feat(blobby): add new recordings-blob-ingestion-overflow role

plugin-server/src/capabilities.ts

@@ -19,6 +19,7 @@ export function getPluginServerCapabilities(config: PluginsServerConfig): Plugin
                 processAsyncOnEventHandlers: true,
                 processAsyncWebhooksHandlers: true,
                 sessionRecordingBlobIngestion: true,
+                sessionRecordingBlobOverflowIngestion: config.SESSION_RECORDING_OVERFLOW_ENABLED,
                 personOverrides: true,
                 appManagementSingleton: true,
                 preflightSchedules: true,
@@ -55,7 +56,11 @@ export function getPluginServerCapabilities(config: PluginsServerConfig): Plugin
                 sessionRecordingBlobIngestion: true,
                 ...sharedCapabilities,
             }
-
+        case PluginServerMode.recordings_blob_ingestion_overflow:
+            return {
+                sessionRecordingBlobOverflowIngestion: true,
+                ...sharedCapabilities,
+            }
         case PluginServerMode.recordings_ingestion_v3:
             return {
                 sessionRecordingV3Ingestion: true,

plugin-server/src/config/kafka-topics.ts

@@ -29,6 +29,8 @@ export const KAFKA_PERSON_OVERRIDE = `${prefix}clickhouse_person_override${suffi
 
 // read session recording snapshot items
 export const KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS = `${prefix}session_recording_snapshot_item_events${suffix}`
+export const KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_OVERFLOW = `${prefix}session_recording_snapshot_item_overflow${suffix}`
+
 // write session recording and replay events to ClickHouse
 export const KAFKA_CLICKHOUSE_SESSION_RECORDING_EVENTS = `${prefix}clickhouse_session_recording_events${suffix}`
 export const KAFKA_CLICKHOUSE_SESSION_REPLAY_EVENTS = `${prefix}clickhouse_session_replay_events${suffix}`

plugin-server/src/main/ingestion-queues/session-recording/session-recordings-consumer.ts

@@ -6,7 +6,10 @@ import { CODES, features, KafkaConsumer, librdkafkaVersion, Message, TopicPartit
 import { Counter, Gauge, Histogram } from 'prom-client'
 
 import { sessionRecordingConsumerConfig } from '../../../config/config'
-import { KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS } from '../../../config/kafka-topics'
+import {
+    KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS,
+    KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_OVERFLOW,
+} from '../../../config/kafka-topics'
 import { BatchConsumer, startBatchConsumer } from '../../../kafka/batch-consumer'
 import { createRdConnectionConfigFromEnvVars, createRdProducerConfigFromEnvVars } from '../../../kafka/config'
 import { createKafkaProducer } from '../../../kafka/producer'
@@ -43,6 +46,7 @@ require('@sentry/tracing')
 
 // WARNING: Do not change this - it will essentially reset the consumer
 const KAFKA_CONSUMER_GROUP_ID = 'session-recordings-blob'
+const KAFKA_CONSUMER_GROUP_ID_OVERFLOW = 'session-recordings-blob-overflow'
 const KAFKA_CONSUMER_SESSION_TIMEOUT_MS = 30000
 const SHUTDOWN_FLUSH_TIMEOUT_MS = 30000
 const CAPTURE_OVERFLOW_REDIS_KEY = '@posthog/capture-overflow/replay'
@@ -141,7 +145,8 @@ export class SessionRecordingIngester {
     teamsRefresher: BackgroundRefresher<Record<string, TeamIDWithConfig>>
     latestOffsetsRefresher: BackgroundRefresher<Record<number, number | undefined>>
     config: PluginsServerConfig
-    topic = KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS
+    topic: string
+    consumerGroupId: string
     totalNumPartitions = 0
     isStopping = false
 
