LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: [PATCH-2.4 0/2] Threshold Limitation & Overflow Server Pool

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: [PATCH-2.4 0/2] Threshold Limitation & Overflow Server Pool
Cc: Horms <horms@xxxxxxxxxxxx>
Cc: wensong@xxxxxxxxxxxx
From: Roberto Nibali <ratz@xxxxxx>
Date: Mon, 05 Dec 2005 15:24:30 +0100
Hi guys,

Thanks to Horms for the review, it's much appreciated.

Other than that, I wish someone could comment on it and give some
feedback. Oh, and of course you need the upgraded ipvsadm source. This
makes the following patchset:

[1/2]: the kernel part
[2/2]: the ipvsadm enhanced with the needed functionality
Hi Ratz,

I read over your patches a couple of times and I couldn't find anything
terribly disturbing in them. Here are some more general comments,
mostly minor.

* 80 columns wide (there seems to be one fragment that ever so slightly overflows)

I'll see to it.

* Can you break out the functionally different parts of the patches
  into separate patches, especially the hprio scheduler and the cleanups.

Most definitely.

* For upstream I think this will have to be a 2.6 patch

100% agreed.

* Seems like I can use your framework to add my quiescent/offline flags
  idea. Perhaps ip_vs_overflow should be ip_vs_available or something
  like that. Not a big deal, we can always change it later.

Which idea? Something that cannot be handled with the overflow pool?

* As far as I can see your code won't break binary compatibility with
  ipvsadm. But its not much use without an updated ipvsadm.

I'm afraid it does, so long as we check the the size of the *_user structs in the binary. TLV's could help here. I had to blow up the ip_vs_rule_user, ip_vs_dest_user and ip_vs_service_user structs, so without at least re-compiling ipvsadm, this is a no-go. This is a general problem when one relies on kernel includes for user space applications. Plus we have two mixed methods of represending/displaying IPVS related data in user space:

1. via proc-fs
2. via getsockopt

The breaking ABI could be solved if we gathered all information via proc-fs (which in itself is ugly), however we also need the kernel include to transport commands to the kernel, for which there is (currently) no other possibility, except netlink transport.

Wensong, are you processing patches for ipvsadm or should I ?

Guess he is.

Lastly, not strictly related, this mail is the last mail
for an unreviewed patch. If I've missed one, please let me know.

Nope, not that I would know of; besides I would have forgotten anyway :).

Responding to the orignial thread with me CCed should grab my attention.

I'll do that in future.

Best regards,
Roberto Nibali, ratz
--
-------------------------------------------------------------
addr://Kasinostrasse 30, CH-5001 Aarau tel://++41 62 823 9355
http://www.terreactive.com             fax://++41 62 823 9356
-------------------------------------------------------------
terreActive AG                       Wir sichern Ihren Erfolg
-------------------------------------------------------------

<Prev in Thread] Current Thread [Next in Thread>