prisma-risk / tsoracle / 26419802249
95%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 77
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 95.85%. Remained the same
26419802249

push

github

web-flow 
fix(stress): surface scheme-less paxos leader hints in the topology fixture (#446)

The paxos stress topology built its toolkit peers with an `http://`-schemed
endpoint, which `StandaloneHost` surfaces verbatim as
`LeaderState::Follower::leader_endpoint`. That violates the scheme-less
`host:port` contract the server's `run_leader_watch` guard enforces (a TLS
client silently drops `http(s)://` redirect hints), so once a follower had
observed a stable leader long enough to surface its hint, the guard's
`debug_assert` panicked the watch task. The panic propagates out of
`serve_with_listener` as a returned error; the topology's server task then ends
and drops the `StandaloneHost`, whose `PaxosRunner::Drop` stops that node's
OmniPaxos tick task. Under slow CI both followers reached that state before
`kill_leader`, leaving no quorum to re-elect — surfacing as the
`kill_leader_triggers_reelection` re-election timeout. On a fast host the test
finished before any follower settled, so it only flaked under CI timing.

Extract the peer construction into `build_toolkit_peers` emitting a scheme-less
`host:port` (the in-process `MeshSink` routes by node id, so the endpoint string
is only ever used as a leader hint, never dialed). Pin the contract with a
deterministic unit test at the construction site: the runtime guard fires only
when a follower has observed a stable leader, which is timing-dependent, whereas
the construction is timing-independent.

11270 of 11758 relevant lines covered (95.85%)

510121.78 hits per line

Jobs
ID Job ID Ran Files Coverage
1 26419802249.1 77
95.85
 GitHub Action Run
Source Files on build 26419802249