1 import { ensureDir, readdir, remove } from 'fs-extra'
2 import { join } from 'path'
3 import { io, Socket } from 'socket.io-client'
4 import { pick } from '@shared/core-utils'
5 import { PeerTubeProblemDocument, ServerErrorCode } from '@shared/models'
6 import { PeerTubeServer as PeerTubeServerCommand } from '@shared/server-commands'
7 import { ConfigManager } from '../shared'
8 import { IPCServer } from '../shared/ipc'
9 import { logger } from '../shared/logger'
10 import { JobWithToken, processJob } from './process'
12 type PeerTubeServer = PeerTubeServerCommand & {
15 runnerDescription?: string
18 export class RunnerServer {
19 private static instance: RunnerServer
21 private servers: PeerTubeServer[] = []
22 private processingJobs: { job: JobWithToken, server: PeerTubeServer }[] = []
24 private checkingAvailableJobs = false
26 private cleaningUp = false
28 private readonly sockets = new Map<PeerTubeServer, Socket>()
30 private constructor () {}
33 logger.info('Running PeerTube runner in server mode')
35 await ConfigManager.Instance.load()
37 for (const registered of ConfigManager.Instance.getConfig().registeredInstances) {
38 const serverCommand = new PeerTubeServerCommand({ url: registered.url })
40 this.loadServer(Object.assign(serverCommand, registered))
42 logger.info(`Loading registered instance ${registered.url}`)
46 const ipcServer = new IPCServer()
48 await ipcServer.run(this)
50 logger.error('Cannot start local socket for IPC communication', err)
55 for (const code of [ 'SIGINT', 'SIGUSR1', 'SIGUSR2', 'uncaughtException' ]) {
56 process.on(code, async (err, origin) => {
57 if (code === 'uncaughtException') {
58 logger.error({ err, origin }, 'uncaughtException')
66 await ensureDir(ConfigManager.Instance.getTranscodingDirectory())
67 await this.cleanupTMP()
69 logger.info(`Using ${ConfigManager.Instance.getTranscodingDirectory()} for transcoding directory`)
71 await this.checkAvailableJobs()
74 // ---------------------------------------------------------------------------
76 async registerRunner (options: {
78 registrationToken: string
80 runnerDescription?: string
82 const { url, registrationToken, runnerName, runnerDescription } = options
84 logger.info(`Registering runner ${runnerName} on ${url}...`)
86 const serverCommand = new PeerTubeServerCommand({ url })
87 const { runnerToken } = await serverCommand.runners.register({ name: runnerName, description: runnerDescription, registrationToken })
89 const server: PeerTubeServer = Object.assign(serverCommand, {
95 this.loadServer(server)
96 await this.saveRegisteredInstancesInConf()
98 logger.info(`Registered runner ${runnerName} on ${url}`)
100 await this.checkAvailableJobs()
103 private loadServer (server: PeerTubeServer) {
104 this.servers.push(server)
106 const url = server.url + '/runners'
107 const socket = io(url, {
109 runnerToken: server.runnerToken
111 transports: [ 'websocket' ]
114 socket.on('connect_error', err => logger.warn({ err }, `Cannot connect to ${url} socket`))
115 socket.on('connect', () => logger.info(`Connected to ${url} socket`))
116 socket.on('available-jobs', () => this.checkAvailableJobs())
118 this.sockets.set(server, socket)
121 async unregisterRunner (options: {
124 const { url } = options
126 const server = this.servers.find(s => s.url === url)
128 logger.error(`Unknown server ${url} to unregister`)
132 logger.info(`Unregistering runner ${server.runnerName} on ${url}...`)
135 await server.runners.unregister({ runnerToken: server.runnerToken })
137 logger.error({ err }, `Cannot unregister runner ${server.runnerName} on ${url}`)
140 this.unloadServer(server)
141 await this.saveRegisteredInstancesInConf()
143 logger.info(`Unregistered runner ${server.runnerName} on ${server.url}`)
146 private unloadServer (server: PeerTubeServer) {
147 this.servers = this.servers.filter(s => s !== server)
149 const socket = this.sockets.get(server)
152 this.sockets.delete(server)
157 servers: this.servers.map(s => {
160 runnerName: s.runnerName,
161 runnerDescription: s.runnerDescription
167 // ---------------------------------------------------------------------------
169 private async checkAvailableJobs () {
170 if (this.checkingAvailableJobs) return
172 logger.info('Checking available jobs')
174 this.checkingAvailableJobs = true
176 for (const server of this.servers) {
178 const job = await this.requestJob(server)
181 await this.tryToExecuteJobAsync(server, job)
183 if ((err.res?.body as PeerTubeProblemDocument)?.code === ServerErrorCode.UNKNOWN_RUNNER_TOKEN) {
184 logger.error({ err }, `Unregistering ${server.url} as the runner token ${server.runnerToken} is invalid`)
186 await this.unregisterRunner({ url: server.url })
190 logger.error({ err }, `Cannot request/accept job on ${server.url} for runner ${server.runnerName}`)
194 this.checkingAvailableJobs = false
197 private async requestJob (server: PeerTubeServer) {
198 logger.debug(`Requesting jobs on ${server.url} for runner ${server.runnerName}`)
200 const { availableJobs } = await server.runnerJobs.request({ runnerToken: server.runnerToken })
202 if (availableJobs.length === 0) {
203 logger.debug(`No job available on ${server.url} for runner ${server.runnerName}`)
207 return availableJobs[0]
210 private async tryToExecuteJobAsync (server: PeerTubeServer, jobToAccept: { uuid: string }) {
211 if (this.processingJobs.length >= ConfigManager.Instance.getConfig().jobs.concurrency) return
213 const { job } = await server.runnerJobs.accept({ runnerToken: server.runnerToken, jobUUID: jobToAccept.uuid })
215 const processingJob = { job, server }
216 this.processingJobs.push(processingJob)
218 processJob({ server, job, runnerToken: server.runnerToken })
220 logger.error({ err }, 'Cannot process job')
222 server.runnerJobs.error({ jobToken: job.jobToken, jobUUID: job.uuid, runnerToken: server.runnerToken, message: err.message })
223 .catch(err2 => logger.error({ err: err2 }, 'Cannot abort job after error'))
226 this.processingJobs = this.processingJobs.filter(p => p !== processingJob)
228 return this.checkAvailableJobs()
232 // ---------------------------------------------------------------------------
234 private saveRegisteredInstancesInConf () {
235 const data = this.servers.map(s => {
236 return pick(s, [ 'url', 'runnerToken', 'runnerName', 'runnerDescription' ])
239 return ConfigManager.Instance.setRegisteredInstances(data)
242 // ---------------------------------------------------------------------------
244 private async cleanupTMP () {
245 const files = await readdir(ConfigManager.Instance.getTranscodingDirectory())
247 for (const file of files) {
248 await remove(join(ConfigManager.Instance.getTranscodingDirectory(), file))
252 private async onExit () {
253 if (this.cleaningUp) return
254 this.cleaningUp = true
256 logger.info('Cleaning up after program exit')
259 for (const { server, job } of this.processingJobs) {
260 await server.runnerJobs.abort({
261 jobToken: job.jobToken,
263 reason: 'Runner stopped',
264 runnerToken: server.runnerToken
268 await this.cleanupTMP()
277 static get Instance () {
278 return this.instance || (this.instance = new this())