feat: 补齐 HWLAB 手动调度能力

This commit is contained in:
Codex
2026-06-01 11:40:08 +08:00
parent eb5e0f57a6
commit 62846f6369
25 changed files with 1159 additions and 70 deletions
+178 -6
View File
@@ -1,4 +1,4 @@
import type { BackendProfile, BackendTurnResult, CommandRecord, CreateCommandInput, CreateRunInput, FailureKind, JsonRecord, RunEvent, RunnerRecord, RunRecord, TerminalStatus } from "../common/types.js";
import type { BackendProfile, BackendTurnResult, CommandRecord, CreateCommandInput, CreateRunInput, FailureKind, JsonRecord, 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";
@@ -26,16 +26,38 @@ export interface AgentRunStore {
getCommand(commandId: string): MaybePromise<CommandRecord>;
listCommands(runId: string, afterSeq: number, limit: number): MaybePromise<CommandRecord[]>;
registerRunner(input: Partial<RunnerRecord>): MaybePromise<RunnerRecord>;
listRunnerJobs(runId: string, commandId?: string): MaybePromise<RunnerJobRecord[]>;
getRunnerJobByIdempotencyKey(runId: string, idempotencyKey: string, payloadHash: string): MaybePromise<RunnerJobRecord | null>;
saveRunnerJob(input: SaveRunnerJobInput): MaybePromise<RunnerJobRecord>;
claimRun(runId: string, runnerId: string, leaseMs: number): MaybePromise<RunRecord>;
heartbeat(runId: string, runnerId: string, leaseMs: number): MaybePromise<RunRecord>;
ackCommand(commandId: string): MaybePromise<CommandRecord>;
finishCommand(commandId: string, result: Pick<BackendTurnResult, "terminalStatus" | "failureKind" | "failureMessage">): MaybePromise<CommandRecord>;
appendEvent(runId: string, type: RunEvent["type"], payload: JsonRecord): MaybePromise<RunEvent>;
finishRun(runId: string, result: Pick<BackendTurnResult, "terminalStatus" | "failureKind" | "failureMessage">): MaybePromise<RunRecord>;
finishRun(runId: string, result: Pick<BackendTurnResult, "terminalStatus" | "failureKind" | "failureMessage" | "threadId" | "turnId">): MaybePromise<RunRecord>;
cancelRun(runId: string, reason?: string): MaybePromise<RunRecord>;
cancelCommand(commandId: string, reason?: string): MaybePromise<CommandRecord>;
getSession(sessionId: string): MaybePromise<SessionRecord | null>;
backends(): MaybePromise<JsonRecord[]>;
close?(): MaybePromise<void>;
}
export interface SaveRunnerJobInput {
runId: string;
commandId: string;
idempotencyKey?: string | null;
payloadHash: string;
attemptId: string;
runnerId: string;
namespace: string;
jobName: string;
managerUrl: string;
image: string;
sourceCommit: string;
serviceAccountName?: string | null;
result: JsonRecord;
}
export async function openAgentRunStoreFromEnv(env: NodeJS.ProcessEnv = process.env): Promise<AgentRunStore> {
const databaseUrl = env.DATABASE_URL?.trim();
if (databaseUrl) {
@@ -52,6 +74,8 @@ export class MemoryAgentRunStore implements AgentRunStore {
private readonly commands = new Map<string, CommandRecord>();
private readonly eventsByRun = new Map<string, RunEvent[]>();
private readonly runners = new Map<string, RunnerRecord>();
private readonly sessions = new Map<string, SessionRecord>();
private readonly runnerJobs = new Map<string, RunnerJobRecord>();
health(): StoreHealth {
return { adapter: "memory-self-test", ready: true, reachable: true, migrationReady: true, migrationId: "memory-self-test", failureKind: null, message: null, credentialValuesPrinted: false };
@@ -59,10 +83,11 @@ export class MemoryAgentRunStore implements AgentRunStore {
createRun(input: CreateRunInput): RunRecord {
const at = nowIso();
const run: RunRecord = { ...input, id: newId("run"), status: "pending", terminalStatus: null, failureKind: null, failureMessage: null, createdAt: at, updatedAt: at, claimedBy: null, leaseExpiresAt: null };
const sessionRef = this.resolveSessionForRun(input, at);
