Warning from filesystem driver: mkdir(): File exists in /data/www/wrbm/corporate/prod/vendor/tedivm/stash/src/Stash/Driver/FileSystem.php on line 234

[gggeek]

This is probably related to a race, where 2 processes might be trying to create the same directory at the same time.
My suggestion: if the mkdir call fails, check again if the dir exists and, if it does, use it

[gggeek]

PS: might relate to #302

[tedivm]

I have an idea on how to solve this (which will, unfortunately, only work on windows). However, when people start seeing this it's generally a sign that their traffic has increased to the point where they'd benefit from running the sqlite driver instead of the filesystem one.

[DavidGoodwin]

I still see this problem at line 221 - https://github.com/tedious/Stash/blob/main/src/Stash/Driver/FileSystem.php#L221C13-L221C76

PHP Error: mkdir(): File exists from /x/y/vendor/tedivm/stash/src/Stash/Driver/FileSystem.php on line 221 {"errno":2,"errstr":"mkdir(): File exists","errfile":"/x/y/vendor/tedivm/stash/src/Stash/Driver/FileSystem.php","errline":221,

In our case, we seem to have amassed a ~30GB stash cache over 24 hours using the filesystem driver ( 'find' says this is about 1.8m files ).

I'm curious how running the purge/cleanup routine on the cache will work with PHP's stat cache when there are multiple processes involved. I'd normally have littered similar code with clearstatcache calls but I don't see any of these.

@tedivm you mentioned using Sqlite for high traffic instances; I seem to think sqlite often has a single process lock, so calling purge() might cause callers to be blocked?

I suspect I'll move back to using a Redis backend anyway.

[tedivm]

If you're at the point where you've got 30gb in 24 hours I would use redis.

[DavidGoodwin]

Yes - we are/were using Redis - but we hadn't changed the default eviction policy and a busy period led us to filling up the redis cache ... so as a hot fix I moved one thing to use a filesystem cache (hence seeing the above mkdir thing!)

I was just wondering if having a table a bit like this to help people choose cache backends (my pros/cons are likely to be incorrect)

Driver	Pros	Cons
FileSystem	Simple to setup, good for low concurrency and smaller caches	Requires manual maintenance schedule (possibly time consuming), may struggle under high concurrency, not distributed.
Redis	more scalable	requires service/server, might be a spof
APC	v. fast, no network dependency	very limited size, per-process (?) so limited scalability
Memcache	network service, scalable	requires memory, not as feature rich as redis
Sqlite	good for mostly read caches, no network dependency	may not handle high concurrency well, requires maintenance schedule