feat: 实现 Queue Q2 受控 dispatch
This commit is contained in:
@@ -5,7 +5,7 @@ import { AgentRunError } from "../common/errors.js";
|
||||
import { redactJson } from "../common/redaction.js";
|
||||
import type { BackendProfile, BackendTurnResult, CommandRecord, CommandState, CreateCommandInput, CreateQueueTaskInput, CreateRunInput, EventType, FailureKind, JsonRecord, JsonValue, QueueCommanderSnapshot, QueueReadCursorRecord, QueueStats, QueueTaskListResult, QueueTaskRecord, QueueTaskState, RunEvent, RunnerJobRecord, RunnerRecord, RunRecord, RunStatus, SessionRecord, SessionRef, TerminalStatus } from "../common/types.js";
|
||||
import { newId, nowIso, stableHash } from "../common/validation.js";
|
||||
import type { AgentRunStore, ListQueueTasksInput, SaveRunnerJobInput, StoreHealth } from "./store.js";
|
||||
import type { AgentRunStore, ListQueueTasksInput, SaveRunnerJobInput, StoreHealth, UpdateQueueTaskAttemptInput } from "./store.js";
|
||||
import { assertSessionBoundary, buildQueueStats, clampQueueLimit, commandStateFromTerminal, isLeaseExpired, isTerminalCommandState, isTerminalQueueTaskState, isTerminalRunStatus, parseQueueCursor, queueTaskSort, sessionRefFromRecord, statusFromTerminal, summarizeResourceBundleRef, summarizeSessionRef } from "./store.js";
|
||||
import { backendCapabilitiesSqlValues, mergeBackendCapability } from "../common/backend-profiles.js";
|
||||
import { normalizeRunEventPayload, requireEventType } from "../common/events.js";
|
||||
@@ -606,6 +606,23 @@ CREATE TABLE IF NOT EXISTS agentrun_schema_migrations (
|
||||
return queueTaskFromRow(row);
|
||||
}
|
||||
|
||||
async updateQueueTaskAttempt(taskId: string, input: UpdateQueueTaskAttemptInput): Promise<QueueTaskRecord> {
|
||||
return this.withTransaction(async (client) => {
|
||||
const existing = await client.query("SELECT * FROM agentrun_queue_tasks WHERE id = $1 FOR UPDATE", [taskId]);
|
||||
const row = existing.rows[0];
|
||||
if (!row) throw new AgentRunError("schema-invalid", `queue task ${taskId} was not found`, { httpStatus: 404 });
|
||||
const task = queueTaskFromRow(row);
|
||||
if (isTerminalQueueTaskState(task.state)) throw new AgentRunError(task.state === "cancelled" ? "cancelled" : "schema-invalid", `queue task ${taskId} is already terminal: ${task.state}`, { httpStatus: 409 });
|
||||
const version = await this.nextQueueVersion(client);
|
||||
const at = nowIso();
|
||||
const updated = await client.query(
|
||||
"UPDATE agentrun_queue_tasks SET state = $2, latest_attempt = $3::jsonb, session_path = $4, version = $5, updated_at = $6 WHERE id = $1 RETURNING *",
|
||||
[taskId, input.state, JSON.stringify(input.latestAttempt), input.sessionPath, version, at],
|
||||
);
|
||||
return queueTaskFromRow(updated.rows[0]);
|
||||
});
|
||||
}
|
||||
|
||||
async cancelQueueTask(taskId: string, reason = "cancel requested"): Promise<QueueTaskRecord> {
|
||||
return this.withTransaction(async (client) => {
|
||||
const existing = await client.query("SELECT * FROM agentrun_queue_tasks WHERE id = $1 FOR UPDATE", [taskId]);
|
||||
|
||||
@@ -0,0 +1,135 @@
|
||||
import { AgentRunError } from "../common/errors.js";
|
||||
import type { CreateRunInput, JsonRecord, JsonValue, QueueAttemptRef, QueueDispatchResult, QueueTaskRecord, QueueTaskState, RunRecord } from "../common/types.js";
