update read committed docs around handling retries

[michae2]

Michael Erickson (michae2) commented:

In 23.2, under read committed isolation, applications could still encounter spurious 40001 errors in rare cases, such as lease transfers. In 24.1, under read committed isolation, we expect applications to encounter no spurious 40001 errors. (The only 40001 errors applications might encounter are all avoidable through workload changes, such as avoiding deadlocks or increasing results_buffer_size.)

This lack of 40001 errors is a special property of read committed isolation, and is definitely worth documenting.

I'm not sure exactly which pages should be updated, but from a quick glance maybe the following:

• https://www.cockroachlabs.com/docs/v24.1/transactions
• https://www.cockroachlabs.com/docs/v24.1/transaction-retry-error-reference
• https://www.cockroachlabs.com/docs/v24.1/read-committed

Jira Issue: DOC-9869

[exalate-issue-sync]

Shannon Bradshaw (shannonbradshaw) commented:
Richard Loveland I think this should be you, but let me know if it should be Ryan Kuo

[exalate-issue-sync]

Richard Loveland (rmloveland) commented:
thanks Shannon Bradshaw

Ryan Kuo I think this could go either way - historically I've done a bunch of transaction retry work but you probably have more context on the READ COMMITTED stuff right now. What do you think?

[michae2]

A relevant slack thread, for some context:

• https://cockroachlabs.slack.com/archives/C04E47S4KK9/p1705703315115239

[exalate-issue-sync]

Ryan Kuo (taroface) commented:
I agree and am happy to take this on – there was already some mention of this in the Read Committed doc that we commented out at the last minute, so this would essentially pick up on that work.

[exalate-issue-sync]

Ryan Kuo (taroface) commented:
Also, I think this should be considered part of 24.1 roadmap work, so I’ll link it to the epic.

[exalate-issue-sync]

Shannon Bradshaw (shannonbradshaw) commented:
Thanks, all!