Question

if we have an UPDATE with a sub-SELECT, can the subquery execute concurrently or not under READ COMMITTED isolation?

In other words, is there a race condition present in the following:

update list set [state] = 'active' 
where
    id = (select top 1 id from list where [state] = 'ready' order by id)

In yet other words, if many connections are simulataneously executing this SQL, can we guarantee that one row is in fact updated per invocation (so long as rows in 'ready' state exist)?

Was it helpful?

Solution

The answer is yes, there is a race condition and two transactions may execute the subquery concurrently, leading to the same row being subsequently updated twice.

This can be fixed by rewriting the update as

update TEMP 
set [state] = 'active' 
from 
    (select top 1 * from list where [state] = 'ready' order by id) TEMP

I frankly don't know why this should be different, but it is. SQL Server will now take update locks ("intent to update") when executing the subquery, preventing concurrent transactions from picking the same row.

Licensed under: CC-BY-SA with attribution
Not affiliated with StackOverflow
scroll top