const run: RunRecord = { ...input, sessionRef, resourceBundleRef: input.resourceBundleRef ?? null, id: newId("run"), status: "pending", terminalStatus: null, failureKind: null, failureMessage: null, createdAt: at, updatedAt: at, claimedBy: null, leaseExpiresAt: null };
this.runs.set(run.id, run);
this.eventsByRun.set(run.id, []);
this.appendEvent(run.id, "backend_status", { phase: "run-created", backendProfile: run.backendProfile });
this.appendEvent(run.id, "backend_status", { phase: "run-created", backendProfile: run.backendProfile, sessionRef: summarizeSessionRef(run.sessionRef ?? null), resourceBundleRef: summarizeResourceBundleRef(run.resourceBundleRef ?? null) });
return run;
}
@@ -113,9 +138,35 @@ export class MemoryAgentRunStore implements AgentRunStore {
return runner;
}
listRunnerJobs(runId: string, commandId?: string): RunnerJobRecord[] {
this.getRun(runId);
return Array.from(this.runnerJobs.values())
.filter((job) => job.runId === runId && (!commandId || job.commandId === commandId))
.sort((a, b) => a.createdAt.localeCompare(b.createdAt));
}
getRunnerJobByIdempotencyKey(runId: string, idempotencyKey: string, payloadHash: string): RunnerJobRecord | null {
const existing = Array.from(this.runnerJobs.values()).find((job) => job.runId === runId && job.idempotencyKey === idempotencyKey);
if (!existing) return null;
if (existing.payloadHash !== payloadHash) throw new AgentRunError("schema-invalid", "runner job idempotency key reused with different payload", { httpStatus: 409 });
return existing;
}
saveRunnerJob(input: SaveRunnerJobInput): RunnerJobRecord {
if (input.idempotencyKey) {
const existing = this.getRunnerJobByIdempotencyKey(input.runId, input.idempotencyKey, input.payloadHash);
if (existing) return existing;
}
const at = nowIso();
const record: RunnerJobRecord = { ...input, id: newId("rjob"), idempotencyKey: input.idempotencyKey ?? null, serviceAccountName: input.serviceAccountName ?? null, createdAt: at, updatedAt: at };
this.runnerJobs.set(record.id, record);
return record;
}
claimRun(runId: string, runnerId: string, leaseMs: number): RunRecord {
const run = this.getRun(runId);
if (run.claimedBy && run.claimedBy !== runnerId && run.status !== "completed" && run.status !== "failed" && run.status !== "cancelled") throw new AgentRunError("runner-lease-conflict", `run ${runId} is already claimed`, { httpStatus: 409 });
if (isTerminalRunStatus(run.status)) throw new AgentRunError(run.failureKind ?? (run.status === "cancelled" ? "cancelled" : "schema-invalid"), `run ${runId} is already terminal: ${run.status}`, { httpStatus: 409 });
if (run.claimedBy && run.claimedBy !== runnerId) throw new AgentRunError("runner-lease-conflict", `run ${runId} is already claimed`, { httpStatus: 409 });
const next = this.updateRun(runId, { status: "claimed", claimedBy: runnerId, leaseExpiresAt: new Date(Date.now() + leaseMs).toISOString() });
this.appendEvent(runId, "backend_status", { phase: "run-claimed", runnerId });
return next;
@@ -123,12 +174,14 @@ export class MemoryAgentRunStore implements AgentRunStore {
heartbeat(runId: string, runnerId: string, leaseMs: number): RunRecord {
const run = this.getRun(runId);
if (isTerminalRunStatus(run.status)) return run;
if (run.claimedBy !== runnerId) throw new AgentRunError("runner-lease-conflict", `run ${runId} is not claimed by ${runnerId}`, { httpStatus: 409 });
return this.updateRun(runId, { leaseExpiresAt: new Date(Date.now() + leaseMs).toISOString() });
}
ackCommand(commandId: string): CommandRecord {
const command = this.getCommand(commandId);
if (isTerminalCommandState(command.state) || command.state === "acknowledged") return command;
const next = { ...command, state: "acknowledged" as const, acknowledgedAt: nowIso(), updatedAt: nowIso() };
this.commands.set(commandId, next);