@@ -156,11 +161,16 @@ export class SessionRecordingIngester {
         private globalServerConfig: PluginsServerConfig,
         private postgres: PostgresRouter,
         private objectStorage: ObjectStorage,
+        private consumeOverflow: boolean,
         captureRedis: Redis | undefined
     ) {
         this.debugPartition = globalServerConfig.SESSION_RECORDING_DEBUG_PARTITION
             ? parseInt(globalServerConfig.SESSION_RECORDING_DEBUG_PARTITION)
             : undefined
+        this.topic = consumeOverflow
+            ? KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_OVERFLOW
+            : KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS
+        this.consumerGroupId = this.consumeOverflow ? KAFKA_CONSUMER_GROUP_ID_OVERFLOW : KAFKA_CONSUMER_GROUP_ID
 
         // NOTE: globalServerConfig contains the default pluginServer values, typically not pointing at dedicated resources like kafka or redis
         // We still connect to some of the non-dedicated resources such as postgres or the Replay events kafka.
@@ -169,7 +179,7 @@ export class SessionRecordingIngester {
 
         this.realtimeManager = new RealtimeManager(this.redisPool, this.config)
 
-        if (globalServerConfig.SESSION_RECORDING_OVERFLOW_ENABLED && captureRedis) {
+        if (globalServerConfig.SESSION_RECORDING_OVERFLOW_ENABLED && captureRedis && !consumeOverflow) {
             this.overflowDetection = new OverflowManager(
                 globalServerConfig.SESSION_RECORDING_OVERFLOW_BUCKET_CAPACITY,
                 globalServerConfig.SESSION_RECORDING_OVERFLOW_BUCKET_REPLENISH_RATE,
@@ -207,7 +217,7 @@ export class SessionRecordingIngester {
         this.latestOffsetsRefresher = new BackgroundRefresher(async () => {
             const results = await Promise.all(
                 this.assignedTopicPartitions.map(({ partition }) =>
-                    queryWatermarkOffsets(this.connectedBatchConsumer, partition).catch((err) => {
+                    queryWatermarkOffsets(this.connectedBatchConsumer, this.topic, partition).catch((err) => {
                         // NOTE: This can error due to a timeout or the consumer being disconnected, not stop the process
                         // as it is currently only used for reporting lag.
                         captureException(err)
@@ -259,8 +269,6 @@ export class SessionRecordingIngester {
 
         const { team_id, session_id } = event
         const key = `${team_id}-${session_id}`
-        // TODO: use this for session key too if it's safe to do so
-        const overflowKey = `${team_id}:${session_id}`
 
         const { partition, highOffset } = event.metadata
         const isDebug = this.debugPartition === partition
@@ -285,11 +293,7 @@ export class SessionRecordingIngester {
 
         // Check that we are not below the high-water mark for this partition (another consumer may have flushed further than us when revoking)
         if (
-            await this.persistentHighWaterMarker.isBelowHighWaterMark(
-                event.metadata,
-                KAFKA_CONSUMER_GROUP_ID,
-                highOffset
-            )
+            await this.persistentHighWaterMarker.isBelowHighWaterMark(event.metadata, this.consumerGroupId, highOffset)
         ) {
             dropEvent('high_water_mark_partition')
             return
@@ -318,7 +322,7 @@ export class SessionRecordingIngester {
 
         await Promise.allSettled([
             this.sessions[key]?.add(event),
-            this.overflowDetection?.observe(overflowKey, event.metadata.rawSize, event.metadata.timestamp),
+            this.overflowDetection?.observe(session_id, event.metadata.rawSize, event.metadata.timestamp),
         ])
     }
 
@@ -486,8 +490,8 @@ export class SessionRecordingIngester {
         const replayClusterConnectionConfig = createRdConnectionConfigFromEnvVars(this.config)
         this.batchConsumer = await startBatchConsumer({
             connectionConfig: replayClusterConnectionConfig,
-            groupId: KAFKA_CONSUMER_GROUP_ID,
-            topic: KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS,
+            groupId: this.consumerGroupId,
+            topic: this.topic,
             autoCommit: false,
             sessionTimeout: KAFKA_CONSUMER_SESSION_TIMEOUT_MS,
             maxPollIntervalMs: this.config.KAFKA_CONSUMPTION_MAX_POLL_INTERVAL_MS,
@@ -510,7 +514,7 @@ export class SessionRecordingIngester {
             debug: this.config.SESSION_RECORDING_KAFKA_DEBUG,
         })
 
