How Share Everything Systems Fence Off Zombies
Fencing is a useful pattern in many distributed systems, but is especially critical in Share Everything systems. Share Everything systems are no longer common, but I think they are making a comeback!
So in this episode I explain what are Share Everything systems, how they are different from the more common Share Nothing systems, their benefits and why I believe they are making a comeback. Most important - We'll discuss fencing. I'll cover two key fencing techniques, and why STONITH is so popular and also very challenging to implement correctly.
SaaS Developer Slack: https://launchpass.com/all-about-saas
The STONITH tweet that I mentioned, with all the awesome responses: https://twitter.com/gwenshap/status/1...
EBS now allows attaching volumes on multiple nodes: https://docs.aws.amazon.com/AWSEC2/la...
Zombie Fencing in Kafka Streams: https://www.confluent.io/blog/transac...
STONITH in Oracle RAC: https://kevinclosson.net/category/sto...
Hardware fencing in Oracle Linux: https://docs.oracle.com/en/operating-...
00:00 intro
1:27 What's fencing?
2:00 Examples of share nothing systems
3:13 What's Share Everything?
5:24 Benefits of Share Everything
6:00 Shared Disks in the Cloud
7:50 Challenges of Share Everything
9:40 Fencing with Epochs
12:12 Fencing by Shooting
14:51 Challenges in implementing STONITH
17:57 Getting nodes to shoot themselves
19:07 Throwing Hardware at the Problem
20:00 Shooting nodes - the cloud way!
21:18 Choosing a Fencing Method
23:52 To summarize