184 lines
4.7 KiB
TypeScript
184 lines
4.7 KiB
TypeScript
import type { ChildMessage, ParentMessage, RunJob } from "../runtime/helpers";
|
|
import { WorkerRpcClient } from "./rpc";
|
|
import { WorkerRunner } from "../runtime";
|
|
|
|
const parentPort = process.parentPort;
|
|
|
|
if (!parentPort) {
|
|
throw new Error("AgentCompany worker requires process.parentPort");
|
|
}
|
|
|
|
interface ActiveRun {
|
|
runId: string;
|
|
agentId: string;
|
|
workspaceKey: string | null;
|
|
}
|
|
|
|
/**
|
|
* Single source of truth for scheduling invariants: one active run per
|
|
* agent and one active run per exclusive workspace lock. Prior layout
|
|
* kept three parallel structures (`activeByAgent`, `activeByWorkspace`,
|
|
* queue) that had to be updated in lockstep. The Map-of-ActiveRun keyed
|
|
* by runId is the only authoritative record; agent/workspace busy checks
|
|
* iterate it.
|
|
*/
|
|
const active = new Map<string, ActiveRun>();
|
|
const queue: RunJob[] = [];
|
|
const cancelledRuns = new Set<string>();
|
|
let shuttingDown = false;
|
|
let drainAckResolver: (() => void) | null = null;
|
|
|
|
function post(message: ChildMessage) {
|
|
parentPort.postMessage(message);
|
|
}
|
|
|
|
const rpc = new WorkerRpcClient(post);
|
|
const runner = new WorkerRunner({
|
|
post,
|
|
rpc,
|
|
isCancelled: (runId) => cancelledRuns.has(runId),
|
|
});
|
|
|
|
function workspaceKey(job: RunJob): string | null {
|
|
if (job.workspaceMode !== "exclusive") return null;
|
|
return job.workspaceId ?? job.workspacePath ?? `task:${job.taskId}`;
|
|
}
|
|
|
|
function isAgentBusy(agentId: string): boolean {
|
|
for (const run of active.values()) {
|
|
if (run.agentId === agentId) return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
function isWorkspaceBusy(lockKey: string): boolean {
|
|
for (const run of active.values()) {
|
|
if (run.workspaceKey === lockKey) return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
function enqueue(job: RunJob) {
|
|
if (shuttingDown) return;
|
|
queue.push(job);
|
|
post({
|
|
type: "run-status",
|
|
runId: job.runId,
|
|
status: "queued",
|
|
message: "Run queued in orchestration worker.",
|
|
});
|
|
tickQueue();
|
|
}
|
|
|
|
/**
|
|
* Claim-before-dispatch: for each queued job whose agent + workspace are
|
|
* free, register the claim in `active` synchronously *before* calling
|
|
* `startRun`. No async window between claim and dispatch.
|
|
*/
|
|
function tickQueue() {
|
|
if (shuttingDown) return;
|
|
let i = 0;
|
|
while (i < queue.length) {
|
|
const job = queue[i];
|
|
if (isAgentBusy(job.agentId)) { i += 1; continue; }
|
|
const lockKey = workspaceKey(job);
|
|
if (lockKey && isWorkspaceBusy(lockKey)) { i += 1; continue; }
|
|
active.set(job.runId, { runId: job.runId, agentId: job.agentId, workspaceKey: lockKey });
|
|
queue.splice(i, 1);
|
|
void startRun(job);
|
|
}
|
|
}
|
|
|
|
async function startRun(job: RunJob) {
|
|
try {
|
|
await runner.run(job);
|
|
} finally {
|
|
cancelledRuns.delete(job.runId);
|
|
}
|
|
}
|
|
|
|
function acknowledgeRunCompletion(runId: string) {
|
|
const released = active.delete(runId);
|
|
if (released) tickQueue();
|
|
if (shuttingDown && active.size === 0 && drainAckResolver) {
|
|
drainAckResolver();
|
|
drainAckResolver = null;
|
|
}
|
|
}
|
|
|
|
function waitForCompletionAcks(): Promise<void> {
|
|
if (active.size === 0) return Promise.resolve();
|
|
return new Promise((resolve) => {
|
|
drainAckResolver = resolve;
|
|
});
|
|
}
|
|
|
|
function cancelRun(runId: string) {
|
|
const queuedIndex = queue.findIndex((entry) => entry.runId === runId);
|
|
if (queuedIndex >= 0) {
|
|
queue.splice(queuedIndex, 1);
|
|
post({
|
|
type: "run-finished",
|
|
runId,
|
|
status: "cancelled",
|
|
summary: "Cancelled before execution.",
|
|
errorMessage: null,
|
|
exitCode: null,
|
|
signal: null,
|
|
model: null,
|
|
usageStatus: "aborted_before_usage",
|
|
inputTokens: null,
|
|
outputTokens: null,
|
|
});
|
|
return;
|
|
}
|
|
|
|
cancelledRuns.add(runId);
|
|
runner.cancel(runId);
|
|
}
|
|
|
|
parentPort.on("message", (event) => {
|
|
const message = event.data as ParentMessage;
|
|
if (message.type === "enqueue-run") {
|
|
enqueue(message.payload);
|
|
return;
|
|
}
|
|
if (message.type === "cancel-run") {
|
|
cancelRun(message.runId);
|
|
return;
|
|
}
|
|
if (message.type === "run-completion-ack") {
|
|
acknowledgeRunCompletion(message.runId);
|
|
return;
|
|
}
|
|
if (message.type === "steer-run") {
|
|
const accepted = runner.steer(message.runId, message.message);
|
|
post({
|
|
type: "steer-ack",
|
|
runId: message.runId,
|
|
correlationId: message.correlationId,
|
|
accepted,
|
|
});
|
|
return;
|
|
}
|
|
if (message.type === "rpc-response") {
|
|
rpc.resolve(message.rpcId, message.response);
|
|
return;
|
|
}
|
|
if (message.type === "shutdown") {
|
|
if (shuttingDown) return;
|
|
shuttingDown = true;
|
|
void (async () => {
|
|
queue.splice(0);
|
|
await runner.cancelAll();
|
|
await waitForCompletionAcks();
|
|
rpc.cancelAll("Worker shutting down.");
|
|
post({ type: "shutdown-ready" });
|
|
})();
|
|
return;
|
|
}
|
|
if (message.type === "shutdown-exit") {
|
|
process.exit(0);
|
|
}
|
|
});
|