-        this.totalNumPartitions = (await getPartitionsForTopic(this.connectedBatchConsumer)).length
+        this.totalNumPartitions = (await getPartitionsForTopic(this.connectedBatchConsumer, this.topic)).length
 
         addSentryBreadcrumbsEventListeners(this.batchConsumer.consumer)
 
@@ -820,7 +824,7 @@ export class SessionRecordingIngester {
                 })
 
                 // Store the committed offset to the persistent store to avoid rebalance issues
-                await this.persistentHighWaterMarker.add(tp, KAFKA_CONSUMER_GROUP_ID, highestOffsetToCommit)
+                await this.persistentHighWaterMarker.add(tp, this.consumerGroupId, highestOffsetToCommit)
                 // Clear all session offsets below the committed offset (as we know they have been flushed)
                 await this.sessionHighWaterMarker.clear(tp, highestOffsetToCommit)
                 gaugeOffsetCommitted.set({ partition }, highestOffsetToCommit)

plugin-server/src/main/ingestion-queues/session-recording/utils.ts

@@ -4,7 +4,6 @@ import { KafkaConsumer, Message, MessageHeader, PartitionMetadata, TopicPartitio
 import path from 'path'
 import { Counter } from 'prom-client'
 
-import { KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS } from '../../../config/kafka-topics'
 import { PipelineEvent, RawEventMessage, RRWebEvent } from '../../../types'
 import { KafkaProducerWrapper } from '../../../utils/db/kafka-producer-wrapper'
 import { status } from '../../../utils/status'
@@ -36,6 +35,7 @@ export const bufferFileDir = (root: string) => path.join(root, 'session-buffer-f
 
 export const queryWatermarkOffsets = (
     kafkaConsumer: KafkaConsumer | undefined,
+    topic: string,
     partition: number,
     timeout = 10000
 ): Promise<[number, number]> => {
@@ -44,20 +44,15 @@ export const queryWatermarkOffsets = (
             return reject('Not connected')
         }
 
-        kafkaConsumer.queryWatermarkOffsets(
-            KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS,
-            partition,
-            timeout,
-            (err, offsets) => {
-                if (err) {
-                    captureException(err)
-                    status.error('🔥', 'Failed to query kafka watermark offsets', err)
-                    return reject(err)
-                }
-
-                resolve([partition, offsets.highOffset])
+        kafkaConsumer.queryWatermarkOffsets(topic, partition, timeout, (err, offsets) => {
+            if (err) {
+                captureException(err)
+                status.error('🔥', 'Failed to query kafka watermark offsets', err)
+                return reject(err)
             }
-        )
+
+            resolve([partition, offsets.highOffset])
+        })
     })
 }
 
@@ -89,7 +84,7 @@ export const queryCommittedOffsets = (
 
 export const getPartitionsForTopic = (
     kafkaConsumer: KafkaConsumer | undefined,
-    topic = KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS
+    topic: string
 ): Promise<PartitionMetadata[]> => {
     return new Promise<PartitionMetadata[]>((resolve, reject) => {
         if (!kafkaConsumer) {

plugin-server/src/main/pluginsServer.ts

@@ -109,6 +109,7 @@ export async function startPluginsServer(
     // meantime.
     let bufferConsumer: Consumer | undefined
     let stopSessionRecordingBlobConsumer: (() => void) | undefined
+    let stopSessionRecordingBlobOverflowConsumer: (() => void) | undefined
     let jobsConsumer: Consumer | undefined
     let schedulerTasksConsumer: Consumer | undefined
 
