[pve-devel] [PATCH 1/2] This patch refactors ZFS's LUN management code, removing the existing LunCmds implementations, in favor of a single lun management infraestructure which just invokes a 'management helper' (which can be either an script or a nat

[Pablo Ruiz]

Hello,

As a followup to this, as promised I've made some tests against a few ssh
servers (with diferent HW/OS specs) and it looks like there's plenty of
room for environment variables.. This are my findings regarding maximun
command line string which can be passed to ssh sucessfully.


   - CentOS-6.x/x86-64 => 256k
   - CentOS-5.x/x86-64 => 128k
   - Debian-Wheezy/x86-64 => 128k
   - Old-Fedora/Sparc64 => 256k
   - Solaris 10/Sparc64 => 256k
   - Linux-dd-wrt/arm => 128k

My tests where done my issuing the following command => ssh
root at 172.21.204.103 "DUMMY=$(perl -e 'print "a"x1024') ls", increasing the
length of 'a' string until reaching the point at which the server would
refuse to execute the command with: Argument list too long

Nontheless, I am still keen on not exposing all proxmox's configuration
settings (which looks a bit of a kludge) and only expose a selected subset
as PMXVARs.

Regards
Pablo



On Wed, Mar 19, 2014 at 10:42 AM, Pablo Ruiz <pablo.ruiz at gmail.com> wrote:

> Hi,
>
> While using command line arguments may seem the obvious approach, in the
> end this is a more fragile mechanism to pass data to helper script, as
> backward/forward compatbility would be harder to achieve once new 'data' is
> to be passed.
>
> Command line arguments need to respect ordering, or define strict argument
> names which should be parsed by the helper by means of getopt or similar
> method. However, environment is a bit more flexible, and we can add more
> variables as needed, just as CGI, AGI, OCF, and other similar protocol do.
>
> Regarding the size limits of environment/ssh-full-command-string, I will
> try to make a few test over various environments and get back to the list.
> If notable restrictions are found, I can restrict the variables passed to
> only those really needed.. instead of passing all config parameters.
>  On Mar 19, 2014 7:17 AM, "Daniel Hunsaker" <danhunsaker at gmail.com> wrote:
>
>> > Just curious - how long are the command lines this patch generates?
>>
>> Well, it looks like that depends on a number of things.  The entire
>> contents of %$scfg are added to the environment as PMXCFG_*, the full path
>> to the ssh key is added to the environment as PMXVAR_SSHKEY (and then added
>> again as an argument to ssh, if ssh is used), and depending on the command
>> being issued, you either get PMXVAR_LUNDEV or PMXVAR_LUNUUID, set to the
>> appropriate value.  So potentially quite long - certainly longer than I
>> expected on my first glance through the code.
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.proxmox.com/pipermail/pve-devel/attachments/20140320/9f065bb5/attachment.htm>