|
||||
import { asRecord } from "../common/validation.js";
|
||||
import type { AgentRunStore } from "./store.js";
|
||||
import { isTerminalQueueTaskState } from "./store.js";
|
||||
import { createKubernetesRunnerJob, type CreateRunnerJobInput, type RunnerJobDefaults } from "./kubernetes-runner-job.js";
|
||||
|
||||
export interface DispatchQueueTaskOptions {
|
||||
store: AgentRunStore;
|
||||
taskId: string;
|
||||
input: JsonRecord;
|
||||
defaults: RunnerJobDefaults;
|
||||
}
|
||||
|
||||
export async function dispatchQueueTask(options: DispatchQueueTaskOptions): Promise<QueueDispatchResult> {
|
||||
const task = await options.store.getQueueTask(options.taskId);
|
||||
assertDispatchable(task);
|
||||
|
||||
const run = await options.store.createRun(buildRunInput(task, options.input));
|
||||
const command = await options.store.createCommand(run.id, { type: "turn", payload: task.payload, idempotencyKey: `queue:${task.id}:command` });
|
||||
const runnerJob = await createKubernetesRunnerJob({
|
||||
store: options.store,
|
||||
runId: run.id,
|
||||
input: buildRunnerJobInput(task, command.id, options.input),
|
||||
defaults: options.defaults,
|
||||
});
|
||||
|
||||
const attemptId = stringFrom(runnerJob, "attemptId");
|
||||
const runnerJobs = await options.store.listRunnerJobs(run.id, command.id);
|
||||
const runnerJobRecord = runnerJobs.find((item) => item.attemptId === attemptId) ?? runnerJobs.at(-1) ?? null;
|
||||
const sessionId = run.sessionRef?.sessionId ?? null;
|
||||
const sessionPath = sessionId ? `/api/v1/sessions/${encodeURIComponent(sessionId)}` : null;
|
||||
const latestAttempt: QueueAttemptRef = {
|
||||
attemptId,
|
||||
state: "running",
|
||||
runId: run.id,
|
||||
commandId: command.id,
|
||||
runnerJobId: runnerJobRecord?.id ?? null,
|
||||
sessionId,
|
||||
sessionPath,
|
||||
};
|
||||
const updatedTask = await options.store.updateQueueTaskAttempt(task.id, { state: "running", latestAttempt, sessionPath });
|
||||
await options.store.appendEvent(run.id, "backend_status", { phase: "queue-dispatched", taskId: task.id, queue: task.queue, lane: task.lane, attemptId, runnerJobId: runnerJobRecord?.id ?? null, sessionPath });
|
||||
|
||||
return {
|
||||
action: "queue-dispatch",
|
||||
mutation: true,
|
||||
task: updatedTask,
|
||||
run,
|
||||
command,
|
||||
runnerJob,
|
||||
latestAttempt,
|
||||
pollCommands: {
|
||||
queue: `./scripts/agentrun queue show ${task.id}`,
|
||||
run: `./scripts/agentrun runs show ${run.id}`,
|
||||
command: `./scripts/agentrun commands show ${command.id} --run-id ${run.id}`,
|
||||
events: `./scripts/agentrun runs events ${run.id} --after-seq 0 --limit 100`,
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
export async function refreshQueueTaskFromCore(store: AgentRunStore, taskId: string): Promise<QueueTaskRecord> {
|
||||
const task = await store.getQueueTask(taskId);
|
||||
if (isTerminalQueueTaskState(task.state)) return task;
|
||||
const latestAttempt = task.latestAttempt;
|
||||
if (!latestAttempt?.runId || !latestAttempt.commandId) throw new AgentRunError("schema-invalid", `queue task ${taskId} has no Core attempt to refresh`, { httpStatus: 409 });
|
||||
const [run, command] = await Promise.all([store.getRun(latestAttempt.runId), store.getCommand(latestAttempt.commandId)]);
