Babel Loop Timeouts

benchmarks/clickhouse/e2e.kafka.benchmark.ts

@@ -1,11 +1,11 @@
 import { performance } from 'perf_hooks'
 
-import { createPosthog, DummyPostHog } from '../../src/extensions/posthog'
 import { KAFKA_EVENTS_PLUGIN_INGESTION } from '../../src/ingestion/topics'
 import { startPluginsServer } from '../../src/server'
 import { LogLevel, PluginsServerConfig, Queue } from '../../src/types'
 import { PluginsServer } from '../../src/types'
 import { delay, UUIDT } from '../../src/utils'
+import { createPosthog, DummyPostHog } from '../../src/vm/extensions/posthog'
 import { makePiscina } from '../../src/worker/piscina'
 import { resetTestDatabaseClickhouse } from '../../tests/helpers/clickhouse'
 import { resetKafka } from '../../tests/helpers/kafka'
@@ -84,7 +84,7 @@ describe('e2e kafka & clickhouse benchmark', () => {
 
         const n = (n: number) => `${Math.round(n * 100) / 100}`
         console.log(
-            `[Kafka & ClickHouse] Ingested ${count} events in ${n(timeMs / 1000)}s (${n(
+            `ℹ️️ [Kafka & ClickHouse] Ingested ${count} events in ${n(timeMs / 1000)}s (${n(
                 1000 / (timeMs / count)
             )} events/sec, ${n(timeMs / count)}ms per event)`
         )

benchmarks/clickhouse/e2e.timeout.benchmark.ts

@@ -0,0 +1,97 @@
+import { performance } from 'perf_hooks'
+
+import { KAFKA_EVENTS_PLUGIN_INGESTION } from '../../src/ingestion/topics'
+import { startPluginsServer } from '../../src/server'
+import { ClickHouseEvent, LogLevel, PluginsServerConfig, Queue } from '../../src/types'
+import { PluginsServer } from '../../src/types'
+import { delay, UUIDT } from '../../src/utils'
+import { createPosthog, DummyPostHog } from '../../src/vm/extensions/posthog'
+import { makePiscina } from '../../src/worker/piscina'
+import { resetTestDatabaseClickhouse } from '../../tests/helpers/clickhouse'
+import { resetKafka } from '../../tests/helpers/kafka'
+import { pluginConfig39 } from '../../tests/helpers/plugins'
+import { resetTestDatabase } from '../../tests/helpers/sql'
+import { delayUntilEventIngested } from '../../tests/shared/process-event'
+
+jest.setTimeout(600000) // 10min timeout
+
+const extraServerConfig: Partial<PluginsServerConfig> = {
+    KAFKA_ENABLED: true,
+    KAFKA_HOSTS: process.env.KAFKA_HOSTS || 'kafka:9092',
+    WORKER_CONCURRENCY: 4,
+    TASK_TIMEOUT: 5,
+    PLUGIN_SERVER_INGESTION: true,
+    KAFKA_CONSUMPTION_TOPIC: KAFKA_EVENTS_PLUGIN_INGESTION,
+    KAFKA_BATCH_PARALELL_PROCESSING: true,
+    LOG_LEVEL: LogLevel.Log,
+}
+
+describe('e2e kafka processing timeout benchmark', () => {
+    let queue: Queue
+    let server: PluginsServer
+    let stopServer: () => Promise<void>
+    let posthog: DummyPostHog
+
+    beforeEach(async () => {
+        await resetTestDatabase(`
+            async function processEvent (event) {
+                await new Promise(resolve => __jestSetTimeout(() => resolve(), 15000 * Math.random()))
+                event.properties.timeout = 'no timeout'
+                return event
+            }
+        `)
+        await resetKafka(extraServerConfig)
+        await resetTestDatabaseClickhouse(extraServerConfig)
+
+        const startResponse = await startPluginsServer(extraServerConfig, makePiscina)
+        server = startResponse.server
+        stopServer = startResponse.stop
+        queue = startResponse.queue
+
+        posthog = createPosthog(server, pluginConfig39)
+    })
+
+    afterEach(async () => {
+        await stopServer()
+    })
+
+    test('measure performance', async () => {
+        console.debug = () => null
+
+        const count = 3000
+
+        // fill in the queue
+        function createEvent() {
+            const uuid = new UUIDT().toString()
+            posthog.capture('custom event', { name: 'haha', uuid, randomProperty: 'lololo' })
+        }
+        await queue.pause()
+        for (let i = 0; i < count; i++) {
+            createEvent()
+        }
+
+        // hope that 5sec is enough to load kafka with all the events (posthog.capture can't be awaited)
+        await delay(5000)
+        await queue.resume()
+
+        console.log('Starting timer')
+        const startTime = performance.now()
+        await delayUntilEventIngested(() => server.db.fetchEvents(), count, 500, count)
+        const timeMs = performance.now() - startTime
+        console.log('Finished!')
+
+        const n = (n: number) => `${Math.round(n * 100) / 100}`
+        console.log(
+            `ℹ️️ [Kafka & ClickHouse] Ingested ${count} events in ${n(timeMs / 1000)}s (${n(
+                1000 / (timeMs / count)
+            )} events/sec, ${n(timeMs / count)}ms per event)`
+        )
+        const events = (await server.db.fetchEvents()) as ClickHouseEvent[]
+        const passedEvents = events.filter((e) => e.properties.timeout).length
+        console.log(
+            `ℹ️ Out of 3000 events: ${passedEvents} took under 5sec, ${
+                3000 - passedEvents
+            } timed out. This should be a 1:2 ratio.`
+        )
+    })
+})

