Subject: | |
From: | |
Reply To: | ~Stack~ |
Date: | Thu, 9 Mar 2017 20:03:42 -0600 |
Content-Type: | multipart/signed |
Parts/Attachments: |
|
|
Greetings,
There has been some discussion in my group about setting something up to
capture kdump files in a single location. Never messed with it before,
but it seems pretty straight forward as /etc/kdump.conf is very well
documented.
My question: What is the best way to compress for sending to remote scp
in a kdump?
Longer version and configuration:
Set up a couple of SL7 virtual machines to start testing.
Default configuration:
$ sed -e '/^#/d' /etc/kdump.conf
path /var/crash
core_collector makedumpfile -l --message-level 1 -d 31
My changes:
$ sed -e '/^#/d' /etc/kdump.conf
ssh [log in to unmask]
sshkey /root/.ssh/kdump_id_rsa
path /data/kdump_crash
core_collector scp
$ kdumpctl propagate
$ systemctl restart kdump
Whoo! It works...Huh...My crash seems to take a long time...Oh, look at
that, my vmcore is 4GB in size (vm memory size). o_O
I really don't think my other crashes are nearly that large...Reset to
default config and force a crash. 39MB. Ah, the '-l' option of
makedumpfile uses LZO compression. Makes sense.
Looking at the kdump.conf man page, I should be able to use makedumpfile
to compress then send to scp.
$ sed -e '/^#/d' /etc/kdump.conf
ssh [log in to unmask]
sshkey /root/.ssh/kdump_id_rsa
path /data/kdump_crash
core_collector makedumpfile -F -l --message-level 1 -d 31
Hrm. It will only work if I specify '-F' which gives me a 'vmcore.flat'
instead of a normal 'vmcore' which my crash tools don't seem to like.
Looks like I can do conversions from a vmcore.flat file to use the debug
tools, but I don't like that.
I've tinkered with a few other options unsuccessfully. My options look
like: uncompressed (which will be fun with my 512GB nodes!) or
vmcore.flat files.
Any thoughts on the best way to compress the vmcore file while still
sending over scp?
Thanks!
~Stack~
|
|
|