Hi everone (Ingo ? :-)
I experience severe problems with the above raid patches and putting swap on a raid1
array (Pentium 233 MMX, 64 MB RAM, 2940UW, 2 UW IBM DDRS 4 GB scsi disks). The system
crashes consistently under heavy load (bonnie, exorcist etc.) while running fine for 3
days under (almost) no load during installation. My disk layout is as follows:
sd[a|b]1 10 mb dos,
sd[a|b]2 70 mb swap (/dev/md0)
sd[a|b]3 the rest for one big root filesystem (/dev/md1).
After the first crash, even the (necessary) background reconstruction brought the
system down.
My first guess were scsi problems but after I changed almost everything on the
hardware side (controller, cabling, drives, half dozen of Dougs aic7xxx drivers from
stock 2.0.35 over pre 10, pre 16, 5.1.0 and 5.1.1 ) I think it might be a bug in the
new raid code. What leads me to this assumption is the fact, that as soon as I turn
off the swap space, everthing works like a charm (background reconstruction several
times, bonnie with 3GB etc.). Is it possible that there is some memory leak or
something like that (due to the rewriting for linked lists ?) because short heavy load
is fine (e.g. reconstruction of the relativly small swap partition when it itself is
online), but under longer heavy load it takes about 1 minute (independently of what is
actually done: reconstruction, bonnie, exorcist etc.) to crash the system. The only
problem with that theory is that every crash is different ;-): sometimes it seg faults
some apps (bonnie, sendmail) and finally kswapd, sometimes it gives me scsi errors and
the worst is rapidly scrolling errors (yes I tried the panic_on_abort of the aic7xxx
driver with no luck). I don't know how to provide you with better technical
information about the crash, because I never had such problems with soft-raid and
linux before ;-). If you need some info, please let me know ...
Does anybody had the same experience or can tell me from his/her experience, that swap
on raid1 works with the newest patch ? Any help/hints appreciated.
Thanks
Urban
P.S.: By the way: I can reproduce a crash with the following steps: turn swap off,
dynamically remove one mirror and add it again to the array to force a full
reconstruction (~4GB in my case, hot add/remove in general works very nice here!),
turn swap on on /dev/md0 and repeat the last step to crash the system