benchmarks/postgres/e2e.celery.benchmark.ts

@@ -1,10 +1,10 @@
 import { performance } from 'perf_hooks'
 
-import { createPosthog, DummyPostHog } from '../../src/extensions/posthog'
 import { startPluginsServer } from '../../src/server'
 import { LogLevel, PluginsServerConfig, Queue } from '../../src/types'
 import { PluginsServer } from '../../src/types'
 import { delay, UUIDT } from '../../src/utils'
+import { createPosthog, DummyPostHog } from '../../src/vm/extensions/posthog'
 import { makePiscina } from '../../src/worker/piscina'
 import { pluginConfig39 } from '../../tests/helpers/plugins'
 import { resetTestDatabase } from '../../tests/helpers/sql'

benchmarks/vm/memory.benchmark.ts

@@ -2,7 +2,7 @@ import { PluginEvent } from '@posthog/plugin-scaffold/src/types'
 
 import { createServer } from '../../src/server'
 import { Plugin, PluginConfig, PluginConfigVMReponse } from '../../src/types'
-import { createPluginConfigVM } from '../../src/vm'
+import { createPluginConfigVM } from '../../src/vm/vm'
 import { commonOrganizationId } from '../../tests/helpers/plugins'
 
 jest.mock('../../src/sql')
@@ -47,6 +47,7 @@ const mockConfig: PluginConfig = {
 }
 
 test('test vm memory usage', async () => {
+    const debug = false
     const numVMs = 1000
     const numEventsPerVM = 100
 
@@ -76,27 +77,31 @@ test('test vm memory usage', async () => {
         const vm = await createPluginConfigVM(server, mockConfig, indexJs)
         vms.push(vm)
 
-        const nowUsed = getUsed()
-        console.log(
-            `Used: ${nowUsed} MB, diff ${nowUsed - used} (${(nowUsed - usedAtStart) / (i + 1)} * ${
-                i + 1
-            } used since the start)`
-        )
-        used = nowUsed
+        if (debug || i === numVMs - 1) {
+            const nowUsed = getUsed()
+            console.log(
+                `Used: ${nowUsed} MB, diff ${nowUsed - used} (${(nowUsed - usedAtStart) / (i + 1)} * ${
+                    i + 1
+                } used since the start)`
+            )
+            used = nowUsed
+        }
     }
 
     for (let i = 0; i < numEventsPerVM; i++) {
         for (let j = 0; j < numVMs; j++) {
             await vms[j].methods.processEvent(createEvent(i + j))
         }
-        global.gc()
-        const nowUsed = getUsed()
-        console.log(
-            `Run ${i}. Used: ${nowUsed} MB, diff ${nowUsed - used} (${nowUsed - usedAtStart} used since the start, ${
-                (nowUsed - usedAtStart) / numVMs
-            } per vm)`
-        )
-        used = nowUsed
+        if (debug || i === numEventsPerVM - 1) {
+            global?.gc?.()
+            const nowUsed = getUsed()
+            console.log(
+                `Run ${i}. Used: ${nowUsed} MB, diff ${nowUsed - used} (${
+                    nowUsed - usedAtStart
+                } used since the start, ${(nowUsed - usedAtStart) / numVMs} per vm)`
+            )
+            used = nowUsed
+        }
     }
 
