Yesterday I stepped on an old behavior of the kernel I was not aware of. I wonder
if other have experienced the problem it is causing to vservers. Here is what it
does.
You create a bunch of vserver, all using an IP (or more) address of the same IP
network. This is fairly common (typical in fact). Say you have the following vservers
A: 192.168.0.1/255.255.255.0
B: 192.168.0.2/255.255.255.0
C: ...
D: ...
Then you start all those vservers in the above order. Then you stop vserver A (or
a restart). When you do stop a vserver, the corresponding IP aliases are removed. In
the above example, the vserver script does
ifconfig eth0:A down
There is a catch here. eth0:A was the first IP aliases created by the kernel
for the 192.168.0.0/255.255.255.0 network. It sets a flag telling this is the
"main" IP number of this network. All the other IP aliases (created for B, C and D
vservers) will have the opposite flag IFA_F_SECONDARY.
Whenever the kernel drop a "main" IP aliases, all the related (secondary) IP
aliases (same network) are also dropped.
Said differently, you start A,B,C and D and the ifconfig reports
eth0:A, eth0:B, eth0:C and eth0:D
Then you do "vserver A stop" and the above IP aliases are gone. The vservers
are left not much functionnal. For sure, you can restore their functionnality simply
by doing
vserver B exec ls
vserver C exec ls
vserver D exec ls
( or vserver B enter, then exit).
------------
Now the behavior is not a bug in the kernel. It is a feature and has existed this
way for many years (I have seen post about this dating 2001). How should
we handle the problem. I know that for vserver 1.1 project, we may be using
a different strategy (virtual network device), but for upcoming vserver 1.0, we
need a solution as this behavior is quite unexpected by any sysadmin.
One solution for the vserver command would be to record the state of ifconfig,
perform the "ifconfig eth0:A down" operation and restore all the missing
IP aliases.
This solution is easy to implement, but weird at best. Further it is not atomic,
meaning
that 2 admins may be stopping 2 vservers at the same time and having one
"resurecting the IP alias" of the second.
--------------------
Another solution would be to change the kernel. For example, when we delete the
main IP alias of a network, instead of deleting the other, we remove the SECONDARY
flag from the first other IP alias in the list.
---------------------------------------------------------
Jacques Gelinas <[EMAIL PROTECTED]>
vserver: run general purpose virtual servers on one box, full speed!
http://www.solucorp.qc.ca/miscprj/s_context.hc
_______________________________________________
Vserver mailing list
[EMAIL PROTECTED]
http://www.solucorp.qc.ca/mailman/listinfo/vserver
