Skip to contentSkip to conversation

nocontrold: Eliminating ocfs2_controld v4

PATCH New Goldwyn Rodrigues 4 participants 8 messages Oct 18, 2013 – Nov 4, 2013 OCFS2

lore
Share & tools
lore
#0 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

Hi,

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

This feature requires modification in the userspace ocfs2-tools. The changes can be found at:
https://github.com/goldwynr/ocfs2-tools branch: nocontrold Currently, not many checks are present in the userspace code, but that would change soon.

These changes were developed on linux-stable 3.11.y. However, the changes are applicable to the current upstream as well. If you wish to give the entire kernel a spin, the link is:

https://github.com/goldwynr/linux-stable branch: nocontrold

Changes since v3:
* Added version negotiation using DLM lock
* Used strlcpy() instead of memcpy()

Changes since v2:
* Joel's comments: patch re-factoring

Changes since v1:
* Backward compatibility with ocfs2_controld

Signature · 1 line
Goldwyn
#1 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

Quoted 23 lines · On Fri, 18 Oct 2013 09:44:54 -0500 Goldwyn Rodrigues <[email protected]> wrote:

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

This feature requires modification in the userspace ocfs2-tools. The changes can be found at:
https://github.com/goldwynr/ocfs2-tools branch: nocontrold Currently, not many checks are present in the userspace code, but that would change soon.

Quoted 5 lines

These changes were developed on linux-stable 3.11.y. However, the changes are applicable to the current upstream as well. If you wish to give the entire kernel a spin, the link is:

https://github.com/goldwynr/linux-stable branch: nocontrold

I queued this up so it will get some linux-next exposure when Stephen gets back to his desk. But I don't feel I can take it further without suitable input from the other ocfs2 developers (please).

maintainer
#2 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

Quoted 29 lines · Andrew Morton
Quoted 27 lines · On Fri, 18 Oct 2013 09:44:54 -0500 Goldwyn Rodrigues <[email protected]> wrote:

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

This feature requires modification in the userspace ocfs2-tools. The changes can be found at:
https://github.com/goldwynr/ocfs2-tools branch: nocontrold Currently, not many checks are present in the userspace code, but that would change soon.

These changes were developed on linux-stable 3.11.y. However, the changes are applicable to the current upstream as well. If you wish to give the entire kernel a spin, the link is:

Quoted 5 lines
Quoted 1 line

https://github.com/goldwynr/linux-stable branch: nocontrold

I queued this up so it will get some linux-next exposure when Stephen gets back to his desk. But I don't feel I can take it further without suitable input from the other ocfs2 developers (please).

I thought I'd been pretty clear on the previous rounds. The code had some significant issues in its approach to compatibility (it wasn't). I haven't had a chance to look at this round yet, but I intend to soon.

Please do not forward this code without an explicit Acked-by from Mark or I.

Joel

Signature · 6 lines
Viro's Razor:
	Any race condition, no matter how unlikely, will occur just
	often enough to bite you.

			http://www.jlbec.org/
			jlbec at evilplan.org
#3 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

Hi Joel/Mark,

Quoted 36 lines · Joel Becker
Quoted 29 lines · Andrew Morton
Quoted 27 lines · On Fri, 18 Oct 2013 09:44:54 -0500 Goldwyn Rodrigues <[email protected]> wrote:

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

This feature requires modification in the userspace ocfs2-tools. The changes can be found at:
https://github.com/goldwynr/ocfs2-tools branch: nocontrold Currently, not many checks are present in the userspace code, but that would change soon.

These changes were developed on linux-stable 3.11.y. However, the changes are applicable to the current upstream as well. If you wish to give the entire kernel a spin, the link is:

Quoted 5 lines
Quoted 1 line

https://github.com/goldwynr/linux-stable branch: nocontrold

I queued this up so it will get some linux-next exposure when Stephen gets back to his desk. But I don't feel I can take it further without suitable input from the other ocfs2 developers (please).

Quoted 7 lines

I thought I'd been pretty clear on the previous rounds. The code had some significant issues in its approach to compatibility (it wasn't). I haven't had a chance to look at this round yet, but I intend to soon.

Please do not forward this code without an explicit Acked-by from Mark or I.

Did you get a chance to review this?