     await closeServer()

benchmarks/vm/worker.benchmark.ts

@@ -94,11 +94,14 @@ test('piscina worker benchmark', async () => {
             `,
         },
         {
-            testName: 'for200k',
+            // used to be 'for200k', but since we inject Date.now() code into
+            // the for/while/do loops, to throw if they are too long, running
+            // those comparisons 200k * 10k * runs * threads times is bit too much
+            testName: 'for2k',
             events: 10000,
             testCode: `
                 function processEvent (event, meta) {
-                    let j = 0; for(let i = 0; i < 200000; i++) { j = i };
+                    let j = 0; for(let i = 0; i < 2000; i++) { j = i };
                     event.properties = { "somewhere": "over the rainbow" };
                     return event
                 }

package.json

@@ -43,6 +43,7 @@
     "repository": "https://github.com/PostHog/posthog-plugin-server",
     "license": "MIT",
     "dependencies": {
+        "@babel/standalone": "^7.12.16",
         "@google-cloud/bigquery": "^5.5.0",
         "@posthog/clickhouse": "^1.7.0",
         "@sentry/node": "^5.29.0",
@@ -72,6 +73,7 @@
         "@babel/preset-typescript": "^7.8.3",
         "@posthog/plugin-scaffold": "0.2.8",
         "@types/adm-zip": "^0.4.33",
+        "@types/babel__standalone": "^7.1.3",
         "@types/ioredis": "^4.17.7",
         "@types/jest": "^26.0.15",
         "@types/luxon": "^1.25.0",
@@ -86,7 +88,6 @@
         "@types/yargs": "^15.0.9",
         "@typescript-eslint/eslint-plugin": "^4.14.0",
         "@typescript-eslint/parser": "^4.14.0",
-        "babel-core": "^7.0.0-bridge.0",
         "babel-eslint": "^10.1.0",
         "eslint": "^7.18.0",
         "eslint-config-prettier": "^7.2.0",

src/config.ts

@@ -24,7 +24,7 @@ export function getDefaultConfig(): PluginsServerConfig {
         KAFKA_CLIENT_CERT_KEY_B64: null,
         KAFKA_TRUSTED_CERT_B64: null,
         KAFKA_CONSUMPTION_TOPIC: null,
-        KAFKA_BATCH_PARALELL_PROCESSING: false,
+        KAFKA_BATCH_PARALELL_PROCESSING: true,
         PLUGIN_SERVER_INGESTION: false,
         PLUGINS_CELERY_QUEUE: 'posthog-plugins',
         REDIS_URL: 'redis://127.0.0.1',
@@ -34,6 +34,7 @@ export function getDefaultConfig(): PluginsServerConfig {
         WEB_PORT: 3008,
         WEB_HOSTNAME: '0.0.0.0',
         WORKER_CONCURRENCY: coreCount,
+        TASK_TIMEOUT: 30,
         TASKS_PER_WORKER: 10,
         LOG_LEVEL: LogLevel.Info,
         SENTRY_DSN: null,
@@ -63,6 +64,7 @@ export function getConfigHelp(): Record<keyof PluginsServerConfig, string> {
         WEB_PORT: 'port for web server to listen on',
         WEB_HOSTNAME: 'hostname for web server to listen on',
         WORKER_CONCURRENCY: 'number of concurrent worker threads',
+        TASK_TIMEOUT: 'How many seconds until tasks are timed out',
         TASKS_PER_WORKER: 'number of parallel tasks per worker thread',
         LOG_LEVEL: 'minimum log level',
         KAFKA_ENABLED: 'use Kafka instead of Celery to ingest events',

src/ingestion/kafka-queue.ts

@@ -36,7 +36,7 @@ export class KafkaQueue implements Queue {
         isRunning,
         isStale,
     }: EachBatchPayload): Promise<void> {
-        const batchProcessingTimer = new Date()
+        const batchStartTimer = new Date()
 
         const uuidOrder = new Map<string, number>()
         const uuidOffset = new Map<string, string>()
@@ -63,21 +63,33 @@ export class KafkaQueue implements Queue {
         )
 
         const processingTimeout = timeoutGuard(
-            `Took too long to run plugins on ${pluginEvents.length} events! Timeout after 30 sec!`
+            `Still running plugins on ${pluginEvents.length} events. Timeout warning after 30 sec!`
         )
         const batches = groupIntoBatches(pluginEvents, maxBatchSize)
-        const processedEvents = (await Promise.all(batches.map(this.processEventBatch))).flat()
+        const processedEvents = (
+            await Promise.all(
+                batches.map(async (batch) => {
+                    const timer = new Date()
+                    const processedBatch = this.processEventBatch(batch)
+                    this.pluginsServer.statsd?.timing('kafka_queue.single_event_batch', timer)
+                    return processedBatch
+                })
+            )
+        ).flat()
 
         clearTimeout(processingTimeout)
 
