1
95%
main: 95%

Ran 23 May 2026 11:27PM UTC

Files 74

Run time 2s

Badge

Embed ▾

Committed 23 May 2026 11:23PM UTC coverage: 94.721% (+0.01%) from 94.71%

Job # 26346282296.1

Build Type

push

github

Committed by

web-flow

Commit Message

fix(driver-paxos): mint a fresh apply-shutdown Notify per start (#232)

StandaloneHost reused a single Arc<Notify> for apply-task shutdown across
its whole lifetime, and stop() signalled it with notify_one(). notify_one
stores a permit when no task is parked on notified(), so a stop() with no
running apply task (stop-before-start, double-stop, or stop after the task
already exited) left a stale permit behind. The next start() spawned an
apply task that consumed the permit on its first select! turn and broke out
of the loop immediately. The runner kept deciding entries, but with no apply
task to drain them the in-memory high-water never advanced and reads waiting
on the apply notifier (current_high_water/submit_advance) hung indefinitely.

Mint a fresh Notify in each start() and publish it through an Option that
stop() takes. A per-start Notify confines every permit to the task it was
minted for, and the Option lets stop() signal nothing when no task is live,
so no permit can survive to poison a later start(). This preserves the
existing notify_one-over-notify_waiters behaviour that handles a stop() while
the apply task is mid-drain.

Add a regression test that poisons each host with a pre-start stop(), then
asserts a decided Advance still folds into every node's high-water.

Coverage Stats

10210 of 10779 relevant lines covered (94.72%)

1497615.96 hits per line

prisma-risk / tsoracle / 26346282296 / 1
95%
main: 95%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 26346282296.1

prisma-risk / tsoracle / 26346282296 / 1 95% main: 95%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 26346282296.1

prisma-risk / tsoracle / 26346282296 / 1
95%
main: 95%

README BADGES
x