fix: wait for stale runner lease before replacement claim

This commit is contained in:
Codex
2026-06-03 23:49:40 +08:00
parent 0dfe709fd4
commit 0d040a33c2
3 changed files with 114 additions and 2 deletions
+74 -1
View File
@@ -19,6 +19,8 @@ export interface RunnerOnceOptions extends BackendAdapterOptions {
logPath?: string;
idleTimeoutMs?: number;
pollIntervalMs?: number;
claimRetryTimeoutMs?: number;
claimRetryIntervalMs?: number;
oneShot?: boolean;
}
@@ -50,7 +52,7 @@ export async function runOnce(options: RunnerOnceOptions): Promise<JsonRecord> {
}) as RunnerRecord;
let claimed: RunRecord;
try {
claimed = await api.claim(options.runId, runner.id, leaseMs);
claimed = await claimRunWithLeaseRecovery(api, options, runner, attemptId, leaseMs);
await api.heartbeat(options.runId, runner.id, leaseMs);
} catch (error) {
const failureKind = failureKindFromError(error);
@@ -355,6 +357,77 @@ async function appendBestEffort(api: RunnerManagerApi, runId: string, event: Bac
}
}
async function claimRunWithLeaseRecovery(api: RunnerManagerApi, options: RunnerOnceOptions, runner: RunnerRecord, attemptId: string, leaseMs: number): Promise<RunRecord> {
const startedAt = Date.now();
const timeoutMs = normalizeClaimRetryTimeoutMs(options.claimRetryTimeoutMs, leaseMs);
const intervalMs = normalizeClaimRetryIntervalMs(options.claimRetryIntervalMs);
let waitingEventWritten = false;
let lastError: unknown = null;
while (Date.now() - startedAt <= timeoutMs) {
try {
const claimed = await api.claim(options.runId, runner.id, leaseMs);
if (waitingEventWritten) {
await appendBestEffort(api, options.runId, {
type: "backend_status",
payload: { phase: "runner-claim-recovered", attemptId, runnerId: runner.id, waitedMs: Date.now() - startedAt },
});
}
return claimed;
} catch (error) {
if (failureKindFromError(error) !== "runner-lease-conflict") throw error;
lastError = error;
const run = await getRunBestEffort(api, options.runId);
if (!waitingEventWritten) {
waitingEventWritten = true;
await appendBestEffort(api, options.runId, {
type: "backend_status",
payload: {
phase: "runner-claim-waiting-for-stale-lease",
attemptId,
runnerId: runner.id,
claimedBy: run?.claimedBy ?? null,
leaseExpiresAt: run?.leaseExpiresAt ?? null,
retryTimeoutMs: timeoutMs,
},
});
}
const remainingMs = timeoutMs - (Date.now() - startedAt);
if (remainingMs <= 0) break;
await sleep(Math.min(remainingMs, claimRetryDelayMs(run, intervalMs)));
}
}
throw lastError ?? new AgentRunError("runner-lease-conflict", `run ${options.runId} could not be claimed before retry timeout`, { httpStatus: 409 });
}
async function getRunBestEffort(api: RunnerManagerApi, runId: string): Promise<RunRecord | null> {
try {
return await api.getRun(runId);
} catch {
return null;
}
}
function claimRetryDelayMs(run: RunRecord | null, intervalMs: number): number {
const leaseExpiresAt = run?.leaseExpiresAt ? Date.parse(run.leaseExpiresAt) : NaN;
if (Number.isFinite(leaseExpiresAt)) {
const untilExpiryMs = leaseExpiresAt - Date.now() + 100;
if (untilExpiryMs > 0) return Math.max(25, Math.min(intervalMs, untilExpiryMs));
}
return intervalMs;
}
function normalizeClaimRetryTimeoutMs(value: number | undefined, leaseMs: number): number {
if (Number.isFinite(value ?? NaN)) return Math.max(0, Math.floor(value!));
return Math.max(leaseMs + Math.max(5_000, Math.floor(leaseMs / 3)), leaseMs);
}
function normalizeClaimRetryIntervalMs(value: number | undefined): number {
if (!Number.isFinite(value ?? NaN)) return 5_000;
return Math.max(25, Math.min(30_000, Math.floor(value!)));
}
function annotateCommandEvent(event: BackendEvent, commandId: string, attemptId: string, runnerId: string): BackendEvent {
return { ...event, payload: { ...event.payload, commandId, attemptId, runnerId } };
}