@@ -151,6 +152,7 @@ export async function startPluginsServer(
             bufferConsumer?.disconnect(),
             jobsConsumer?.disconnect(),
             stopSessionRecordingBlobConsumer?.(),
+            stopSessionRecordingBlobOverflowConsumer?.(),
             schedulerTasksConsumer?.disconnect(),
             personOverridesPeriodicTask?.stop(),
         ])
@@ -446,7 +448,7 @@ export async function startPluginsServer(
                 throw new Error("Can't start session recording blob ingestion without object storage")
             }
             // NOTE: We intentionally pass in the original serverConfig as the ingester uses both kafkas
-            const ingester = new SessionRecordingIngester(serverConfig, postgres, s3, captureRedis)
+            const ingester = new SessionRecordingIngester(serverConfig, postgres, s3, false, captureRedis)
             await ingester.start()
 
             const batchConsumer = ingester.batchConsumer
@@ -458,6 +460,28 @@ export async function startPluginsServer(
             }
         }
 
+        if (capabilities.sessionRecordingBlobOverflowIngestion) {
+            const recordingConsumerConfig = sessionRecordingConsumerConfig(serverConfig)
+            const postgres = hub?.postgres ?? new PostgresRouter(serverConfig)
+            const s3 = hub?.objectStorage ?? getObjectStorage(recordingConsumerConfig)
+
+            if (!s3) {
+                throw new Error("Can't start session recording blob ingestion without object storage")
+            }
+            // NOTE: We intentionally pass in the original serverConfig as the ingester uses both kafkas
+            // NOTE: We don't pass captureRedis to disable overflow computation on the overflow topic
+            const ingester = new SessionRecordingIngester(serverConfig, postgres, s3, true, undefined)
+            await ingester.start()
+
+            const batchConsumer = ingester.batchConsumer
+
+            if (batchConsumer) {
+                stopSessionRecordingBlobOverflowConsumer = () => ingester.stop()
+                shutdownOnConsumerExit(batchConsumer)
+                healthChecks['session-recordings-blob-overflow'] = () => ingester.isHealthy() ?? false
+            }
+        }
+
         if (capabilities.sessionRecordingV3Ingestion) {
             const recordingConsumerConfig = sessionRecordingConsumerConfig(serverConfig)
             const postgres = hub?.postgres ?? new PostgresRouter(serverConfig)

plugin-server/src/types.ts

@@ -77,6 +77,7 @@ export enum PluginServerMode {
     scheduler = 'scheduler',
     analytics_ingestion = 'analytics-ingestion',
     recordings_blob_ingestion = 'recordings-blob-ingestion',
+    recordings_blob_ingestion_overflow = 'recordings-blob-ingestion-overflow',
     recordings_ingestion_v3 = 'recordings-ingestion-v3',
     person_overrides = 'person-overrides',
 }
@@ -306,6 +307,7 @@ export interface PluginServerCapabilities {
     processAsyncOnEventHandlers?: boolean
     processAsyncWebhooksHandlers?: boolean
     sessionRecordingBlobIngestion?: boolean
+    sessionRecordingBlobOverflowIngestion?: boolean
     sessionRecordingV3Ingestion?: boolean
     personOverrides?: boolean
     appManagementSingleton?: boolean

plugin-server/tests/main/ingestion-queues/session-recording/fixtures.ts

@@ -1,6 +1,5 @@
 import { Message } from 'node-rdkafka'
 
-import { KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS } from '../../../../src/config/kafka-topics'
 import { IncomingRecordingMessage } from '../../../../src/main/ingestion-queues/session-recording/types'
 import jsonFullSnapshot from './data/snapshot-full.json'
 
@@ -41,13 +40,14 @@ export function createIncomingRecordingMessage(
 }
 
 export function createKafkaMessage(
+    topic: string,
     token: number | string,
     messageOverrides: Partial<Message> = {},
     eventProperties: Record<string, any> = {}
 ): Message {
     return {
         partition: 1,
-        topic: KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS,
+        topic,
         offset: 0,
         timestamp: messageOverrides.timestamp ?? Date.now(),
         size: 1,
@@ -72,6 +72,6 @@ export function createKafkaMessage(
     }
 }
 
-export function createTP(partition: number, topic = KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS) {
+export function createTP(partition: number, topic: string) {
     return { topic, partition }
 }

plugin-server/tests/main/ingestion-queues/session-recording/session-recordings-consumer-v3.test.ts

