apssh 0.22.0

Asynchroneous Parallel ssh

The apssh binary

Purpose

apssh is a tool that aims at running commands remotely using ssh on a large number of target nodes at once. It is thus comparable to parallel-ssh, except that it is written on top of asyncio and asyncssh.

In addition, apssh comes with a class SshJob that can be used in conjunction with asynciojobs to write scenarios that are more elaborate than just sending the same command on a bunch of hosts. This is presented in more details in README-jobs.md.

This document, along with API reference doc, and changelog, is available at http://apssh.readthedocs.io/

How to get it

Requirement

apssh requires python-3.5, as it uses the latest syntax constructions async def and await instead of the former @asyncio.coroutine and yield from idioms.

Installation

[sudo] pip3 install apssh

2 major modes : well-known commands, or local scripts

Usual mode

The usual way to run a command that is already present on the remote systems, is to do e.g. this (we'll see the -t option right away)

apssh -t host1 -t host2 hostname

Script mode : using a local script that gets copied over

Now if you need to run a more convoluted command, you can of course quote meta characters as ; and the like, and struggle your way using the same technique.

There is however an other way to achieve this, by writing a local script (usually a shell or python script) and use the -s/--script option, to have apssh copy it on the target nodes before executing it, like e.g.:

apssh -t host1 -t host2 --script mymacros.sh one two

• You can also use this option and provide your own script directly on the command line

$ apssh -s -t r2lab.infra --script 'arg1=$1; shift; arg2=$1; shift; echo exchanged $arg2 $arg1' one two
faraday.inria.fr:exchanged two one
bemol.pl.sophia.inria.fr:exchanged two one

This will have the effect to perform the following on each target node :

• create if needed a directory named ~/.apssh-remote
• copy the local file mymacros.sh - or your inline script - into that remote dir
• run .apssh/mymacros.sh one two remotely in the home directory

Note that in this mode:

• the first argument of the commands part (here mymacros.sh) should denote a file that exists locally, or be a valid script;
• it does not have to sit in the local directory but will be installed right under ~/.apssh-remote regardless;
• the remote file will be created in mode o755;
• the command executed remotely has its cwd set to the remote home directory.

Global return code

apssh returns 0 if and only if all remote commands complete and return 0 themselves; otherwise it returns 1.

Scope selection

Adding names : the -t or --target option

To run the command true on hosts host1 and host2 as well on all hostnames contained in file hosts.list, do this:

$ apssh -t host1 -t hosts.list -t host2 true

As a matter of fact you can use the --target option to refer to

• the name of an existing file: in this case, the file is read, lines with a # are considered comments and ignored, all the rest is considered a list of hostnames; you can have several hostnames on one line if you want;
• the name of an existing directory: in this case, all the simple files present in this directory are considered hostnames (see the --mark option below to see how this feature allows to easily select nodes that are actually online and reachable);
• otherwise, the string is considered a hostname itself, or possibly several space-separated hostnames.
• NOTE that files and directories are also searched in ~/.apssh, so that these shorthands can be defined globally.

So in practice, assuming that:

• directory hosts.outputs contains only 2 files named tutu and toto, and
• hosts.file is a text file containing the single line foo bar,

then if you run

apssh -t host1 -t "host2 host3" -t hosts.file -t hosts.dir true

it will cause the true command to be run on hosts host1, host2, host3, foo, bar, toto and tutu.

Excluding names : the -x or --exclude option

You can specify exclusions, the logic is exactly the same; exclusions are parsed first, and then hostnames from --target will be actually added only if they are not excluded. Which means the order in which you define targets and excludes does not matter.

