Memory Leak - rsync/libata/smbmount?

onewheelskyward · #1 25th March 2005, 06:09 PM

I have an FC3 server that's been running fine for months. I made some changes to it's role recently (it now does full backups of our Windows PCs), and if I leave a backup running long enough I can peg the physical memory at 1 Gig of RAM usage within 24 hours. I have had several hard lock ups in the last few days.

Here's a summary of the changes:

Updated to 2.6.11.5 kernel
Installed a Promise SATA Non-RAID controller
Installed 4 drives on SATA controller in a software RAID-5
I've been running rsync against smbmounted drives for hours at a time.

Does anyone know if either libata or sata_promise has any documented memory leaks that would cause this problem? I suspect rsync and smbmount too since I've been syncing around 50 gigs of data per run. MD is the last thing on my list because I think it's been the most tested out of all of the new elements on my machine.

Can anyone point me to a memory profiler that allows me to view my current RAM in a useful way? I've checked out valgrind and others, but they're all application specific. I may try using it on my next rsync just to see if I can figure anything out.

onewheelskyward · #2 26th March 2005, 12:35 AM

For what it's worth, here's a dump of free and meminfo:

Code:

[user@localhost]$ free
             total       used       free     shared    buffers     cached
Mem:       1017520    1004836      12684          0      40100     589308
-/+ buffers/cache:     375428     642092
Swap:      2031608        240    2031368

[user@localhost]$ cat /proc/meminfo
MemTotal:      1017520 kB
MemFree:         13312 kB
Buffers:         39396 kB
Cached:         589536 kB
SwapCached:          0 kB
Active:         438616 kB
Inactive:       498516 kB
HighTotal:      113852 kB
HighFree:          144 kB
LowTotal:       903668 kB
LowFree:         13168 kB
SwapTotal:     2031608 kB
SwapFree:      2031368 kB
Dirty:          197816 kB
Writeback:           0 kB
Mapped:         417768 kB
Slab:            33332 kB
CommitLimit:   2540368 kB
Committed_AS:   705980 kB
PageTables:       4316 kB
VmallocTotal:   114680 kB
VmallocUsed:     10696 kB
VmallocChunk:   103300 kB
HugePages_Total:     0
HugePages_Free:      0
Hugepagesize:     4096 kB

It looks like more than half of RAM is dedicated to caching, which makes sense if you consider the amount of file xfers I've been doing. However, with 40% of RAM being 'inactive', why hasn't anything hit swap?

Jman · #3 27th March 2005, 04:24 AM

The only thing that indicates a problem of what you said is the lockups. What is going on when that happens, the backup?

You have enough memory that I'm not surprised swap is unused.

onewheelskyward · #4 5th April 2005, 12:03 AM

The lockups were occuring when I was running more than one rsync (each one runs 3 processes). I turned off the Preemptible Kernel and Kernel IRQ Balancing, and my system is now stable enough to use. I'm glad it didn't have anything to do with the SATA driver, that one had me a bit worried.