bugfix, move reconcile to job scheduler

This commit is contained in:
Alex Holliday committed 2026-06-25 11:47:22 -07:00
1 parent b21ef60cda
commit ef353dcb36
9 files changed
+88 -71

No files matched your search

@@ -10,6 +10,7 @@ const QueueWorkerSchema = new Schema<QueueWorkerDocument>(
{
_id: { type: String, required: true },
mode: { type: String, enum: QueueModes, required: true },
processesJobs: { type: Boolean, required: true },
lastSeenAt: { type: Date, default: Date.now, expires: WORKER_TTL_SECONDS },
},
{ timestamps: true }
@@ -2,7 +2,7 @@ import type { QueueMode } from "@/domain/app-settings/app-settings.type.js";
import type { QueueWorker } from "@/domain/queue-workers/queue-worker.type.js";
export interface IQueueWorkersRepository {
upsert(workerId: string, mode: QueueMode): Promise<void>;
upsert(workerId: string, mode: QueueMode, processesJobs: boolean): Promise<void>;
findRecent(maxAgeMs: number): Promise<QueueWorker[]>;
deleteById(workerId: string): Promise<void>;
}
@@ -11,12 +11,13 @@ class MongoQueueWorkersRepository implements IQueueWorkersRepository {
return {
workerId: doc._id,
mode: doc.mode,
processesJobs: doc.processesJobs,
lastSeenAt: doc.lastSeenAt.getTime(),
};
};
upsert = async (workerId: string, mode: QueueMode): Promise<void> => {
await QueueWorkerModel.updateOne({ _id: workerId }, { $set: { mode, lastSeenAt: new Date() } }, { upsert: true });
upsert = async (workerId: string, mode: QueueMode, processesJobs: boolean): Promise<void> => {
await QueueWorkerModel.updateOne({ _id: workerId }, { $set: { mode, processesJobs, lastSeenAt: new Date() } }, { upsert: true });
};
findRecent = async (maxAgeMs: number): Promise<QueueWorker[]> => {
@@ -1,11 +1,9 @@
import { type QueueMode } from "@/domain/app-settings/app-settings.type.js";
// Heartbeat presence record for a scheduler instance. _id is the scheduler's
// workerId (hostname:pid:uuid). The TTL index on lastSeenAt garbage-collects
// records for instances that have stopped heartbeating.
export interface QueueWorkerDocument {
_id: string;
mode: QueueMode;
processesJobs: boolean;
lastSeenAt: Date;
createdAt: Date;
updatedAt: Date;
@@ -14,5 +12,6 @@ export interface QueueWorkerDocument {
export type QueueWorker = {
workerId: string; // hostname:pid:uuid
mode: QueueMode;
processesJobs: boolean;
lastSeenAt: number; // epoch ms of the last heartbeat
};
+2 -1
View File
@@ -75,7 +75,8 @@ const startApp = async () => {
// Primary node does not process jobs, only need scheduler
// ***********************
else {
scheduler = new JobScheduler(shared.jobsRepository, shared.queueWorkersRepository, shared.workerId);
scheduler = new JobScheduler(shared.jobsRepository, shared.queueWorkersRepository, shared.monitorsRepository, queueMode, logger, shared.workerId);
await scheduler.init(); // register in the queue_workers registry + reconcile (seed the jobs collection)
}
const services = buildApi(shared, scheduler);
+16 -61
View File
@@ -1,6 +1,5 @@
import { supportsGeoCheck } from "@/domain/monitors/monitor.types.js";
import { IQueueWorker } from "@/worker/worker.interface.js";
