mptyl/ThothII
1
0
Fork 0
Code Issues 25 Pull Requests Actions Packages Projects Releases 1 Wiki Activity

fix: make session resume atomic across restarts

Browse Source
This commit is contained in:
User
2026-07-15 00:42:35 +02:00
parent 2b4797f133
commit 08b1f4909e
12 changed files with 1082 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/src/pi/pi-process-manager.ts
+40 -27
View File
@@ -68,9 +68,14 @@ export class PiProcessManager {
cwd: this.cfg.harnessDir,
env,
});
// Log stderr for debugging (was silently drained)
child.stderr.on("data", (d: Buffer) => console.error(`[pi:${sessionId}] stderr:`, d.toString().trim()));
return child;
try {
// Log stderr for debugging (was silently drained)
child.stderr.on("data", (d: Buffer) => console.error(`[pi:${sessionId}] stderr:`, d.toString().trim()));
return child;
} catch (error) {
try { child.kill(); } catch { /* preserve the initialization error */ }
throw error;
}
}
count(): number { return this.runtimes.size; }
@@ -91,31 +96,39 @@ export class PiProcessManager {
const author = o.author ?? "dev@local";
const provider = canonicalPiProvider(o.provider ?? this.cfg.defaults.provider);
const child = this.spawnFn(sessionId, author, provider);
const rpc = new RpcClient(child);
const bridge = new SessionBridge(rpc);
const rt: SessionRuntime = { rpc, bridge, child };
bridge.beginTurn();
this.runtimes.set(sessionId, rt);
// Identity-checked: a stale child's exit must not evict a newer runtime.
// Expected teardowns (teardown()/respawn) delete the runtime from the map BEFORE the
// exit event fires, so reaching this branch with `rt` still mapped means the child
// died on its own: tell the client, or the UI spins forever waiting for a turn end.
child.on("exit", (code) => {
console.error(`[pi:${sessionId}] exited code=${code ?? "?"} mapped=${this.runtimes.get(sessionId) === rt}`);
if (this.runtimes.get(sessionId) === rt) {
rt.bridge.markFailed();
this.runtimes.delete(sessionId);
rt.bridge.emitClientEvent({
type: "info",
level: "error",
text: `Pi process exited unexpectedly (code ${code ?? "?"})`,
});
rt.bridge.emitClientEvent({ type: "system_event", event: "session_failed" });
rt.bridge.emitClientEvent({ type: "system_event", event: "agent_end" });
}
});
let rt: SessionRuntime | undefined;
try {
const rpc = new RpcClient(child);
const bridge = new SessionBridge(rpc);
const runtime: SessionRuntime = { rpc, bridge, child };
rt = runtime;
bridge.beginTurn();
this.runtimes.set(sessionId, runtime);
// Identity-checked: a stale child's exit must not evict a newer runtime.
// Expected teardowns (teardown()/respawn) delete the runtime from the map BEFORE the
// exit event fires, so reaching this branch with `rt` still mapped means the child
// died on its own: tell the client, or the UI spins forever waiting for a turn end.
child.on("exit", (code) => {
console.error(`[pi:${sessionId}] exited code=${code ?? "?"} mapped=${this.runtimes.get(sessionId) === runtime}`);
if (this.runtimes.get(sessionId) === runtime) {
runtime.bridge.markFailed();
this.runtimes.delete(sessionId);
runtime.bridge.emitClientEvent({
type: "info",
level: "error",
text: `Pi process exited unexpectedly (code ${code ?? "?"})`,
});
runtime.bridge.emitClientEvent({ type: "system_event", event: "session_failed" });
runtime.bridge.emitClientEvent({ type: "system_event", event: "agent_end" });
}
});
return rt;
return runtime;
} catch (error) {
if (rt && this.runtimes.get(sessionId) === rt) this.runtimes.delete(sessionId);
try { child.kill(); } catch { /* preserve the initialization error */ }
throw error;
}
}
/** Configure model and thinking. Safe to run alongside deterministic retrieval. */