return next;
@@ -136,6 +189,7 @@ export class MemoryAgentRunStore implements AgentRunStore {
finishCommand(commandId: string, result: Pick<BackendTurnResult, "terminalStatus" | "failureKind" | "failureMessage">): CommandRecord {
const command = this.getCommand(commandId);
if (isTerminalCommandState(command.state)) return command;
const next = { ...command, state: commandStateFromTerminal(result.terminalStatus), updatedAt: nowIso() };
this.commands.set(commandId, next);
this.appendEvent(command.runId, "backend_status", { phase: "command-terminal", commandId, state: next.state, terminalStatus: result.terminalStatus, failureKind: result.failureKind });
@@ -151,13 +205,45 @@ export class MemoryAgentRunStore implements AgentRunStore {
return event;
}
finishRun(runId: string, result: Pick<BackendTurnResult, "terminalStatus" | "failureKind" | "failureMessage">): RunRecord {
finishRun(runId: string, result: Pick<BackendTurnResult, "terminalStatus" | "failureKind" | "failureMessage" | "threadId" | "turnId">): RunRecord {
const existing = this.getRun(runId);
if (isTerminalRunStatus(existing.status)) return existing;
const status = statusFromTerminal(result.terminalStatus);
const next = this.updateRun(runId, { status, terminalStatus: result.terminalStatus, failureKind: result.failureKind, failureMessage: result.failureMessage });
if (result.threadId && next.sessionRef?.sessionId) this.upsertSessionThread(next, result.threadId, result.turnId ?? null);
this.appendEvent(runId, "terminal_status", { terminalStatus: result.terminalStatus, failureKind: result.failureKind, message: result.failureMessage });
return next;
}
cancelRun(runId: string, reason = "cancel requested"): RunRecord {
const run = this.getRun(runId);
if (isTerminalRunStatus(run.status)) return run;
const at = nowIso();
for (const command of Array.from(this.commands.values()).filter((item) => item.runId === runId && !isTerminalCommandState(item.state))) {
const cancelled = { ...command, state: "cancelled" as const, updatedAt: at };
this.commands.set(command.id, cancelled);
this.appendEvent(runId, "backend_status", { phase: "command-terminal", commandId: command.id, state: "cancelled", terminalStatus: "cancelled", failureKind: "cancelled" });
}
this.appendEvent(runId, "backend_status", { phase: "cancel-requested", reason });
const next = this.updateRun(runId, { status: "cancelled", terminalStatus: "cancelled", failureKind: "cancelled", failureMessage: reason });
this.appendEvent(runId, "terminal_status", { terminalStatus: "cancelled", failureKind: "cancelled", message: reason });
return next;
}
cancelCommand(commandId: string, reason = "cancel requested"): CommandRecord {
const command = this.getCommand(commandId);
if (isTerminalCommandState(command.state)) return command;
const next = { ...command, state: "cancelled" as const, updatedAt: nowIso() };
this.commands.set(commandId, next);
this.appendEvent(command.runId, "backend_status", { phase: "command-terminal", commandId, state: "cancelled", terminalStatus: "cancelled", failureKind: "cancelled" });
this.cancelRun(command.runId, reason);
return next;
}
getSession(sessionId: string): SessionRecord | null {
return this.sessions.get(sessionId) ?? null;
}
backends(): JsonRecord[] {
return backendCapabilities();
}
@@ -168,6 +254,48 @@ export class MemoryAgentRunStore implements AgentRunStore {
this.runs.set(runId, next);
return next;
}
private resolveSessionForRun(input: CreateRunInput, at: string): SessionRef | null {
if (!input.sessionRef) return null;
const existing = this.sessions.get(input.sessionRef.sessionId);
if (existing) return sessionRefFromRecord(existing, input.sessionRef);
const record: SessionRecord = {
sessionId: input.sessionRef.sessionId,
tenantId: input.tenantId,