|
||||
if (command.runId !== run.id) throw new AgentRunError("schema-invalid", `queue task ${taskId} attempt references mismatched run and command`, { httpStatus: 409 });
|
||||
const state = queueStateFromRun(run);
|
||||
const sessionId = run.sessionRef?.sessionId ?? latestAttempt.sessionId ?? null;
|
||||
const sessionPath = sessionId ? `/api/v1/sessions/${encodeURIComponent(sessionId)}` : latestAttempt.sessionPath ?? null;
|
||||
const nextAttempt: QueueAttemptRef = { ...latestAttempt, state, sessionId, sessionPath };
|
||||
return store.updateQueueTaskAttempt(task.id, { state, latestAttempt: nextAttempt, sessionPath });
|
||||
}
|
||||
|
||||
function assertDispatchable(task: QueueTaskRecord): void {
|
||||
if (isTerminalQueueTaskState(task.state)) throw new AgentRunError(task.state === "cancelled" ? "cancelled" : "schema-invalid", `queue task ${task.id} is already terminal: ${task.state}`, { httpStatus: 409 });
|
||||
if (task.state !== "pending") throw new AgentRunError("schema-invalid", `queue task ${task.id} is not pending: ${task.state}`, { httpStatus: 409 });
|
||||
if (!task.workspaceRef) throw new AgentRunError("schema-invalid", "queue dispatch requires workspaceRef", { httpStatus: 400 });
|
||||
if (!task.providerId) throw new AgentRunError("schema-invalid", "queue dispatch requires providerId", { httpStatus: 400 });
|
||||
if (!task.executionPolicy) throw new AgentRunError("schema-invalid", "queue dispatch requires executionPolicy", { httpStatus: 400 });
|
||||
}
|
||||
|
||||
function queueStateFromRun(run: RunRecord): QueueTaskState {
|
||||
if (run.status === "completed") return "completed";
|
||||
if (run.status === "failed") return "failed";
|
||||
if (run.status === "blocked") return "blocked";
|
||||
if (run.status === "cancelled") return "cancelled";
|
||||
return "running";
|
||||
}
|
||||
|
||||
function buildRunInput(task: QueueTaskRecord, input: JsonRecord): CreateRunInput {
|
||||
if (!task.workspaceRef || !task.providerId || !task.executionPolicy) throw new AgentRunError("schema-invalid", "queue dispatch task is missing runtime fields", { httpStatus: 400 });
|
||||
const traceSink = input.traceSink === undefined ? { kind: "queue", taskId: task.id, queue: task.queue, lane: task.lane } : input.traceSink;
|
||||
return {
|
||||
tenantId: task.tenantId,
|
||||
projectId: task.projectId,
|
||||
workspaceRef: task.workspaceRef,
|
||||
resourceBundleRef: task.resourceBundleRef,
|
||||
providerId: task.providerId,
|
||||
backendProfile: task.backendProfile,
|
||||
executionPolicy: task.executionPolicy,
|
||||
traceSink: traceSink as JsonValue,
|
||||
};
|
||||
}
|
||||
|
||||
function buildRunnerJobInput(task: QueueTaskRecord, commandId: string, input: JsonRecord): CreateRunnerJobInput {
|
||||
const jobInput: CreateRunnerJobInput = { commandId, idempotencyKey: optionalString(input.idempotencyKey) ?? `queue:${task.id}:runner-job` };
|
||||
const copyString = (inputKey: string, outputKey: keyof CreateRunnerJobInput): void => {
|
||||
const value = optionalString(input[inputKey]);
|
||||
if (value) jobInput[outputKey] = value as never;
|
||||
};
|
||||
copyString("managerUrl", "managerUrl");
|
||||
copyString("image", "image");
|
||||
copyString("namespace", "namespace");
