Questions about Etcd: Changing Raft Algorithm and Quorum Principles

[sakthiskv]

Hello all,

I am new to etcd,I have a few questions about Etcd:

Is it possible to change the Raft algorithm in Etcd?

• Can Raft be replaced or customized with another consensus algorithm?

Adjusting CP Focus in CAP Theorem

• Etcd prioritizes consistency (C) and partition tolerance (P). Can it be configured to focus more on availability (A)?

Modifying Quorum Rules

• In a 3-node cluster, writes fail if 2 nodes are down. Is there a way to allow writes with only 1 node up?

Looking forward to any insights or suggestions. Thanks!

[jmhbnz]

Hi @sakthiskv - Thanks for your questions

Can Raft be replaced or customized with another consensus algorithm?

Not practically - etcd is heavily reliant on the existing raft implementation in etcd-io/raft.

Etcd prioritizes consistency (C) and partition tolerance (P). Can it be configured to focus more on availability (A)?

A distributed etcd cluster is already highly available - Please provide some examples or descriptions on how you would see availability being improved in etcd.

In a 3-node cluster, writes fail if 2 nodes are down. Is there a way to allow writes with only 1 node up?

No - a cluster majority is required for consensus please refer to the raft paper https://raft.github.io & https://raft.github.io/raft.pdf

[sakthiskv]

Thank you for your reply @jmhbnz,

When a write operation is made in etcd, quorum confirmation is required. How does etcd handle scenarios like network failures, unresponsive nodes, or high latency? Is there any way to optimize or prioritize availability during such conditions without sacrificing functionality entirely?

For example:

• In a 3-node setup, if 2 nodes go down, the remaining node can't process writes.
• Is there flexibility to adjust quorum rules to allow writes in such cases, even if it risks consistency?

Also, are there configurations or setups to prioritize availability over consistency for specific workloads? I'd appreciate any insights or best practices!

[jmhbnz]

How does etcd handle scenarios like network failures, unresponsive nodes, or high latency?

Suggest exploring documentation :

• https://etcd.io/docs/v3.5/faq/#what-is-failure-tolerance
• https://etcd.io/docs/v3.5/tuning/
• https://etcd.io/docs/v3.5/learning/
• http://thesecretlivesofdata.com/raft/

Is there any way to optimize or prioritize availability during such conditions without sacrificing functionality entirely?

etcd is a strongly consistent distributed system. If you don't have distributed consensus/quorum (i.e. a majority of etcd nodes within a cluster in agreement) then you can't accept writes.

Is there flexibility to adjust quorum rules to allow writes in such cases, even if it risks consistency?

Not really - if you would prefer to have an eventually consistent system and don't need strong consistency then etcd might not be the best choice for your use case.