Re: cvs [server aborted]: out of memory; can not reallocate xxx bytes

[Paul Sander]

You wouldn't be running out of swap space, would you?  What happens if you
reduce the number of processes running on that machine, increase the size
of the swap partition, and clean out any swapfs filesystems you might have?

RCS has compilations options that trade high performance (memory mapped I/O
or loading complete files into allocated memory) vs. small foot print (make
many passes over large temporary files).  I took a (very) quick look at
the RCS-related code in CVS, and it appears that the RCS library does not
implement this trade-off; it loads the entire RCS file into in-memory data
structures.  Unless this trade-off is implemented, it seems you'll have
to supply sufficient swap space to process the largest of your files.

--- Forwarded mail from address@hidden

One of my users complained she got this message whenever she checked out or 
updated in her module.  I found that she had committed two of her binaries; 
one was 11Mb, the other was 7Mb.  When I removed those files, the 
checkout/update worked. 

I am running cvs 1.11 pserver with "-T /tmpcvs", where /tmpcvs is a 3Gb 
filesystem.  There is 2Gb of RAM on the server, which is HP 10.20.  /tmp is 
1.1Mb (but I hope it isn't being used).  The "xxx bytes" that can not be 
reallocated seems to be fixed depending on what file I am trying to checkout. 
 For the file of size 11,456,355, xxx is 21968.  For the file of size 
7094501, xxx is 14262.

Funny thing is, there are larger files that I *CAN* check out without a 
problem.  What appears to be a distinguishing feature is that they only have 
a single revision in the tree (1.1).  In fact, I can checkout 1.1 of the 2 
files that gave me trouble above.

So I'm guessing that CVS is running out of something while trying to 
reconstruct revision 1.4 (in the above example).  Is it really memory, or 
could it be temp disk space (which I was hoping would be /tmpcvs, as 
specified on the inetd.conf command line, and not /tmp)?  This might sound 
like a stupid question, but I thought I remember a problem with the size of 
the history file that also caused this message to appear.  Wait.  I seem to 
remember something about the maximum amount of memory an application is 
allowed to request.....

--- End of forwarded message from address@hidden