[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[netCDF #PKT-462504]: Re: [netcdfgroup] netcdf-4 open/close memory leak

> OK, so I ran today's netcdf snapshot2010012708 on a linux (x86_64)
> box, too, and I still see memory increasing for tst_files2.  I ran 500
> cycles (output attached). (Note that for get_mem_used1, the output is
> for 'rss=' in place of 'size='.  The difference with using 'size=' is
> that the increases are less often but larger, e.g., about 30-40 zeros
> and then a 132 kB increase  I also tried it with the hdf5-1.8.4-snap7
> with same results.
> 
> I also verified zero memory increase with the open and close
> statements commented out. So I think I don't see any change from
> snapshot2010011908.  Hmmmmmm.
> 
> -- Ted

Ted,

I owe you one beer.

Turns out there were *two* memory growth problems, one for reading, one for 
writing. You had a hold of the read one, and Charles at LLNL had a hold of the 
write one. Now both are fixed.

I have also finished a review of all the write and read code to make sure that 
there are no other memory problems of this particularly insidious kind. These 
blasted varmints can't be detected by the two methods I am using. I am looking 
into adding some code that will check with the HDF5 library whether any of 
these things are open after a netcdf file is closed. Devilishly tricky little 
blighters!

But our current method of looking at the output of the ps command is now 
showing zero growth for reads and writes.

So please give the snapshot a try and see how it works for you. Make sure you 
get the one I just generated.

Thanks!

Ed



Ticket Details
===================
Ticket ID: PKT-462504
Department: Support netCDF
Priority: Critical
Status: Closed