gdb on a node

Wade Hampton wade.hampton at
Mon Nov 11 07:19:51 PST 2002


I'm trying to debug a process on a remote node using Scyld
28CZ4.  For example, I was trying to get ganglia to work
on the cluster.

As PIDs appear on the head node, I figured that a gdb "attach"
would not work but it more or less did:

    bpsh 0 myprog

    ps eax | grep myprog

    get pid from that and then:

    bpsh 0 gdb my prog
    gdb> attach myprog's pid

This worked, but:

1)  It requires the program to be running first, possibly requiring changes
     to the program startup code.

2)  Typing ctrl-c following a gdb "continue" terminates the bpsh and the 
     not pausing gdb as one would expect.

What I found that does not work, but based on my understanding,
should work:

    copy file to an NFS partition which the nodes have mounted:


    build the program with -ggdb to enable debugging....

    from the head node:

        cd /data/xfer/myprog
        bpsh 0 gdb myprog

    At this point, gdb should be running on the remote node with
    stdio forwarded to me.  It should be debugging my program.

        break main    -- or set any other breakpoint that should be hit

    The program NEVER hits the beakpoint and runs to completion....

Any ideas?   Help would be most appreciated.

Wade Hampton

More information about the Beowulf mailing list