bergman on 10 Aug 2011 09:20:32 -0700

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: [PLUG] ClusterSSH & friends

In the message dated: Wed, 10 Aug 2011 01:35:27 EDT,
The pithy ruminations from JP Vossen on 
<[PLUG] ClusterSSH & friends> were:
=> Last night at PLUG N we were talking about running the same command on a 
=> bunch of servers.  Here is some follow-up for that.

I use clusterssh on a daily basis. Prior to that I've used clusterm[1]
and the Solaris Cluster Server "cssh" package[2] when I admin'ed Solaris
and Veritas clusters.

Some Linux terminal emulators (roxterm? konsole?) have a "multiplexing" option
that allows you to send output from one terminal to multiple terminals, but
they don't natively integrate the ssh parts.

I've hacked^H^H^H^H^H^H written a front-end script to the ClusterSSH "cssh"
command to enable loops over specified hosts, rather than starting interactive

That front-end wrapper solves most of my complaints with clusterssh,
the only remaining issues are the hard-coded dependency on xterm as
the terminal emulator, rather than allowing a user-specified program,
and a minor bug that causes the Nth terminal window to be opened with
a smaller size than the others.


=> Then there is "Execute commands simultaneously on multiple servers Using 
=> PSSH/Cluster SSH/Multixterm" [2] and "Parallel SSH execution and a 
=> single shell to control them all" [3].

Yeah, I tried pssh for a while...I don't recall why I prefered clusterssh.

I actually like clusterm a lot, but it doesn't play well with recent versions
of perl.

There's a recent round-up of parallel ssh tools here:
=> And of course, CFEngine [4], Puppet [5] and chef [6] and a comparison of 

I use cfengine as well. It's slightly related to the ssh tools, but
only in the general sense of system management...there's a good case
for having both types of tools in any sufficiently complex environment.

In general, I use cssh to do "interactive" tasks on multiple servers in sync,
where those servers are not managed by cfengine. The cfengine boxes (19 at
present, soon to be ~90) don't get any interactive changes or management--the
changes are all applied to the cfengine server. I'll login to the cfengine
managed boxes only to make sure that the changes were applied as really, really, really don't want to mix administration through
a config management system and interactive (manual) management.

=> If anyone (Kevin?) dives into these, let us know how it works out.

I'd be glad to give more info about clusterm if anyone's interested.




	it makes me feel icky to link to "" when referencing
	something from Solaris

Philadelphia Linux Users Group         --
Announcements -
General Discussion  --