Signature · 1 line
Goldwyn
#4 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

In reply to Joel Becker
Quoted 32 lines · Joel Becker
Quoted 31 lines · Andrew Morton
Quoted 28 lines · On Fri, 18 Oct 2013 09:44:54 -0500 Goldwyn Rodrigues <[email protected]> wrote:

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

This feature requires modification in the userspace ocfs2-tools. The changes can be found at:
https://github.com/goldwynr/ocfs2-tools branch: nocontrold Currently, not many checks are present in the userspace code, but that would change soon.

These changes were developed on linux-stable 3.11.y. However, the changes are applicable to the current upstream as well. If you wish to give the entire kernel a spin, the link is:

https://github.com/goldwynr/linux-stable branch: nocontrold

Quoted 7 lines
Quoted 3 lines

I queued this up so it will get some linux-next exposure when Stephen gets back to his desk. But I don't feel I can take it further without suitable input from the other ocfs2 developers (please).

I thought I'd been pretty clear on the previous rounds. The code had some significant issues in its approach to compatibility (it wasn't). I haven't had a chance to look at this round yet, but I intend to soon.

FYI this round seems to include backwards compat code.

Quoted 2 lines

Please do not forward this code without an explicit Acked-by from Mark or I.

By forward I assume you mean to Linus? I agree this doesn't want to go upstream yet but since the backwards compat code is in now I think it's probably a good thing that this gets tested in linux-next.

	--Mark

--
Mark Fasheh

#5 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

In reply to Goldwyn Rodrigues
Quoted 6 lines · Goldwyn Rodrigues

Hi,

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

Thanks again for doing this work. I'm about halfway done with the patches when I write this but things seem to be coming along well.

Quoted 11 lines

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

Can you give us some brief details on how backwards compatibility was tested? I have a feeling that it would alleviate some concerns we had about that when the 1st series hit ocfs2-devel.

	--Mark

--
Mark Fasheh

#6 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

Quoted 21 lines · Mark Fasheh
Quoted 6 lines · Goldwyn Rodrigues

Hi,

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

Thanks again for doing this work. I'm about halfway done with the patches when I write this but things seem to be coming along well.

Quoted 9 lines

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

Quoted 6 lines
Quoted 2 lines

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

Can you give us some brief details on how backwards compatibility was tested? I have a feeling that it would alleviate some concerns we had about that when the 1st series hit ocfs2-devel.

I ran the code with an unmodified ocfs2-tools/libdlm and it worked fine and I got the expected message of upgrading dlm/ocfs2-tools.

I checked the reverse (older kernel, newer tools) as well, and mount returned ESRCH because ocfs2_controld was not running.

Signature · 1 line
Goldwyn
#7 lore raw

cover nocontrold: Eliminating ocfs2_controld v4

Quoted 29 lines · Goldwyn Rodrigues
Quoted 21 lines · Mark Fasheh
Quoted 6 lines · Goldwyn Rodrigues

Hi,

This is an effort of removing ocfs2_controld.pcmk and getting ocfs2 DLM handling up to the times with respect to DLM (>=4.0.1) and corosync (2.3.x). AFAIK, cman also is being phased out for a unified corosync cluster stack.

Thanks again for doing this work. I'm about halfway done with the patches when I write this but things seem to be coming along well.

Quoted 9 lines

fs/dlm performs all the functions with respect to fencing and node management and provides the API's to do so for ocfs2. For all future references, DLM stands for fs/dlm code.

The advantages are:
+ No need to run an additional userspace daemon (ocfs2_controld)
+ No contrrold devince handling and controld protocol
+ Shifting responsibilities of node management to DLM layer

Quoted 6 lines
Quoted 2 lines

For backward compatibility, we are keeping the controld handling code. Once enough time has passed we can remove a significant portion of the code.

Can you give us some brief details on how backwards compatibility was tested? I have a feeling that it would alleviate some concerns we had about that when the 1st series hit ocfs2-devel.

Quoted 5 lines

I ran the code with an unmodified ocfs2-tools/libdlm and it worked fine and I got the expected message of upgrading dlm/ocfs2-tools.

I checked the reverse (older kernel, newer tools) as well, and mount returned ESRCH because ocfs2_controld was not running.

Ok awesome - put this in your next 0/N e-mail please :) It's totally the type of thing I'd love to see up front.

Thanks,

	--Mark
Quoted 4 lines
Signature · 2 lines
-- 
Goldwyn

--
Mark Fasheh