| Re: Dangerous updates? [message #7740 is a reply to message #7357] |
Mon, 23 October 2006 16:10  |
rollinw
Messages: 25
Registered: June 2006
Location: Santa Barbara, California
|
Junior Member |
|
|
After spending quite a bit of time on this problem, I have managed to reproduce part of the problem. I have created a VPS that will not start up properly. In this case, the HN did not seem to be affected. Here is the sequence that has led to the defective VPSs:
1. Inside the SUSE VPS bring up the Yast (ncurses) inteface.
(This is a SuSE 9.3 SMP x86_64 kernel).
2. Find a mirror that has updates (RPM patches).
3. Run the online update application.
4. After yast compiles its list of updates, select those to be
installed. We chose 2 categories: recommended and security
5. Yast gives a warning that there is a kernel patch and asks if
this should be skipped. We chose to install.
6. During the installation of updates, Yast gives an error
regarding the kernel update, and at this point we chose the
Skip option. There were 45 update RPMs installed.
7. Yast appeared to complete the update successfully.
8. The VPS continues to run OK after the update until it is
stopped by vzctl.
9. From this point on, the VPS does not start properly. Vzctl
start does not give an error message, but only a skeletion
system comes up. It appears that the boot process does not
complete correctly. In particular, init 3 does not start
up the processes in /etc/init.d One can enter the VPS and
start the processes manually, including the system logger,
the network, sshd, etc.
I have gone through the Yast logs and the /var/log/messages file,
but so far I have only part of the answer. Yast does attempt to
install a version of the Suse kernel. It creates a /boot directory and attempts to write a vmlinux kernel file and an initrd. This is the point at which it fails. But it does install a new kernel RPM, which is older than the version running on the HN.
Besides the kernel install problem, there seems to be a problem resulting from the linux-utils RPM. This is suggested from the time stamp when this was installed. Here are the contents of the messages log that show the failure(s) that occurred:
------------------------
Oct 13 21:54:07 vpssuse1 usermod[32604]: default group changed - account=nobody, uid=65534, gid=65533, old gid=65533, by=0
Oct 13 21:54:13 vpssuse1 groupadd[32681]: group already exists - group=haldaemon, by=0
Oct 13 21:54:13 vpssuse1 useradd[32682]: account already exists - account=haldaemon, by=0
Oct 13 21:54:16 vpssuse1 init: Trying to re-exec init
Oct 13 21:54:16 vpssuse1 init: no more processes left in this runlevel
Oct 13 21:54:20 vpssuse1 init: Trying to re-exec init
Oct 13 21:54:20 vpssuse1 init: no more processes left in this runlevel
Oct 13 21:54:37 vpssuse1 groupadd[1422]: group already exists - group=sshd, by=0
Oct 13 21:54:37 vpssuse1 useradd[1423]: account already exists - account=sshd, by=0
Oct 13 22:00:01 vpssuse1 run-crons[1607]: mcelog returned 1
Oct 13 22:01:47 vpssuse1 syslog-ng[19545]: new configuration initialized
--------------------------------
I am trying to understand how vzctl starts a VPS. I examined quite a bit of the code and discovered that some time after vzctl mounts the VPS, it calls a routine, execvp() that at some point calls the /sbin/init inside the VPS. I have not been able to discover the other steps in the VPS startup.
Repeating my earlier comment, this update does not appear to have damaged the HN, because I can still create new VPSs and start them correctly.
Rollin
|
|
|
|
OpenVZ Forum
Members
Search
Help
Register
Login
Home
