Menu
▾
▴
Re: [Evms-devel] Possible to create a degraded RAID5 array with EVMS?
|
From: James L. <jam...@ho...> - 2005-11-17 23:57:16
|
Well it looks like it's the fact that I'd set the partitions to raid
autodetect rather than the standard Linux patition type that's started
causing problems then...
Also, as someone's pointed out in this thread, the version of EVMS (2.5.2)
which is shipped with Ubuntu is out of date. A quick look at CVS shows that
the raid5_mgr.c file has changed, with a fix to memory corruption when
running a degraded RAID5 array. So I guess it's possible this is in fact
now fixed. If not, then you'd hope that this would be dealt with a little
more gracefully by EVMS... I'll see if I can run with those extra diags if I
have time (am a bit swamped at work, sorting out this sort of thing!).
It might be a good idea to document more clearly the fact that EVMS doesn't
like linux raid autodetect arrays: I certainly didn't spot it, and doing a
text search in the user guide for "autodetect", or for "0xfd" doesn't come
up with anything...
Anyway thanks for the help, and I'll let you know how I get on.
James
>From: Steve Dobbelstein <st...@us...>
>To: "James Lee" <jam...@ho...>
>CC: evm...@li...
>Subject: Re: [Evms-devel] Possible to create a degraded RAID5 array with
>EVMS?
>Date: Thu, 17 Nov 2005 10:16:06 -0600
>
>"James Lee" <jam...@ho...> wrote on 11/16/2005 06:08:14 PM:
>
> > Thanks for the reply Steve.
> >
> > I've retried setting up a degraded RAID5 array as described below, and
>have
> > hit a different error in EVMS ("*** glibc detected *** free(): invalid
>next
> > size (fast): 0x0819d720 ***"). Procedure was:
> >
> >
> > 1) Delete all md arrays and all segments from /dev/sdb (using EVMS).
> > Worked fine.
>
>Yea.
>
> > 2) Close and reopen EVMS to check all is still well. It is. Close it
>down
> > again.
>
>Good.
>
> > 3) Using cfdisk, create two logical (not primary) partitions on
>/dev/sdb
>
> > (which had just free space), 100MB each (these are /dev/sdb5 and
>/dev/sdb6).
> > Mark them as type 0xfd (Linux RAID autodetect). Save changes and
>again
>
> > reload and close EVMS to check it's still happy; which it is.
>
>Don't set the partition type to RAID autodetect. Use the normal Linux
>partition type. The kernel MD autodetect conflicts with EVMS's discovery
>and activation. Somewhere in the documentation (I can't find it at the
>moment) it should say that you should disable RAID autodetect if you are
>going to use EVMS to manage your software-RAID arrays.
>
> > 4) Use mdadm to create a degraded RAID5 array using the following
>command:
> > "mdadm --create /dev/md0 --level=5 --raid-devices=3 /dev/sdb5 /dev/sdb6
> > missing"
>
>Looks good.
>
> > 5) mdadm report that this array gets created successfully (in a
>degraded
>
> > state):
> > james@ubuntu-fileserver:~$ sudo mdadm --detail /dev/md0
> > /dev/md0:
> > Version : 00.90.01
> > Creation Time : Wed Nov 16 23:05:41 2005
> > Raid Level : raid5
> > Array Size : 192512 (188.00 MiB 197.13 MB)
> > Device Size : 96256 (94.00 MiB 98.57 MB)
> > Raid Devices : 3
> > Total Devices : 2
> > Preferred Minor : 0
> > Persistence : Superblock is persistent
> >
> > Update Time : Wed Nov 16 23:05:41 2005
> > State : clean, degraded
> > Active Devices : 2
> > Working Devices : 2
> > Failed Devices : 0
> > Spare Devices : 0
> >
> > Layout : left-symmetric
> > Chunk Size : 64K
> >
> > UUID : 6f0fa83a:ae0b4a99:a199ce36:9e21815c
> > Events : 0.4
> >
> > Number Major Minor RaidDevice State
> > 0 8 21 0 active sync
>/dev/.static/dev/sdb5
> > 1 8 22 1 active sync
>/dev/.static/dev/sdb6
> > 2 0 0 - removed
> >
> > 6) Start EVMS GUI. Gives me a warning about /md/md0 being in a
>degraded
>
> > state (which is fine). It doesn't give me the option to create an EVMS
> > volume from the /md/md0 array (which is not good). It sees the array,
>and
> > reports it as both degraded and corrupted (it shouldn't be, right?).
>
>Correct. EVMS should discover the array is degraded, but it should not be
>corrupt.
>
> > 7) Close down EVMS and restart it. Again, I get a warning about the
>array
> > being degraded. Now I do have the option to create a volume from the
>array.
> > On trying to do this however, I get the following error message:
> > *** glibc detected *** free(): invalid next size (fast): 0x0819d720 ***
> > and EVMS hangs completely.
> >
> > I've attached a gzipped copy of the evms-engine.log file.
> >
> > Hope this helps...
>
>It does. The last thing in the log is a call to raid5_free_private_data()
>in the MD plug-in. That function has a bunch of calls to free memory. I
>suspect one of those is going bad. It is not apparent from the code which
>one is bad. In fact, for each call to free memory, the code checks for a
>NULL pointer first, frees the non-NULL pointer, and then sets the pointer
>to NULL. Looks safe to me. My guess is there is some code elsewhere that
>has already freed some memory but has not set the pointer to NULL,
>resulting in a double free.
>
>EVMS has some memory debugging code, but it has to be built in. Can you do
>me the favor of another run? Rebuild EVMS with the debug support.
>./configure --with-debug <your-other-options>
>make clean install
>
>Then run evmsgui with the debug level set to "everything", e.g., "evmsgui
>-d everything". That will generate a rather large log file since it will
>log *everything*, which includes all calls to allocate and free memory.
>I'm hoping the memory allocation debugging code will catch a double free
>and that the log will contain the trace information to find out who did it.
>
> > What is the relationship between the EVMS MD plugin and mdadm by the
>way?
>
> > Do they use the same code, or are they entirely different programs that
> > interface in their own way with the Linux RAID driver? Should arrays
> > created by mdadm and array created by EVMS be identical once they have
>been
> > created?
>
>EVMS and mdadm are entirely different programs that interface in their own
>way with the Linux RAID driver. Naturally, they share common knowledge
>about the on-disk data structures and the behavior of the MD code in the
>kernel. Arrays created by mdadm should be understood by EVMS and vice
>versa.
>
>Steve D.
>
>
>
>-------------------------------------------------------
>This SF.Net email is sponsored by the JBoss Inc. Get Certified Today
>Register for a JBoss Training Course. Free Certification Exam
>for All Training Attendees Through End of 2005. For more info visit:
>http://ads.osdn.com/?ad_id=7628&alloc_id=16845&op=click
>_______________________________________________
>Evms-devel mailing list
>Evm...@li...
>To subscribe/unsubscribe, please visit:
>https://lists.sourceforge.net/lists/listinfo/evms-devel
|
