Sepia: How LinkedIn Mobile Made Integration Testing Fast and Reliable in Node.js

LinkedIn Mobile runs a node.js server powering clients on three platforms: iPhone, Android and mobile web. The server aggregates data from downstream services using HTTP, services such as Voldemort data stores and rest.li web services. At first, emulating a client and exercising the server API made sense to validate the server. As the integration test suite expanded to over 200 tests, and the number of developers grew, the test suite became too slow and unreliable to depend on. The downstream HTTP requests incurred network latencies and were subject to the health of the downstream services. This talk walks through how LinkedIn is addressing the challenges of scaling a complex testing infrastructure in order to release quality code, and how we developed a open-source tool called sepia to help other companies do the same.

• The LinkedIn Mobile Architecture
• The server API, downstream HTTP requests
• The integration test suite
• Why the integration test suite is unreliable

• Old attempts and a solution
• Why vcr.js doesn’t address our use case
• Why not shim http#request?
• Recording fixtures and playing them back

• Fixture resolution edge-cases
• Non-repeatable data in HTTP requests
• Non-referentially transparent HTTP requests

• Success!

• Battle-testing sepia
• Supporting multiple authentication schemes
• Caching old fixtures, recording new ones
• Bypassing fixtures for certain HTTP requests
• Test-specific fixture directories

• How YOU can use sepia to make your software more robust