[Gluster-devel] GlusterFS core dump (v1.2.3)

[Ian Grimstead]

I think something's up - I have just uninstalled 1.2.3 on each node, then
obtained & built glusterfs--mainline--2.4; looks like you need to run
autogen.sh to produce the "configure" script?
 
Did that, installed... now I'm not getting any directories under my
glusterfs mount, apart from a single (and empty) "/home". So I have:
 
/media/glusterfs => gluster FS mount point
 
ls -al /media/glusterfs gives:
 
total 24
drwxr-xr-x  3 root root 4096 May 30 14:57 .
drwxr-xr-x  3 root root 4096 May 11 17:32 ..
drwxr-xr-x  4 root root 4096 Jun 11 13:46 home

But, ls -al /media/glusterfs/home gives:

total 0

Rather worrying. However, if I "cd" to a known directory, it works. This dir
is also empty. If I "cat" a known file, it's there!
 
Wierd. Log from client just shows "stat-prefetch:flush on: /" 3 times after
initial handshake success of servers...
 
I've turned off prefetch caching on the client - no difference. :o(
 
Any idea what could have caused this?
 
Ian

  _____  

From: anand.avati at gmail.com [mailto:anand.avati at gmail.com] On Behalf Of
Anand Avati
Sent: 23 June 2007 19:05
To: Ton van Rosmalen
Cc: I.J.Grimstead at cs.cardiff.ac.uk; gluster-devel at nongnu.org
Subject: Re: [Gluster-devel] GlusterFS core dump (v1.2.3)


Ton,
please upgrade to glusterfs--mainline--2.4-patch-182. even selinux
extensions should work now. just committed a heap overflow bug.

thanks!
avati


2007/6/23, Ton van Rosmalen < ton at netbase.nl>: 

Hi Avati,

Thanks for your attention.

I checked with glusterfs--mainline--2.4 and it shows the same behaviour in
my set-up.
My VIM version is 7.0.109 (on FC6). I must admit I hadn't tested the
permissions yet a reported by Ian. I just checked and mode 0777 gives me a
core dump as well.
In fact, starting vi with a non-existing file dumps core.

Testing with vi(m) further.
vi -R <file>    - works ok
vi -n <file>     - works ok
    write file     - core dump

It looks like any write command from vi(m) crashes GlusterFS.

I checked for the extended attributes as you requested and it said that a
selinux-attribute existed on the file(s). 
I disabled SELinux, deleted all files from the volume and copied them again.


Now it works for me! I can edit any file without problems.
I guess these selinux-attributes aren't ready yet.

Thanks

Ton

Anand Avati schreef: 


Can you confirm that this still happens with a tla checkout from
'glusterfs--mainline--2.4' ? we have tested vim to be working. which version
of vi have you tried with? also is there a confirmed pattern in the file
permission/ownership which is causing this? since the coredump seems to be
in unify_getxattr_cbk, can you also check if the file has any extended
attributes from the backend fs? 

thanks!
avati


2007/6/23, Ton van Rosmalen <ton at netbase.nl>: 


    --- original message copied from archive ---
    Hi there,

    I get a (rather odd) repeatable bug; if I have a file with permissions
    -rw-r--r-- (644), and edit it using "vi", I get:

    E297: Write error in swap file
    E207: Can't delete backup file
    E138: Can't write viminfo file /home/scmijg/.viminfo!

    Next thing you know, glusterfs has core dumped. I've run it in 
    non-daemon
    mode, and "DEBUG" level - no messages, apart from along the lines of
    "glibc
    detected" and "free(): invalid next size (fast)"

    I dropped the core into gdb, ran a backtrace and it listed 
    "unify_getxattr_cbk" in "unify.so" as the culprit. Looks like pointer
    corruption?

    The config is 4 hosts, all hosting their local HDs apart from 1 machine
    which is hosting an external (USB) drive in addition. Everything running

    RHEL, all machines AMD Opteron dual-core...

    Anyone else had this trouble? Otherwise, with other access, Gluster
    seems
    OK... Very odd. Say, using "vi" with a file permissions of "-rwxr-xr-x" 
    works fine...

    Thanks for any help!

    Ian
    -----------------

Hi Ian,

I've got the same problem. I tested with v1.2.3 about 3 months ago and
re-tested with 1.3.0-pre4 this week. 

My config is running on "VM-Ware Linux-FC6" virtual machines. I've
created 3 servers and one client. Normal copy-operations work beautiful.
Using vi leads to the same problems as you're experiencing. 

For me this is a show-stopper. Basic file-editing is a must for me.
I've got no suggestion for a solution. Should I find something I'll post
it to the list.

Regards,

Ton




_______________________________________________
Gluster-devel mailing list
Gluster-devel at nongnu.org
http://lists.nongnu.org/mailman/listinfo/gluster-devel





-- 
Anand V. Avati 




-- 
Anand V. Avati