import { type Job, type JobSeed, type JobType, jobId, LOCK_MS } from "@/domain/jobs/job.type.js";
import { type Job, type JobType, LOCK_MS } from "@/domain/jobs/job.type.js";
import { ILogger } from "@/utils/logger.js";
import { IJobsRepository } from "@/domain/jobs/job.repository.interface.js";
import { IMonitorsRepository } from "@/domain/monitors/monitor.repository.interface.js";
@@ -20,9 +19,7 @@ const SERVICE_NAME = "JobQueue";
const POLL_MS = 250; // base poll interval while a loop is actively claiming work
const POLL_MAX_MS = 5000; // back off poll interval to 5s if no jobs
const WORKER_STALE_MS = WORKER_TTL_SECONDS * 1000; // a worker counts as alive if seen within this window
const HEARTBEAT_MS = WORKER_STALE_MS / 3; // Worker can miss two beats without being considered stale
const LOCK_RENEW_MS = LOCK_MS / 3; // renew an in-flight job's lock at 1/3 the lease, so it survives two missed renewals
const CLEANUP_INTERVAL_MS = 24 * 60 * 60 * 1000;
const DRAIN_TIMEOUT_MS = 25_000;
const DRAIN_POLL_MS = 200;
@@ -48,11 +45,13 @@ export class DBQueueWorker extends JobScheduler implements IQueueWorker {
"cleanup-retention": 0,
};
private processesJobs: boolean = false;
constructor(
private logger: ILogger,
logger: ILogger, // forwarded to super
private isDbConnected: () => boolean,
jobsRepository: IJobsRepository, // ← no modifier: forwarded to super
private monitorsRepository: IMonitorsRepository,
jobsRepository: IJobsRepository, // forwarded
monitorsRepository: IMonitorsRepository, // forwarded
private checksRepository: IChecksRepository,
private checkService: ICheckService,
private bufferService: IBufferService,
@@ -61,12 +60,13 @@ export class DBQueueWorker extends JobScheduler implements IQueueWorker {
private geoCheckPipeline: ICheckPipeline,
private dispatcher: IReactorDispatcher,
private helper: IWorkerHelper,
queueWorkersRepository: IQueueWorkersRepository, // ← no modifier: forwarded
private queueMode: QueueMode,
private queuePrimaryProcesses: boolean,
queueWorkersRepository: IQueueWorkersRepository, // forwarded
queueMode: QueueMode, // forwarded
queuePrimaryProcesses: boolean, // consumed immediately
workerId: string
) {
super(jobsRepository, queueWorkersRepository, workerId);
super(jobsRepository, queueWorkersRepository, monitorsRepository, queueMode, logger, workerId);
this.processesJobs = queuePrimaryProcesses === true || queueMode === "worker";
}
get serviceName() {
@@ -115,15 +115,6 @@ export class DBQueueWorker extends JobScheduler implements IQueueWorker {
private getInFlightCount = () => Object.values(this.inFlight).reduce((sum, n) => sum + n, 0);
private toCleanupJob = (type: "cleanup-orphaned" | "cleanup-retention", now: number): JobSeed => ({
id: jobId(type, null),
type,
refId: null,
isActive: true,
nextScheduledAt: now,
intervalMs: CLEANUP_INTERVAL_MS,
});
// ********************
// Stage 1: This can eventually be offloaded to a separate service
// ********************
@@ -253,52 +244,16 @@ export class DBQueueWorker extends JobScheduler implements IQueueWorker {
// Wake a worker loop after adding a job that's due immediately.
