Our great sponsors
-
-
artillery
Load testing at cloud-scale. Serverless & distributed out-of-the-box. Load test with Playwright. Load test HTTP APIs, GraphQL, WebSocket, and more. Use any Node.js module. Never fail to scale with Artillery!
-
InfluxDB
Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.
-
Redis
Redis is an in-memory database that persists on disk. The data model is key-value, but many different kind of values are supported: Strings, Lists, Sets, Sorted Sets, Hashes, Streams, HyperLogLogs, Bitmaps.
Finding out what's causing that was not that simple as Jedis does not write a lot of log information when the log level is set to debug. From what we could tell from the Jedis source code it seems that when the Jedis client connects to the configuration endpoint it fetches all known hosts. It stores that in a local cache and figures out the primary node from the set of nodes. Once the primary node is known it will execute operations against the primary node. It does so by connecting to the Amazon-generated domain name/entry and when that entry is down / unreachable it seems it does not use the configuration endpoint to rediscover the nodes or fetch the new cluster topology. We initially thought this had to do with the DNS caching of the JVM, so we also tried to disable caching, but we did not see any effect after that change.
To test the impact we created a simple Lambda function that performs GET operations on a specific set of keys within our cache. As we wanted to test under 'real-world' circumstances we created a small load test with Artillery.io that was sending a constant stream of requests.
The AWS Lambda function had been written in Java and therefore the team looked at different Java-based Redis drivers and eventually decided to use Jedis, a Java client for Redis designed for performance and ease of use. One of the reasons they chose to use Jedis was that it proved to be a lightweight library. Instantiating the client connection takes somewhere between 200-600ms (depending on the lambda memory settings) which impacts the cold start of the lambda function.