prisma-risk / tsoracle / 28642785977

95%

Build Type

push

github

Committed by web-flow

Commit Message

fix(standalone): fail fast when a peer or admin server task dies (#644)

A peer or admin transport server that terminated with an error was logged and abandoned: the spawned task exited, the consensus driver kept running, and the node became a zombie — no health signal, no non-zero exit, no restart, while peers timed out and elected around it.

Supervise all three server tasks (openraft peer, paxos peer, admin) through TransportHandle::spawn_supervised: the server future runs in its own task, and any unexpected exit — an error, a clean return nobody requested, or a panic — trips a per-node FatalSignal. The bin's serve loop selects on that signal next to the OS shutdown signal; on a trip it skips the graceful drain (the node is already degraded), stops the client server, tears down the remaining transports, and exits non-zero so an orchestrator restarts the node. Cooperative shutdown and handle drop remain non-fatal.

Fixes #616

Signed-off-by: Sebastian Thiebaud <sebastian@prismarisk.com>

Coverage Stats

178 of 185 new or added lines in 6 files covered. (96.22%)

2 existing lines in 1 file now uncovered.

17425 of 18327 relevant lines covered (95.08%)

312614.83 hits per line