+        this.pluginsServer.statsd?.timing('kafka_queue.each_batch.process_events', batchStartTimer)
+        const batchIngestionTimer = new Date()
+
         // Sort in the original order that the events came in, putting any randomly added events to the end.
         // This is so we would resolve the correct kafka offsets in order.
         processedEvents.sort(
             (a, b) => (uuidOrder.get(a.uuid!) || pluginEvents.length) - (uuidOrder.get(b.uuid!) || pluginEvents.length)
         )
 
         const ingestionTimeout = timeoutGuard(
-            `Took too long to ingest ${processedEvents.length} events! Timeout after 30 sec!`
+            `Still ingesting ${processedEvents.length} events. Timeout warning after 30 sec!`
         )
 
         // TODO: add chunking into groups of 500 or so. Might start too many promises at once now
@@ -126,7 +138,8 @@ export class KafkaQueue implements Queue {
 
         clearTimeout(ingestionTimeout)
 
-        this.pluginsServer.statsd?.timing('kafka_queue.each_batch', batchProcessingTimer)
+        this.pluginsServer.statsd?.timing('kafka_queue.each_batch.ingest_events', batchIngestionTimer)
+        this.pluginsServer.statsd?.timing('kafka_queue.each_batch', batchStartTimer)
         resolveOffset(batch.lastOffset())
         await heartbeat()
         await commitOffsetsIfNecessary()

src/ingestion/utils.ts

@@ -1,6 +1,7 @@
 import * as Sentry from '@sentry/node'
 import crypto from 'crypto'
 
+import { defaultConfig } from '../config'
 import { BasePerson, Element, Person, RawPerson } from '../types'
 
 export function unparsePersonPartial(person: Partial<Person>): Partial<RawPerson> {
@@ -156,9 +157,9 @@ export function chainToElements(chain: string): Element[] {
     return elements
 }
 
-export function timeoutGuard(message: string): NodeJS.Timeout {
+export function timeoutGuard(message: string, timeout = defaultConfig.TASK_TIMEOUT * 1000): NodeJS.Timeout {
     return setTimeout(() => {
         console.log(`⌛⌛⌛ ${message}`)
         Sentry.captureMessage(message)
-    }, 30000)
+    }, timeout)
 }

src/plugins.ts

@@ -7,7 +7,7 @@ import { getPluginAttachmentRows, getPluginConfigRows, getPluginRows } from './s
 import { status } from './status'
 import { PluginConfig, PluginJsonConfig, PluginsServer, PluginTask, TeamId } from './types'
 import { getFileFromArchive } from './utils'
-import { createPluginConfigVM } from './vm'
+import { createPluginConfigVM } from './vm/vm'
 
 export async function setupPlugins(server: PluginsServer): Promise<void> {
     const pluginRows = await getPluginRows(server)

src/types.ts

@@ -22,6 +22,7 @@ export enum LogLevel {
 export interface PluginsServerConfig extends Record<string, any> {
     WORKER_CONCURRENCY: number
     TASKS_PER_WORKER: number
+    TASK_TIMEOUT: number
     CELERY_DEFAULT_QUEUE: string
     DATABASE_URL: string
     CLICKHOUSE_HOST: string

src/utils.ts

@@ -343,3 +343,11 @@ export function groupIntoBatches<T>(array: T[], batchSize: number): T[][] {
     }
     return batches
 }
+
+/** Template literal function, to standardize JS code used internally to form without extraneous indentation. */
+export function code(strings: TemplateStringsArray): string {
+    const stringsConcat = strings.join('…')
+    const indentation = stringsConcat.match(/^\n([ ]*)/)?.[1].length ?? 0
+    const dedentedCode = stringsConcat.replace(new RegExp(`^[ ]{${indentation}}`, 'gm'), '')
+    return dedentedCode.trim()
+}