Merge pull request #5453 from oasisprotocol/peternose/bugfix/runtime-…

…abort

go/worker/compute: Abort runtimes only on timeouts

.changelog/5453.bugfix.md

@@ -0,0 +1 @@
+go/worker/compute: Abort runtimes only on timeouts

go/runtime/host/sandbox/sandbox.go

@@ -446,7 +446,7 @@ func (r *sandboxedRuntime) handleAbortRequest(rq *abortRequest) error {
 		return nil
 	}
 
-	r.logger.Warn("restarting runtime", "force_restart", rq.force, "abbort_err", err, "abort_resp", response)
+	r.logger.Warn("restarting runtime", "force_restart", rq.force, "abort_err", err, "abort_resp", response)
 
 	// Failed to gracefully interrupt the runtime. Kill the runtime and it will be automatically
 	// restarted by the manager after it dies.

go/worker/compute/executor/committee/node.go

@@ -691,9 +691,15 @@ func (n *Node) runtimeExecuteTxBatch(
 	}()
 
 	// Ensure batch execution is bounded.
+	//
+	// Note: We intentionally never abort batch execution when ctx is canceled
+	// (e.g., when a round ends or when a proposal with a higher rank is received)
+	// to prevent runtimes from restarting, as abort requests are currently not
+	// supported. Execution shouldn't take a significant amount of time anyway
+	// unless something is seriously wrong.
 	proposerTimeout := state.Runtime.TxnScheduler.ProposerTimeout
 	callCtx, cancelCallFn := context.WithTimeoutCause(
-		ctx,
+		context.TODO(), // Replace with ctx once runtimes start supporting abort requests.
 		executeBatchTimeoutFactor*proposerTimeout,
 		errors.New("proposer timeout expired"),
 	)