Skip to content

README.md

Latest commit

 

History

History
157 lines (96 loc) · 4.89 KB

README.md

File metadata and controls

157 lines (96 loc) · 4.89 KB
Raw

Yorkie Load Test and Profiling with k6 and pprof

Overview

This document provides an overview of the Yorkie load test script and how to profile the Yorkie server using pprof.

Prerequisites

Install k6: Ensure you have k6 installed on your machine. You can download it from k6.io.

Available Load Tests

There are three types of load tests available:

Basic Presence Test (presence.ts)

  • Tests basic presence functionality without streaming
  • Simulates users updating their presence data via Document API
  • Ideal for testing presence update latency and throughput in PushPull

Presence with Stream Test (presence-with-stream.ts)

  • Tests real-time presence streaming capabilities via Document API
  • Simulates both watchers (who observe changes) and updaters (who make changes)
  • Tests watch stream performance and resource usage

Channel Presence Test (channel-presence.ts)

  • Tests the dedicated Channel API (AttachChannel, DetachChannel)
  • Simulates clients repeatedly attaching and detaching from channels
  • Ideal for testing channel presence count accuracy and attach/detach performance
  • Verifies that presence count increments/decrements correctly under load

Running the Yorkie Server

Start the Yorkie server with pprof enabled:

yorkie server --mongo-connection-uri mongodb://localhost:27017 --pprof-enabled

Running Load Tests

Basic Presence Test

Run the basic presence load test in even mode (distributed across multiple documents):

k6 run -e DOC_KEY_PREFIX=even-1 -e TEST_MODE=even -e CONCURRENCY=500 -e VU_PER_DOCS=10 presence.ts

This runs the test with 50 documents, each with 10 virtual users, for a total of 500 concurrent users.

Run the test in skew mode (all users on a single document):

k6 run -e DOC_KEY_PREFIX=skew-1 -e TEST_MODE=skew -e CONCURRENCY=500 presence.ts

This runs the test with a single document with 500 virtual users.

Presence with Stream Test

Run the advanced presence test with streaming in even mode:

k6 run -e DOC_KEY_PREFIX=stream-1 -e TEST_MODE=even -e CONCURRENCY=500 -e VU_PER_DOCS=10 -e WATCHER_RATIO=0.5 presence-with-stream.ts

This test simulates a realistic scenario with both watchers and updaters:

  • Watchers: Connect to documents and watch for presence changes via streaming
  • Updaters: Continuously make presence updates to documents

Run the stream test in skew mode (high-contention scenario):

k6 run -e DOC_KEY_PREFIX=stream-skew-1 -e TEST_MODE=skew -e CONCURRENCY=500 -e WATCHER_RATIO=0.3 presence-with-stream.ts

This creates a high-contention scenario with 150 watchers and 350 updaters all on the same document.

Channel Presence Test

Run the channel API test in even mode (distributed across multiple channels):

k6 run -e CHANNEL_KEY_PREFIX=channel-1 -e TEST_MODE=even -e CONCURRENCY=500 -e VU_PER_CHANNEL=10 -e ATTACH_ITERATIONS=5 channel-presence.ts

This runs the test with 50 channels, each with 10 virtual users, where each user performs 5 attach/detach cycles.

Run the test in skew mode (all users on a single channel for high contention):

k6 run -e CHANNEL_KEY_PREFIX=channel-skew-1 -e TEST_MODE=skew -e CONCURRENCY=500 -e ATTACH_ITERATIONS=5 channel-presence.ts

This creates a high-contention scenario with 500 users all attaching/detaching from the same channel, ideal for testing presence count accuracy under heavy concurrent access.

Test Parameters

Common Parameters

  • DOC_KEY_PREFIX: Prefix for document keys (creates unique keys for each test run)
  • TEST_MODE: Choose between skew (single document) or even (multiple documents)
  • CONCURRENCY: Number of concurrent users to simulate

Even Mode Parameters

  • VU_PER_DOCS: Number of virtual users per document (only in even mode)

Presence Test Parameters

  • WITH_VERSION_VECTOR: If true, update presence including explicit version vectors

Stream Test Parameters

  • WATCHER_RATIO: Ratio of watchers to total users (0.0 to 1.0). Example: 0.5 = 50% watchers, 50% updaters

Channel Presence Test Parameters

  • CHANNEL_KEY_PREFIX: Prefix for channel keys (creates unique keys for each test run)
  • VU_PER_CHANNEL: Number of virtual users per channel (only in even mode)
  • ATTACH_ITERATIONS: Number of attach/detach cycles each user performs (default: 5)

Profiling with pprof

The Yorkie server runs with pprof enabled on port 8081.

CPU Profiling

Collect CPU profile data:

curl http://localhost:8081/debug/pprof/profile\?seconds\=150 --output cpu.out

Open the interactive pprof web tool:

go tool pprof -http=:9090 cpu.out

Memory Profiling

Collect heap memory profile data (during load test):

curl http://localhost:8081/debug/pprof/heap --output mem.out

Open the interactive pprof web tool:

go tool pprof -http=:9090 mem.out