So for example if you have all the known nodes in PLE in file PLE.nodes, and in a separate file PLE.dns-unknown the subset of the PLE nodes that are actually unknown to DNS, you can skip them by doing:

$ apssh -l root -t PLE.nodes -x PLE.dns-unknown cat /etc/fedora-release

or, equivalently:

$ apssh -l root -x PLE.dns-unknown -t PLE.nodes cat /etc/fedora-release

Max connections: the -w or --window option

By default there is no limit on the number of simultaneous connections, which is likely to be a problem as soon as you go for several tens of hosts, as you would then run into limitations on open connections in your OS or network. Use w or --window to run at most 50 connections at a time

$ apssh -w 50 -t tons-of-nodes true

Users and keys

Running under a different user

Use -l or --login to specify a specific username globally; or give a specific user on a given hostname with @

So e.g. to run as user on host1, but as root on host2 and host3:

$ apssh -l root -t user@host1 -t host2 -t host3 -- true

Keys

Here's how apssh locates private keys:

If no keys are specified using the -i command line option

• (A) if an ssh agent can be reached using the SSH_AUTH_SOCK environment variable, and offers a non-empty list of keys, apssh will use the keys loaded in the agent (NOTE: use ssh-add for managing the keys known to the agent);
• (B) otherwise, apssh will use ~/.ssh/id_rsa and ~/.ssh/id_dsa as far as they exist.

If keys are specified on the command line

• (C) That exact list is used for loading private keys.

In both cases

Note that when loading keys from a file - i.e. in cases (B) and (C) above, a passphrase will be prompted at the terminal for each key that is passphrase-protected. Each passphrase gets prompted once for all the target hosts of course.

It results from all this that passphrase-protected keys can be used in apssh without prompting only if present in an agent.

This behaviour might not be optimal - for example with this logic there is no way to use agent-loaded keys and additional keys. I am eager to receive feedback from users for possible improvements in this area.

Gateway a.k.a. Bouncing a.k.a. Tunnelling

In some cases, the target nodes are not directly addressable from the box that runs apssh, and the ssh traffic needs to go through a gateway. This typically occurs with testbeds where nodes only have private addresses.

For example in the R2lab testbed, you cannot reach nodes directly from the Internet, but you would need to issue something like:

# reaching one individual node with plain ssh
$ ssh [e-mail] ssh root@fit02 hostname
fit02

In such cases, you can specify the gateway username and hostname through the -g or --gateway option. For example for running the above command on several R2lab nodes in one apssh invokation:

$ apssh -g [e-mail] --login root -t "fit02 fit03 fit04" hostname
fit04:fit04
fit02:fit02
fit03:fit03

Note that in this case there is a single ssh connection created to the gateway.

Output formats

Default : on the fly, annotated with hostname

Default is to output every line as they come back, prefixed with associated hostname. As you might expect, stdout goes to stdout and stderr to stderr. Additionally, error messages issued by apssh itself, like e.g. when a host cannot be reached, also goes on stderr.

$ apssh -l root -t alive -- grep VERSION_ID /etc/os-release
[e-mail]:22 - Connection failed Disconnect Error: Permission denied
host2.planetlab.informatik.tu-darmstadt.de:VERSION_ID=23
host3.planetlab.informatik.tu-darmstadt.de:VERSION_ID=23
planetlab1.xeno.cl.cam.ac.uk:VERSION_ID=23
planetlab2.xeno.cl.cam.ac.uk:VERSION_ID=23

In the above trasnscript, there were 5 target hostnames, one of which being unreachable. The line with Permission denied goes on stderr, the other ones on stdout.

Your own format

You can specify a format with the --format option (see apssh --help); there also are a few predefined formats for convenience:

• -r/--raw (equivalent to --format '@line@') output is produced as it comes from the host, with no annotation as to which node the line is originating from.
• -tc/--time-colon-format is equivalent to --format '%H-%M-%S:@host@:@line@'.

Subdir : store outputs individually in a dedicated dir

Alternatively, the -o or -d options allow to select a specific subdir and to store results in files named after each hostname. In this case, stdout is expected to contain a single line that says in which directory results are to be found (this is useful mostly with -d, since with -o you can predict this in advance)

• Specifying -o it is possible to redirect outputs in a separate directory, in one file per hostname.
• The -d option behaves like -o with a name computed from the current time.