// Otherwise the loop could be delayed by POLL_MAX_MS if it is already fully backed off
// Seed queue
private reconcile = async () => {
const now = Date.now();
const monitors = (await this.monitorsRepository.findAll()) ?? [];
for (const monitor of monitors) {
await this.jobsRepository.upsertJob(this.toCheckJob(monitor, now));
if (supportsGeoCheck(monitor.type) && monitor.geoCheckEnabled) {
await this.jobsRepository.upsertJob(this.toGeoCheckJob(monitor, now));
}
}
// Both cleanup jobs run immediately on every startup, then once a day
await this.jobsRepository.upsertCleanupJob(this.toCleanupJob("cleanup-orphaned", now));
await this.jobsRepository.upsertCleanupJob(this.toCleanupJob("cleanup-retention", now));
};
// Register worker
private heartbeat = async () => {
try {
await this.queueWorkersRepository.upsert(this.workerId, this.queueMode);
} catch (error: unknown) {
this.logger.warn({
message: error instanceof Error ? error.message : String(error),
service: SERVICE_NAME,
method: "heartbeat",
});
}
};
init = async () => {
this.stopped = false;
await this.heartbeat();
this.timers.set("heartbeat", setInterval(this.heartbeat, HEARTBEAT_MS));
if (this.queueMode === "primary") {
await this.reconcile();
}
if (this.queuePrimaryProcesses === true || this.queueMode === "worker") {
override async init() {
await super.init(); // heartbeat + (primary?) reconcile
if (this.processesJobs) {
for (const type of Object.keys(this.inFlight) as JobType[]) {
this.startLoop(type);
}
}
this.initComplete = true;
return true;
};
}
override flushQueues = async () => {
await this.shutdown();
@@ -354,6 +309,6 @@ export class DBQueueWorker extends JobScheduler implements IQueueWorker {
countAliveWorkers = async () => {
const workers = await this.queueWorkersRepository.findRecent(WORKER_STALE_MS);
return workers.length;
return workers.filter((worker) => worker.processesJobs).length;
};
}
+1 -1
View File
@@ -45,7 +45,7 @@ export class HealthServer implements IHealthServer {
return (
gauge("checkmate_worker_jobs_in_flight", "Jobs currently being processed by this worker", health.inFlight) +
gauge("checkmate_worker_due_backlog", "Due check jobs not currently locked (cluster-wide)", dueBacklog) +
gauge("checkmate_worker_alive_total", "Queue members seen within the TTL window (cluster-wide)", aliveWorkers) +
gauge("checkmate_worker_alive_total", "Job-processing nodes seen within the TTL window (cluster-wide)", aliveWorkers) +
gauge("checkmate_worker_draining", "1 if this worker is draining, else 0", health.draining ? 1 : 0)
);
};
+1 -1
View File
@@ -77,13 +77,13 @@ export interface IJobScheduler {
getMetrics(): Promise<WorkerMetrics>;
getJobs(pagination: WorkerJobsPagination): Promise<WorkerJobsPage>;
flushQueues(): Promise<{ success: boolean }>;
init(): Promise<boolean>;
drain(): Promise<void>;
shutdown(): Promise<void>;
}
export interface IQueueWorker extends IJobScheduler {
readonly serviceName: string;
init(): Promise<boolean>;
getHealth(): WorkerHealth;
countDueBacklog(): Promise<number>;
countAliveWorkers(): Promise<number>;
+61 -1
View File
@@ -1,20 +1,28 @@
import { IJobScheduler } from "@/worker/worker.interface.js";
import { IJobsRepository } from "@/domain/jobs/job.repository.interface.js";
import { IMonitorsRepository } from "@/domain/monitors/monitor.repository.interface.js";
import { Monitor, supportsGeoCheck } from "@/domain/monitors/monitor.types.js";
import { type Job, type JobSeed, type JobType, jobId } from "@/domain/jobs/job.type.js";
import { IQueueWorkersRepository } from "@/domain/queue-workers/queue-worker.repository.interface.js";
import { WorkerJobsPagination, WorkerJobSummary, WorkerMetrics } from "@/worker/worker.interface.js";
import { WORKER_TTL_SECONDS } from "@/domain/queue-workers/queue-worker.model.js";