|
||||
copyString("attemptId", "attemptId");
|
||||
copyString("runnerId", "runnerId");
|
||||
copyString("sourceCommit", "sourceCommit");
|
||||
copyString("serviceAccountName", "serviceAccountName");
|
||||
if (input.transientEnv !== undefined) {
|
||||
if (!Array.isArray(input.transientEnv)) throw new AgentRunError("schema-invalid", "transientEnv must be an array", { httpStatus: 400 });
|
||||
jobInput.transientEnv = input.transientEnv.map((item, index) => asRecord(item, `transientEnv[${index}]`));
|
||||
}
|
||||
return jobInput;
|
||||
}
|
||||
|
||||
function stringFrom(record: JsonRecord, key: string): string {
|
||||
const value = record[key];
|
||||
if (typeof value !== "string" || value.length === 0) throw new AgentRunError("infra-failed", `runner job response missing ${key}`, { httpStatus: 502 });
|
||||
return value;
|
||||
}
|
||||
|
||||
function optionalString(value: unknown): string | undefined {
|
||||
return typeof value === "string" && value.trim().length > 0 ? value.trim() : undefined;
|
||||
}
|
||||
@@ -7,6 +7,7 @@ import { AgentRunError, errorToJson } from "../common/errors.js";
|
||||
import { asRecord, validateCreateCommand, validateCreateQueueTask, validateCreateRun, validateQueueTaskState } from "../common/validation.js";
|
||||
import type { ApiErrorBody, ApiOkBody, JsonRecord, JsonValue, RunEvent } from "../common/types.js";
|
||||
import { createKubernetesRunnerJob } from "./kubernetes-runner-job.js";
|
||||
import { dispatchQueueTask, refreshQueueTaskFromCore } from "./queue-dispatch.js";
|
||||
import { buildRunResult } from "./result.js";
|
||||
import { runnerJobStatusSummary } from "./runner-job-status.js";
|
||||
|
||||
@@ -81,6 +82,25 @@ async function route({ method, url, body, store, sourceCommit, runnerJobDefaults
|
||||
}
|
||||
const queueTaskMatch = path.match(/^\/api\/v1\/queue\/tasks\/([^/]+)$/u);
|
||||
if (method === "GET" && queueTaskMatch) return await store.getQueueTask(queueTaskMatch[1] ?? "") as unknown as JsonValue;
|
||||
const queueTaskDispatchMatch = path.match(/^\/api\/v1\/queue\/tasks\/([^/]+)\/dispatch$/u);
|
||||
if (method === "POST" && queueTaskDispatchMatch) {
|
||||
const namespace = runnerJobDefaults?.namespace ?? process.env.AGENTRUN_RUNTIME_NAMESPACE ?? "agentrun-v01";
|
||||
return await dispatchQueueTask({
|
||||
store,
|
||||
taskId: queueTaskDispatchMatch[1] ?? "",
|
||||
input: asRecord(body ?? {}, "queueDispatch"),
|
||||
defaults: {
|
||||
namespace,
|
||||
managerUrl: runnerJobDefaults?.managerUrl ?? process.env.AGENTRUN_INTERNAL_MGR_URL ?? `http://agentrun-mgr.${namespace}.svc.cluster.local:8080`,
|
||||
image: runnerJobDefaults?.image ?? process.env.AGENTRUN_RUNNER_IMAGE ?? "",
|
||||
sourceCommit,
|
||||
serviceAccountName: runnerJobDefaults?.serviceAccountName ?? process.env.AGENTRUN_RUNNER_SERVICE_ACCOUNT ?? "agentrun-v01-runner",
|
||||
...(runnerJobDefaults?.kubectlCommand ? { kubectlCommand: runnerJobDefaults.kubectlCommand } : {}),
|
||||
},
|
||||
}) as unknown as JsonValue;
|
||||
}
|
||||
const queueTaskRefreshMatch = path.match(/^\/api\/v1\/queue\/tasks\/([^/]+)\/refresh$/u);
|
||||
if (method === "POST" && queueTaskRefreshMatch) return await refreshQueueTaskFromCore(store, queueTaskRefreshMatch[1] ?? "") as unknown as JsonValue;
