Hi
The way the hobbit init script works (at least on debian), a lot of processes are left running on the host when doing a "/etc/init.d/hobbit-client stop".
Is there any reason for this?
Cheers, Gildas
On Fri, Oct 06, 2006 at 01:19:52PM +0100, Gildas Le Nadan wrote:
The way the hobbit init script works (at least on debian), a lot of processes are left running on the host when doing a "/etc/init.d/hobbit-client stop".
Is there any reason for this?
You mean the vmstat processes ?
The reason is that they run pretty independent of everything else, just collecting data over a 5 minute period. They will terminate by themselves after at most 5 minutes; and if you're shutting down the system, they will get killed.
It's not that it has to be this way, but it was just the easiest way of doing it.
Regards, Henrik
Henrik Stoerner wrote:
On Fri, Oct 06, 2006 at 01:19:52PM +0100, Gildas Le Nadan wrote:
The way the hobbit init script works (at least on debian), a lot of processes are left running on the host when doing a "/etc/init.d/hobbit-client stop".
Is there any reason for this?
You mean the vmstat processes ?
Yes, but also df commands that seem to be stalled, such as:
ssgtest1a:~# date Fri Oct 6 13:36:34 BST 2006 ssgtest1a:~# ps aux | grep hobbit hobbit 21440 0.0 0.0 5620 1188 ? S 13:06 0:00 /bin/sh /usr/lib/hobbit/client/bin/hobbitclient-linux.sh hobbit 21456 0.0 0.0 2580 572 ? S 13:06 0:00 df -Pl -x none -x tmpfs -x shmfs -x unknown -x iso9660 hobbit 21457 0.0 0.0 2624 508 ? S 13:06 0:00 sed -e /^[^ ?][^ ?]*$/{?N?s/[ ?]*\n[ ?]*/ /?}
The reason is that they run pretty independent of everything else, just collecting data over a 5 minute period. They will terminate by themselves after at most 5 minutes; and if you're shutting down the system, they will get killed.
It's not that it has to be this way, but it was just the easiest way of doing it.
Regards, Henrik
One way is to modify/enhance hobbit init script to kill processes own by hobbit after the stop.
In Solaris, ie. kill ps -u hobbit | grep -v PID| awk '{print $1}'
T.J. Yang
From: Gildas Le Nadan <gn1 at sanger.ac.uk> Reply-To: hobbit at hswn.dk To: hobbit at hswn.dk Subject: Re: [hobbit] hobbit client init script Date: Fri, 06 Oct 2006 13:40:12 +0100
Henrik Stoerner wrote:
On Fri, Oct 06, 2006 at 01:19:52PM +0100, Gildas Le Nadan wrote:
The way the hobbit init script works (at least on debian), a lot of processes are left running on the host when doing a "/etc/init.d/hobbit-client stop".
Is there any reason for this?
You mean the vmstat processes ?
Yes, but also df commands that seem to be stalled, such as:
ssgtest1a:~# date Fri Oct 6 13:36:34 BST 2006 ssgtest1a:~# ps aux | grep hobbit hobbit 21440 0.0 0.0 5620 1188 ? S 13:06 0:00 /bin/sh /usr/lib/hobbit/client/bin/hobbitclient-linux.sh hobbit 21456 0.0 0.0 2580 572 ? S 13:06 0:00 df -Pl -x none -x tmpfs -x shmfs -x unknown -x iso9660 hobbit 21457 0.0 0.0 2624 508 ? S 13:06 0:00 sed -e /^[^ ?][^ ?]*$/{?N?s/[ ?]*\n[ ?]*/ /?}
The reason is that they run pretty independent of everything else, just collecting data over a 5 minute period. They will terminate by themselves after at most 5 minutes; and if you're shutting down the system, they will get killed.
It's not that it has to be this way, but it was just the easiest way of doing it.
Regards, Henrik
To unsubscribe from the hobbit list, send an e-mail to hobbit-unsubscribe at hswn.dk
From: Nicolas Dorfsman <ndo at unikservice.com> Reply-To: hobbit at hswn.dk To: hobbit at hswn.dk Subject: Re: [hobbit] hobbit client init script Date: Fri, 6 Oct 2006 15:27:43 +0200
Le 6 oct. 06 � 15:24, T.J. Yang a �crit :
One way is to modify/enhance hobbit init script to kill processes own by hobbit after the stop.
