Alluxio Stress Testing Methods and Practices

The article introduces Alluxio stress testing, originally presented by Alluxio engineer Ding Bowen at an online meetup, and outlines three main topics: the principles of stress testing, the classification of Alluxio stress tools, and practical usage of these tools.

1. Purpose of stress testing – to determine how configuration and parameters affect performance, to identify the system's safe operating boundaries, and to simulate failure scenarios for emergency planning. It also distinguishes non‑stress testing goals such as functional verification and component integration.

2. Sources of pressure in Alluxio – client requests (metadata operations on Master, read/write on Workers, UFS I/O), internal asynchronous and periodic tasks (async persistence, replica checks, worker status reports, health checks), and scale factors (large numbers of files or Workers).

3. Manifestations of pressure – high concurrency, large data volume, and complex operations (e.g., recursive deletes), each stressing different system resources like CPU, memory, storage, or network.

4. Built‑in stress testing framework – defines stress operations (representative file system actions) and stress tasks (repeated execution of operations). Two execution modes are supported: cluster mode (using Job Service to distribute tasks to Job Workers) and standalone mode (running directly on a single node). The framework collects per‑operation latency, calculates throughput, and aggregates statistics.

5. Result calculation methods – time‑based (using --duration ) and count‑based (using --stop-count ) approaches, each suitable for different testing scenarios.

6. Reporting – after a run, each Job Worker reports to the Job Master, which compiles a report containing throughput, latency metrics, and the configuration used.

7. Stress tool classification – seven categories are provided: Master (e.g., StressMasterBench ), Worker ( StressWorkerBench ), Job Service ( StressJobServiceBench ), UFS ( UfsIOBench ), FUSE ( FuseIOBench ), Client (e.g., StressClientBench , CompactionBench ), and RPC (e.g., WorkerHeartbeatBench , RegisterWorkerBench , GetPinnedFileIdsBench ). Example tools and their configurable parameters are described.

8. Running stress tools – tools reside in the alluxio.stress.cli package. Users can execute them via the Alluxio command line, for example: bin/alluxio runClass alluxio.stress.cli.StressMasterBench , and view options with --help .

9. Common issues and debugging – includes handling timeouts, all‑operations‑failed cases, partial failures, and suggests enabling debug logs (found in logs/user/<username>.log for both standalone and cluster modes) to diagnose problems.

The article concludes with a reminder to follow future Alluxio meetups for more technical sharing.