@@ -6,6 +6,7 @@ import path from 'path'
 
 import { waitForExpect } from '../../../../functional_tests/expectations'
 import { defaultConfig } from '../../../../src/config/config'
+import { KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS } from '../../../../src/config/kafka-topics'
 import {
     SessionManagerBufferContext,
     SessionManagerContext,
@@ -76,6 +77,7 @@ describe('ingester', () => {
     let teamToken = ''
     let mockOffsets: Record<number, number> = {}
     let mockCommittedOffsets: Record<number, number> = {}
+    const consumedTopic = KAFKA_SESSION_RECORDING_SNAPSHOT_ITEM_EVENTS
 
     beforeAll(async () => {
         mkdirSync(path.join(config.SESSION_RECORDING_LOCAL_DIRECTORY, 'session-buffer-files'), { recursive: true })
@@ -120,7 +122,7 @@ describe('ingester', () => {
         ingester = new SessionRecordingIngesterV3(config, hub.postgres, hub.objectStorage)
         await ingester.start()
 
-        mockConsumer.assignments.mockImplementation(() => [createTP(0), createTP(1)])
+        mockConsumer.assignments.mockImplementation(() => [createTP(0, consumedTopic), createTP(1, consumedTopic)])
     })
 
     afterEach(async () => {
@@ -139,6 +141,7 @@ describe('ingester', () => {
         mockOffsets[partition]++
 
         return createKafkaMessage(
+            consumedTopic,
             teamToken,
             {
                 partition,
@@ -223,7 +226,7 @@ describe('ingester', () => {
 
         describe('batch event processing', () => {
             it('should batch parse incoming events and batch them to reduce writes', async () => {
-                mockConsumer.assignments.mockImplementation(() => [createTP(1)])
+                mockConsumer.assignments.mockImplementation(() => [createTP(1, consumedTopic)])
                 await ingester.handleEachBatch(
                     [
                         createMessage('session_id_1', 1),
@@ -279,7 +282,11 @@ describe('ingester', () => {
                 const partitionMsgs1 = [createMessage('session_id_1', 1), createMessage('session_id_2', 1)]
                 const partitionMsgs2 = [createMessage('session_id_3', 2), createMessage('session_id_4', 2)]
 
-                mockConsumer.assignments.mockImplementation(() => [createTP(1), createTP(2), createTP(3)])
+                mockConsumer.assignments.mockImplementation(() => [
+                    createTP(1, consumedTopic),
+                    createTP(2, consumedTopic),
+                    createTP(3, consumedTopic),
+                ])
                 await ingester.handleEachBatch([...partitionMsgs1, ...partitionMsgs2], noop)
 
                 expect(getSessions(ingester)).toMatchObject([
@@ -291,12 +298,15 @@ describe('ingester', () => {
 
                 // Call handleEachBatch with both consumers - we simulate the assignments which
                 // is what is responsible for the actual syncing of the sessions
-                mockConsumer.assignments.mockImplementation(() => [createTP(2), createTP(3)])
+                mockConsumer.assignments.mockImplementation(() => [
+                    createTP(2, consumedTopic),
+                    createTP(3, consumedTopic),
+                ])
                 await otherIngester.handleEachBatch(
                     [createMessage('session_id_4', 2), createMessage('session_id_5', 2)],
                     noop
                 )
-                mockConsumer.assignments.mockImplementation(() => [createTP(1)])
+                mockConsumer.assignments.mockImplementation(() => [createTP(1, consumedTopic)])
                 await ingester.handleEachBatch([createMessage('session_id_1', 1)], noop)
 
                 // Should still have the partition 1 sessions that didnt move with added events
@@ -317,8 +327,8 @@ describe('ingester', () => {
                 // non-zero offset because the code can't commit offset 0
                 await ingester.handleEachBatch(
                     [
-                        createKafkaMessage('invalid_token', { offset: 12 }),
-                        createKafkaMessage('invalid_token', { offset: 13 }),
+                        createKafkaMessage(consumedTopic, 'invalid_token', { offset: 12 }),
+                        createKafkaMessage(consumedTopic, 'invalid_token', { offset: 13 }),
                     ],
                     noop
                 )