projectId: input.projectId,
backendProfile: input.backendProfile,
conversationId: input.sessionRef.conversationId ?? null,
threadId: input.sessionRef.threadId ?? null,
metadata: input.sessionRef.metadata ?? {},
createdAt: at,
updatedAt: at,
expiresAt: input.sessionRef.expiresAt ?? null,
};
this.sessions.set(record.sessionId, record);
return sessionRefFromRecord(record, input.sessionRef);
}
private upsertSessionThread(run: RunRecord, threadId: string, turnId: string | null): void {
if (!run.sessionRef?.sessionId) return;
const at = nowIso();
const existing = this.sessions.get(run.sessionRef.sessionId);
const record: SessionRecord = {
sessionId: run.sessionRef.sessionId,
tenantId: run.tenantId,
projectId: run.projectId,
backendProfile: run.backendProfile,
conversationId: run.sessionRef.conversationId ?? existing?.conversationId ?? null,
threadId,
metadata: { ...(existing?.metadata ?? {}), ...(run.sessionRef.metadata ?? {}), ...(turnId ? { lastTurnId: turnId } : {}) },
createdAt: existing?.createdAt ?? at,
updatedAt: at,
expiresAt: run.sessionRef.expiresAt ?? existing?.expiresAt ?? null,
};
this.sessions.set(record.sessionId, record);
const nextSessionRef = sessionRefFromRecord(record, run.sessionRef);
this.updateRun(run.id, { sessionRef: nextSessionRef });
this.appendEvent(run.id, "backend_status", { phase: "session-updated", sessionRef: summarizeSessionRef(nextSessionRef), turnId });
}
}
export function statusFromTerminal(terminalStatus: TerminalStatus): RunRecord["status"] {
@@ -182,3 +310,47 @@ export function commandStateFromTerminal(terminalStatus: TerminalStatus): Comman
if (terminalStatus === "cancelled") return "cancelled";
return "failed";
}
export function isTerminalRunStatus(status: RunRecord["status"]): boolean {
return status === "completed" || status === "failed" || status === "blocked" || status === "cancelled";
}
export function isTerminalCommandState(state: CommandRecord["state"]): boolean {
return state === "completed" || state === "failed" || state === "cancelled";
}
export function sessionRefFromRecord(record: SessionRecord, fallback: SessionRef): SessionRef {
return {
sessionId: record.sessionId,
...(record.conversationId ? { conversationId: record.conversationId } : fallback.conversationId ? { conversationId: fallback.conversationId } : {}),
...(record.threadId ? { threadId: record.threadId } : fallback.threadId ? { threadId: fallback.threadId } : {}),
...(record.expiresAt ? { expiresAt: record.expiresAt } : fallback.expiresAt ? { expiresAt: fallback.expiresAt } : {}),
...(Object.keys(record.metadata).length > 0 ? { metadata: record.metadata } : fallback.metadata ? { metadata: fallback.metadata } : {}),
};
}
export function summarizeSessionRef(sessionRef: SessionRef | null): JsonRecord | null {
if (!sessionRef) return null;
return {
sessionId: sessionRef.sessionId,
conversationId: sessionRef.conversationId ?? null,
threadId: sessionRef.threadId ?? null,
expiresAt: sessionRef.expiresAt ?? null,
metadataKeys: Object.keys(sessionRef.metadata ?? {}).sort(),
valuesPrinted: false,
};
}
export function summarizeResourceBundleRef(resourceBundleRef: RunRecord["resourceBundleRef"] | null | undefined): JsonRecord | null {
if (!resourceBundleRef) return null;
return {
kind: resourceBundleRef.kind,
repoUrl: resourceBundleRef.repoUrl,
commitId: resourceBundleRef.commitId,
subdir: resourceBundleRef.subdir ?? null,
sparsePathCount: resourceBundleRef.sparsePaths?.length ?? 0,
submodules: resourceBundleRef.submodules ?? false,
lfs: resourceBundleRef.lfs ?? false,
credentialRef: resourceBundleRef.credentialRef ? { name: resourceBundleRef.credentialRef.name, namespace: resourceBundleRef.credentialRef.namespace ?? null, keys: resourceBundleRef.credentialRef.keys ?? [], valuesPrinted: false } : null,
};
}