#181 Deadlock under load

3.6.2
open
5
2005-09-05
2005-09-05
Anonymous
No

I have been trying to get the 3.6.2 production release
to work relyably on a Debian 2.4.25 kernel built from
debian source using the config file from the standard
2.4.25-1-686 binary kernel. All works, but when I put
the file system under load I get a huge number of
messages (in messages)

starting with

Sep 4 09:58:23 chain kernel: iSCSI: session d074a000
sending mgmt 1280652 abort for itt 1280615 task
c7970a28 cmnd d270e600 cdb 0x2a to (1 0 0 0) at 7327783
Sep 4 09:58:23 chain kernel: task management command: 81

in batches of about 40, every 30 seconds,

The disc throughput is 0 when this is going on

after several hours of this, the kernel deadlocks (have
to turn the power off and on to get it back)

The configuration for the initiator is as the
distribution with just the target defined,

the test that fails is a default run of bonnie++
also mongo fails

however

dbench (the smb test suite) runs fine with 5 clients. I
suspect because it is not exercising the file system
quite as hard ?

Is this a bug, or am I missing something ?

---

The target I am using for this test is the Ardis Target
(20040211) which is exposing a loopback file system of
5GB. I have tested the Loopback device directly without
any problems. Als the tests are fine direct to disk

so thats
/dev/hda5 -> ext3 tests ok
file->loopback (/dev/loop0) -> ext3 tests ok
file->loopback (/dev/loop0) -> ardis target -> iscsi
initiator (/dev/sda) -> partition (/dev/sda1) -> ext3 fails

The test I am using is

Discussion

  • Logged In: NO

    When you get those error messages does the system respond to
    pings or is there any other network activity or does that
    die too?

     
  • Logged In: NO

    I dont think it responds to pings, I'm logged in at the
    console, and none of the normal lights on the keyboard work
    (numlock, capslock etc), which normally means something is
    really locked up.