$ rm -rf alive.results/
$ apssh -o alive.results -l root -t alive cat /etc/fedora-release
alive.results
$ grep . alive.results/*
alive.results/mars.planetlab.haw-hamburg.de:Fedora release 14 (Laughlin)
alive.results/merkur.planetlab.haw-hamburg.de:Fedora release 14 (Laughlin)
alive.results/planetlab-2.research.netlab.hut.fi:Fedora release 22 (Twenty Two)
alive.results/planetlab1.tlm.unavarra.es:Fedora release 22 (Twenty Two)
alive.results/planetlab1.virtues.fi:Fedora release 14 (Laughlin)

When an output subdir is selected with either -d or -o, the -m or --mark option can be used to request details on the retcod from individual nodes. The way this is exposed in the filesystem under is as follows

• subdir/0ok/hostname will contain 0 for all nodes that could run the command successfully
• subdir/1failed/hostname will contain the actual retcod, for all nodes that were reached but could not successfully run the command, or None for the nodes that were not reached at all.

In the example below, we try to talk to two nodes, one of which is not reachable.

$ subdir=$(apssh --mark -d -l root -t planetlab2.tlm.unavarra.es -t uoepl2.essex.ac.uk cat /etc/fedora-release)
[e-mail][22]:Connection failed:[Errno 8] nodename nor servname provided, or not known

$ echo $subdir
2016-09-01@15:42

$ head -100 $(find $subdir -type f)
==> 2016-09-01@15:42/0ok/planetlab2.tlm.unavarra.es <==
0

==> 2016-09-01@15:42/1failed/uoepl2.essex.ac.uk <==
None

==> 2016-09-01@15:42/planetlab2.tlm.unavarra.es <==
Fedora release 18 (Spherical Cow)

Good practices

• First off, options order matters; apssh will stop interpreting options on your command line at the beginning of the remote command. That is to say, in the following example

$ apssh -t host1 -t file1 -t host2 rpm -aq \| grep libvirt

the -aq option is meant for the remote rpm command, and that's fine because after the rpm token, apssh stops taking options, and passes them to the remote command instead.

• Also note in the example above that you can pass shell specials, like |, <, >, ; and the like, by backslashing them, like this:

$ apssh -l root -t faraday.inria.fr -t r2lab.inria.fr uname -a \; cat /etc/fedora-release /etc/lsb-release 2\> /dev/null
r2lab.inria.fr:Linux r2lab.pl.sophia.inria.fr 4.6.4-201.fc23.x86_64 #1 SMP Tue Jul 12 11:43:59 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux
r2lab.inria.fr:Fedora release 24 (Twenty Four)
faraday.inria.fr:Linux faraday 4.4.0-36-generic #55-Ubuntu SMP Thu Aug 11 18:01:55 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux
faraday.inria.fr:DISTRIB_ID=Ubuntu
faraday.inria.fr:DISTRIB_RELEASE=16.04
faraday.inria.fr:DISTRIB_CODENAME=xenial
faraday.inria.fr:DISTRIB_DESCRIPTION="Ubuntu 16.04.1 LTS"

$ apssh -l root -t PLE.alive.5 -tc uname -r \; hostname
16-47-40:mars.planetlab.haw-hamburg.de:2.6.32-36.onelab.i686
16-47-40:merkur.planetlab.haw-hamburg.de:2.6.32-36.onelab.i686
16-47-40:mars.planetlab.haw-hamburg.de:mars.planetlab.haw-hamburg.de
16-47-40:merkur.planetlab.haw-hamburg.de:merkur.planetlab.haw-hamburg.de
16-47-40:planetlab1.tlm.unavarra.es:4.4.13-200.fc22.x86_64
16-47-40:planetlab1.tlm.unavarra.es:planetlab1.tlm.unavarra.es
16-47-40:planetlab1.virtues.fi:2.6.32-36.onelab.i686
16-47-40:planetlab1.virtues.fi:planetlab1.virtues.fi
16-47-40:planetlab-2.research.netlab.hut.fi:4.2.3-200.fc22.x86_64
16-47-40:planetlab-2.research.netlab.hut.fi:planetlab-2.research.netlab.hut.fi

TODO

• brewing something like appush and appull sounds pretty straightforward, and could turn out most useful; some day probably
• current output system can only properly handle commands output that are text-based; if your remote command produces binary data instead, you must redirect its output on the remote system, and fetch the results later on; note that the binary command apssh has no option for doing that, but the API has 2 objects Pull and Push for doing this in a more elaborate scenario (see README-jobs.md).
• better tests coverage would not hurt !?!
• probably a lot more features are required for more advanced usages, feel free to fill in issues at https://github.com/parmentelat/apssh.