In Solaris, ie. kill
ps -u hobbit | grep -v PID| awk '{print $1}'pkill -u hobbit
Much elegant solution for solaris hobbit init script. Thanks for the tip.
tj
On Fri, Oct 06, 2006 at 01:40:12PM +0100, Gildas Le Nadan wrote:
Henrik Stoerner wrote:
On Fri, Oct 06, 2006 at 01:19:52PM +0100, Gildas Le Nadan wrote:
The way the hobbit init script works (at least on debian), a lot of processes are left running on the host when doing a "/etc/init.d/hobbit-client stop".
Is there any reason for this?
You mean the vmstat processes ?
Yes, but also df commands that seem to be stalled, such as:
"df" commands should not stall. Your system has some sort of problem if they do; the Linux client explicitly uses "df -Pl" to check only local disks (not network-mounted disks), and that should not stall.
Regards, Henrik
Gildas Le Nadan wrote:
Henrik Stoerner wrote:
On Fri, Oct 06, 2006 at 01:19:52PM +0100, Gildas Le Nadan wrote:
The way the hobbit init script works (at least on debian), a lot of processes are left running on the host when doing a "/etc/init.d/hobbit-client stop".
Is there any reason for this?
You mean the vmstat processes ?
Yes, but also df commands that seem to be stalled, such as:
ssgtest1a:~# date Fri Oct 6 13:36:34 BST 2006 ssgtest1a:~# ps aux | grep hobbit hobbit 21440 0.0 0.0 5620 1188 ? S 13:06 0:00 /bin/sh /usr/lib/hobbit/client/bin/hobbitclient-linux.sh hobbit 21456 0.0 0.0 2580 572 ? S 13:06 0:00 df -Pl -x none -x tmpfs -x shmfs -x unknown -x iso9660 hobbit 21457 0.0 0.0 2624 508 ? S 13:06 0:00 sed -e /^[^ ?][^ ?]*$/{?N?s/[ ?]*\n[ ?]*/ /?}
Okay, so I have found the culprit: we have lustre filesystems that hang sometime so the df command is stalled for a while.
I'd like to exclude lustre from the df command on some on the linux hosts only. Is it possible?
The only way I've found to exclude lustre from the df command is to add a "-x lustre_lite" to the df line in /usr/lib/hobbit/client/bin/hobbitclient-linux.sh
Cheers, Gildas
On Fri, Oct 06, 2006 at 04:43:01PM +0100, Gildas Le Nadan wrote:
Okay, so I have found the culprit: we have lustre filesystems that hang sometime so the df command is stalled for a while.
I'd like to exclude lustre from the df command on some on the linux hosts only. Is it possible?
The only way I've found to exclude lustre from the df command is to add a "-x lustre_lite" to the df line in /usr/lib/hobbit/client/bin/hobbitclient-linux.sh
I'm afraid that is the only way of doing that.
Regards, Henrik
Henrik Stoerner wrote:
On Fri, Oct 06, 2006 at 01:19:52PM +0100, Gildas Le Nadan wrote:
The way the hobbit init script works (at least on debian), a lot of processes are left running on the host when doing a "/etc/init.d/hobbit-client stop".
Is there any reason for this?
You mean the vmstat processes ?
The reason is that they run pretty independent of everything else, just collecting data over a 5 minute period. They will terminate by themselves after at most 5 minutes; and if you're shutting down the system, they will get killed.
It's not that it has to be this way, but it was just the easiest way of doing it.
A "kill ps -u hobbit -o pid | grep -v PID" in the initscript works as
well.
While the subject of initscripts is up, I've been using one locally for a while that integrates runclient.sh's actions directly into the code. It's been working well for me, and - combined with more regexing of BBHOME and redirect variables out of the .spec file - might be useful to some as a replacement.
Comments welcome :)
http://mirrors.redwire.net/pub/local-rpms/packages/hobbit/hobbit-client.init... http://mirrors.redwire.net/pub/local-rpms/packages/hobbit/hobbit.initscript.... http://mirrors.redwire.net/pub/local-rpms/packages/hobbit/hobbit.spec
Regards,
Japheth Cleaver, cleaver at redwire.net
participants (5)
-
cleaver@redwire.net
-
gn1@sanger.ac.uk
-
henrik@hswn.dk
-
ndo@unikservice.com
-
tj_yang@hotmail.com