Thread

This question may be related to a resolved thread.

Picking up on another thread that’s closed but not resolved…

Sue stated:

EE uses db_cache as a folder name, not db_cache_delete. Hmmm…

But if you look at delete_directory() in /core/core.functions.php

you’ll see the following:

mv {$path} {$path}_delete

So the delete routine is adding _delete to the end of the directory name, so in fact EE could be creating a directory called db_cache_delete.  But sue doesn’t see it in her screen shots because it is supposed to be immediately deleted.

Thought Fabrice P. would want to know!

Sue if you want to reopen that thread with Fabrice…

Now on to my (related?) problem:

I am seeing lots of the following errors in my error log:

mv: cannot move `/www/system/cache/page_cache' to a subdirectory of itself, `/www/system/cache/page_cache_delete/page_cache'
mv: cannot stat `/www/system/cache/page_cache/45b19e1a82cd4199fdbef449f8af0efb': No such file or directory

just now I see:

rm: cannot remove directory `/www/system/cache/page_cache_delete/4c61b7f8e3c2ad350cd4821e9bde7b68': No such file or directory
rm: reading directory `/www/system/cache/page_cache_delete/668ef11a4e29cd5d609cb405033f86f5': Input/output error

This is in Expression Engine 1.6.8.

The file system is mounted through an NFS.  Permissions look OK to me, but permissions are not really my expertise.

If I had to guess on what is going on I would say that there is some latency in the mv command (or the filesystem), such that even though mv returns 0 (success), when the rm command kicks in the filesystem doesn’t know that the _delete directory exists yet!  Haven’t studied the delete_directory() function carefully enough to totally understand it but this is my hunch.

Hi Peter,

I suspect you are correct about the particular implementation of NFS and its interaction with PHP on your server being the cause, as I haven’t found other instances of this coming up.

Unfortunately there’s not much EE can do about this; it’s something that should be fixed at the server level.

Peter,

Did you ever track down the cause / solution for this issue? I’m seeing it today on one of my client’s sites.

Could it also be two different servers attempting to delete the page_cache_delete directory, and one of them getting there too late?

TTFN
Travis

As Sue and Brandon said, this is probably not an issue with EE as such. Let me move this thread to Howto.

Travis—

No we are not really much further along.  Except that we are occasionally seeing seg faults in the logs, we also are seeing memory on the Apache servers bunching up over time to the point that we have to reboot.  Happens every couple of weeks.  This is a “bad thing”.

We think both of these things may be related to the issues of EE cache on the NFS.  My server admin is looking at the EnableMMAP directive:

http://httpd.apache.org/docs/2.2/mod/core.html#enablemmap

And the EnableSendfile directive:

http://httpd.apache.org/docs/2.2/mod/core.html#enablesendfile

But I don’t think we are ready to state any conclusion.

I am not entirely clear on the mechanism of load balancing on the proxy server.  I suppose it could be the race condition you describe.

I guess we would have to set up a balancer with just one node and see if we are still seeing the issue to know conclusively…

Am regretting the use of EE page caching.  Am also regretting the use of NFS.  Initial review indicates that GFS might have been a better choice for us, but this is really outside of my expertise.

Would be curious if you have any further insight into this.

EE Page caching is a “good thing” (tm) but definitely a tricky thing.  NFS may not be as “good” 😊

Thanks for the info.

TTFN
Travis

Travis—

Is your client on NFS as well?

We make the changes described above to the Apache config this morning, are watching the results and will report back here if we find anything conclusive.

Also, above in my post of May 18, I did not describe the memory issue we are seeing correctly.  The memory is being exhausted on the NFS not on the Apache servers.  Rebooting the Apache servers frees the memory on the NFS.  I don’t really understand how that could happen, but again this is outside my area of expertise.