HAMMER update 23-Feb-2008

[Matthew Dillon]

:I've got a panic when I tried to reboot the machine after running
:some stress tests.  I'm not sure if this one is easily reproducible, but
:here's the last message anyway:
:
:syncing disks... 6 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 
:done
:panic: assertion: node->lock.refs == 0 && node->ondisk == NULL in hammer_flush_buffer_nodes
:mp_lock = 00000000; cpuid = 0
:
:The kernel dump is on my leaf account as ~y0netan1/crash/{kernel,vmcore}.8 .

    Ah, excellent.  I've gotten that one a few times too.  I'll look at your
    core.  I've got a ton of assertions in that path and if things are not
    in exactly the state it expects it will panic.

    There may also be a syncing disks issue on reboot too which I haven't
    tracked down yet.  Sometimes it says 'syncing disk ...  blah blah
    blah blah 1 1 1 1 1 ... then gives up'.

    Both are related to hanging references.  The B-Tree node handling is
    especially complex since it tries really hard to passively cache
    pointers into the B-Tree so it doesn't have to start searches at the
    root every time.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>