|
||||
const queueTaskCancelMatch = path.match(/^\/api\/v1\/queue\/tasks\/([^/]+)\/cancel$/u);
|
||||
if (method === "POST" && queueTaskCancelMatch) {
|
||||
const record = body === null ? {} : asRecord(body, "cancel");
|
||||
|
||||
+16
-1
@@ -1,4 +1,4 @@
|
||||
import type { BackendProfile, BackendTurnResult, CommandRecord, CreateCommandInput, CreateQueueTaskInput, CreateRunInput, FailureKind, JsonRecord, QueueCommanderSnapshot, QueueReadCursorRecord, QueueStats, QueueTaskListResult, QueueTaskRecord, QueueTaskState, RunEvent, RunnerJobRecord, RunnerRecord, RunRecord, SessionRecord, SessionRef, TerminalStatus } from "../common/types.js";
|
||||
import type { BackendProfile, BackendTurnResult, CommandRecord, CreateCommandInput, CreateQueueTaskInput, CreateRunInput, FailureKind, JsonRecord, QueueAttemptRef, QueueCommanderSnapshot, QueueReadCursorRecord, QueueStats, QueueTaskListResult, QueueTaskRecord, QueueTaskState, RunEvent, RunnerJobRecord, RunnerRecord, RunRecord, SessionRecord, SessionRef, TerminalStatus } from "../common/types.js";
|
||||
import { AgentRunError } from "../common/errors.js";
|
||||
import { newId, nowIso, stableHash } from "../common/validation.js";
|
||||
import { redactJson } from "../common/redaction.js";
|
||||
@@ -42,6 +42,7 @@ export interface AgentRunStore {
|
||||
createQueueTask(input: CreateQueueTaskInput): MaybePromise<QueueTaskRecord>;
|
||||
listQueueTasks(input: ListQueueTasksInput): MaybePromise<QueueTaskListResult>;
|
||||
getQueueTask(taskId: string): MaybePromise<QueueTaskRecord>;
|
||||
updateQueueTaskAttempt(taskId: string, input: UpdateQueueTaskAttemptInput): MaybePromise<QueueTaskRecord>;
|
||||
cancelQueueTask(taskId: string, reason?: string): MaybePromise<QueueTaskRecord>;
|
||||
markQueueTaskRead(taskId: string, readerId: string): MaybePromise<QueueReadCursorRecord>;
|
||||
queueStats(queue?: string): MaybePromise<QueueStats>;
|
||||
@@ -58,6 +59,12 @@ export interface ListQueueTasksInput {
|
||||
updatedAfter?: number;
|
||||
}
|
||||
|
||||
export interface UpdateQueueTaskAttemptInput {
|
||||
state: QueueTaskState;
|
||||
latestAttempt: QueueAttemptRef;
|
||||
sessionPath: string | null;
|
||||
}
|
||||
|
||||
export interface SaveRunnerJobInput {
|
||||
runId: string;
|
||||
commandId: string;
|
||||
@@ -299,6 +306,14 @@ export class MemoryAgentRunStore implements AgentRunStore {
|
||||
return task;
|
||||
}
|
||||
|
||||
updateQueueTaskAttempt(taskId: string, input: UpdateQueueTaskAttemptInput): QueueTaskRecord {
|
||||
const task = this.getQueueTask(taskId);
|
||||
if (isTerminalQueueTaskState(task.state)) throw new AgentRunError(task.state === "cancelled" ? "cancelled" : "schema-invalid", `queue task ${taskId} is already terminal: ${task.state}`, { httpStatus: 409 });
|
||||
const next: QueueTaskRecord = { ...task, state: input.state, latestAttempt: input.latestAttempt, sessionPath: input.sessionPath, version: this.nextQueueVersion(), updatedAt: nowIso() };
|
||||
this.queueTasks.set(taskId, next);
|
||||
return next;
|
||||
}
|
||||
|
||||
cancelQueueTask(taskId: string, reason = "cancel requested"): QueueTaskRecord {
|
||||
const task = this.getQueueTask(taskId);
|
||||
if (isTerminalQueueTaskState(task.state)) return task;
|
||||
|
||||
Reference in New Issue
Block a user