|
Can anyone confirm these findings? This problem is getting really obnoxious,
and I'd love to hear if I'm not the only one who has ever seen it. I've
written locking mechanisms into scripts that attempt to keep from doing the
same operation twice to try to circumvent the bug, but those methods aren't
perfect. ... and if one of them happens to give way and take down a node
twice... A press from the reset button on a running Linux machine with a 13
GB drive... and I suddenly find myself in single-user mode with my drive
mounted read-only doing an fsck and hoping that no important data was lost.
:(
Any ideas?
All the best --
Ted Pavlic
tpavlic@xxxxxxxxxxx
----- Original Message -----
From: Ted Pavlic <tpavlic_list@xxxxxxxxxxx>
To: Linux Virtual Server Mailing List <linux-virtualserver@xxxxxxxxxxxx>
Cc: <wensong@xxxxxxxxxxxx>
Sent: Tuesday, July 06, 1999 12:24 PM
Subject: Bug in LVS-0.6
> I think I found a bug in LVS-0.6 for 2.2.10. It looks like when someone
> deletes a real server from the hash, the actual entry that says "hey, I'm
an
> existing real server" isn't deleted. (I haven't had time to look at the
> code, but I'm guessing this is the case)
>
> Try this: (SYNC first! And be sure you don't have processes in the
> background. This may cause you to need to reset your computer.) (Feel free
> to change the IP addresses)
>
> sync
> ipvsadm -C
> ipvsadm -A -t 10.200.0.15:80 -s rr
> ipvsadm -a -t 10.200.0.15:80 -r 216.69.192.196 -g
> ipvsadm -d -t 10.200.0.15:80 -r 216.69.192.196
> ipvsadm -d -t 10.200.0.15:80 -r 216.69.192.196
>
> At that point, if things work correctly, I should get this:
>
> setosckopt failed: No such file or directory
> No such destination
>
> Which is fine. That's what I get if I delete any real server that never
> existed in the first place. However, what I *ACTUALLY* get is this:
>
> IP_VS: ip_vs_del_dest(): destination not found!
> setosckopt failed: No such file or directory
> No such destination
>
> And then the machine locks up. It looks like init still is functioning, as
a
> CNTRL+ALT+DELETE does kick it into runlevel 6, however... it doesn't get
> very far into runlevel 6 until it completely locks up.
>
> Let me know if anyone else can confirm these results.
>
> Try:
>
> adding a virtual server
> adding a real node
> deleting that real node
> deleting it again
>
> If there isn't a bug, then it'll just act as if you tried to delete a
server
> that never existed... but if there is a bug... then your machine will
> lockup.
>
> All the best --
> Ted Pavlic
> tpavlic@xxxxxxxxxxx
>
>
>
>
|
