Call me maybe Jepsen and flaky networks

on YouTube ↗41m

" summary

Introduces Jepsen, a tool that injects network partitions and other failures to test distributed datastores (MongoDB, Cassandra, Elasticsearch, Solr) running over asynchronous unreliable networks. Covers how Jepsen works, the kinds of tests you can write, and uses those tests to explore distributed-consensus subtleties, CAP-related guarantees and real failure modes discovered in Apache Solr and other systems.

AI-written summary · about

" chapters (9)

1. 0:00Start
2. 2:58Network partition
3. 4:11CAP recap
4. 9:41Anatomy of a Jepsen test
5. 15:23A set of integers: cas-set-client
6. 17:00Compare and Set client
7. 25:22Solr - Bugs, bugs & bugs
8. 37:26Conclusion
9. 38:24References

" related talks

• 55mJepsen IV: Hope Springs EternalKyle Kingsbury · GOTO Chicago 2015
• 42mJepsen 9: A Fsyncing FeelingKyle Kingsbury · GOTO Chicago 2018
• 53mKeynote - Jepsen 7: Anna ConcurreninaKyle Kingsbury · Scala Days Copenhagen 2017
• 45mJepsen II: Linearizable BoogalooKyle Kingsbury · Strange Loop 2014
• 43mPartitions for Everyone!Kyle Kingsbury · Strange Loop 2013
• 42mPlanning To FailDave Gardner · PHP UK Conference 2013
• 43mThe Power of Data-Driven Testing: A Deep Dive into Jakarta Persistence Specifications and NoSQLOtavio Santana, Elias Nogueira · EclipseCon 2023
• 44mBeauty and the Beast: your application and distributed systemsEmily Stolfo · rubyday 2019