Please consider the attached path, which adds support for selecting a group of hosts based on descriptor-columns in the host-list. The host list file parser is expanded to support delimiter-separated values where the last column is the hostname. A new flag (-s) is introduced which allows the selection of hosts, each column can be matched by a regular expression. When all columns match the host is selected for the pssh operation.

Example; consider the following hostlist file: yellow fruit userA@banana green fruit userB@watermelon red fruit userC@cherry green veg userD@cucumber yellow veg userE@squash red veg userF@beet

pssh -h hostlist -s ".* veg" ls: will run ls on all "vegetable" hosts pssh -h hostlist -s "red .*": will select all red vegetable/fruit hosts.

The change is largely backwards compatible. When no -s flag is passed it is assumed that the traditional hostlist format is used. It should be noted that when -s is used the "hostname [user]" will not longer be a proper specification, and "[username@]hostname" should be used instead.

The number of "descriptor" columns in the host files is arbitrary, however the -s flag must have the corresponding number of space separated regular expressions.

The primary use case for this feature is for environments that have many different types of hosts and its is often necessary to perform an operation on a particular host-class. Previously one would have had to maintain separate hostlist file for each host-type and it was rather cumbersome.

Comments/suggestions welcome.

It's an interesting idea. Could you give a little more detail about why it's been cumbersome to maintain separate hostlist files? It should be possible to write a pre-processor that reads a file in the "descriptor" format and outputs a set of "classic-style" host files. Also, what happens when lines in the file don't all share the same number of columns? The "descriptor" format for host files seems interesting but a little complicated.

Cumbersomeness is proportional to the number of host classes one has to deal with. Certainly pre-processing is an option, but it adds one extra step. Furthermore one has to be careful when cleaning up so as to ensure that a command is not run on the wrong host (using a stale, derived hostlist). Albeit rather uncommon to run pssh against many different permutations of hostclasses it complicates the management overhead even more. Carrying one file around is a lot easier than dozens.

With current implementation, the behavior around incorrect number of columns is similar to behavior of pssh when it encounters more than 2 fields on a line in unpatched version. It prints a message that there was an error parsing a line and skips it. It is perhaps wise to adjust this behavior for both the patched and the unpatched version.

The feature is difficult to describe in words but it is rather intuitive. Perhaps my description is unclear. I would be more than happy to clarify it.

I think this feature is redundant and is easy to workaround. All you need is grep and process subsitution:

pssh -h <(grep -P ".* veg" hostlist) pssh -h <(grep -P "red .*" hostlist)

That's a cool trick. I think we should add something about that to the examples section of the man page.

hi, on Debian i have to use parallel-ssh rather than pssh directly is that normal?

ok got it ... i debian does not use the last version ... i pip install --upgrade pssh and now it works as :: pssh -Aih bob sudo apt-get update && sudo apt-get upgrade but does not let me select the machine I want.