import { QueueMode } from "@/domain/app-settings/app-settings.type.js";
import { ILogger } from "@/utils/logger.js";
const POLL_MS = 250; // base poll interval while a loop is actively claiming work
const SERVICE_NAME = "JobScheduler";
const WORKER_STALE_MS = WORKER_TTL_SECONDS * 1000; // a worker counts as alive if seen within this window
const HEARTBEAT_MS = WORKER_STALE_MS / 3; // Worker can miss two beats without being considered stale
const CLEANUP_INTERVAL_MS = 24 * 60 * 60 * 1000;
export class JobScheduler implements IJobScheduler {
static SERVICE_NAME = SERVICE_NAME;
constructor(
protected jobsRepository: IJobsRepository,
protected queueWorkersRepository: IQueueWorkersRepository,
protected monitorsRepository: IMonitorsRepository,
protected queueMode: QueueMode,
protected logger: ILogger,
protected readonly workerId: string // same id MongoJobsRepository stamps into lockedBy
) {}
@@ -22,6 +30,7 @@ export class JobScheduler implements IJobScheduler {
return JobScheduler.SERVICE_NAME;
}
protected processesJobs = false; // scheduler-only by default; DBQueueWorker flips this on when it runs job loops
protected stopped = false;
protected draining = false;
protected ticking: Partial<Record<JobType, boolean>> = {}; // Whether a loop is mid tick
@@ -53,6 +62,15 @@ export class JobScheduler implements IJobScheduler {
intervalMs: monitor.geoCheckInterval!,
});
protected toCleanupJob = (type: "cleanup-orphaned" | "cleanup-retention", now: number): JobSeed => ({
id: jobId(type, null),
type,
refId: null,
isActive: true,
nextScheduledAt: now,
intervalMs: CLEANUP_INTERVAL_MS,
});
private toSummary = (job: Job): WorkerJobSummary => ({
monitorId: job.refId ?? job.id,
monitorType: job.type,
@@ -112,7 +130,8 @@ export class JobScheduler implements IJobScheduler {
getMetrics = async (): Promise<WorkerMetrics> => {
const rows = await this.jobsRepository.findAll();
const workers = await this.queueWorkersRepository.findRecent(WORKER_STALE_MS);
const recent = await this.queueWorkersRepository.findRecent(WORKER_STALE_MS);
const workers = recent.filter((worker) => worker.processesJobs);
const now = Date.now();
return rows.reduce<WorkerMetrics>(
(acc, job) => {
@@ -155,6 +174,47 @@ export class JobScheduler implements IJobScheduler {
return { success: true };
};
// init is the scheduler's startup: register in the registry + (if primary) seed the queue.
// Subclasses override to ALSO start processing, calling super.init() first.
// Declared as a prototype method (not an arrow field) so subclasses can reach it via super.init().
async init(): Promise<boolean> {
this.stopped = false;
await this.heartbeat();
this.timers.set("heartbeat", setInterval(this.heartbeat, HEARTBEAT_MS));
if (this.queueMode === "primary") {
await this.reconcile();
}
return true;
}
// Seed queue
protected reconcile = async () => {
const now = Date.now();
const monitors = (await this.monitorsRepository.findAll()) ?? [];
for (const monitor of monitors) {
await this.jobsRepository.upsertJob(this.toCheckJob(monitor, now));
if (supportsGeoCheck(monitor.type) && monitor.geoCheckEnabled) {
await this.jobsRepository.upsertJob(this.toGeoCheckJob(monitor, now));
}
}
// Both cleanup jobs run immediately on every startup, then once a day
await this.jobsRepository.upsertCleanupJob(this.toCleanupJob("cleanup-orphaned", now));
await this.jobsRepository.upsertCleanupJob(this.toCleanupJob("cleanup-retention", now));
};
// Register worker
protected heartbeat = async () => {
try {
await this.queueWorkersRepository.upsert(this.workerId, this.queueMode, this.processesJobs);
} catch (error: unknown) {
this.logger.warn({
message: error instanceof Error ? error.message : String(error),
service: SERVICE_NAME,
method: "heartbeat",
});
}
};
drain = async () => {
this.stopped = true;
this.draining = true;