add_option() racy behaviour

[karl94]

If two or more processes try to add an option via add_option() at the same time with both the same value, the following calls to get_option() may return false as if the option wasn't set.

This happens for example if two processes A and B: both reach past the check in ​https://github.com/WordPress/wordpress-develop/blob/6.0/src/wp-includes/option.php#L622-L624 before any of the two performs the INSERT.
In this scenario, the first process which gets to execute the INSERT will return true and will have the correct value in its object cache. The other process's add_option() will return false and it's object cache will continue to believe the option doesn't exists.

​https://github.com/WordPress/wordpress-develop/blob/6.0/src/wp-includes/option.php#L640-L641
​https://github.com/WordPress/wordpress-develop/blob/6.0/src/wp-includes/wp-db.php#L2116-L2131
​https://dev.mysql.com/doc/refman/8.0/en/insert-on-duplicate.html#:~:text=0%20if%20an%20existing%20row%20is%20set%20to%20its%20current%20values

You can find attached a plugin designed for testing this race condition.

[SergeyBiryukov]

Hi there, welcome back to WordPress Trac!

Just linking to some related tickets here: #38931, #51352, #51372.

[karl94]

@SergeyBiryukov thanks, I think #51486 is also related.

I prepared a preliminary patch which takes into account both issues, could you please take a look at it and let me know what you think?

[SergeyBiryukov]

Thanks for the patch! Moving to 6.1 for review.

[audrasjb]

Unfortunately, this ticket didn’t get enough momentum in 6.1 release cycle… sorry about that, but the proposed patch still needs testing, so let's move it to Future Release milestone, at least until it's properly reviewed and tested.