Skip to content

Latest commit

 

History

History
245 lines (163 loc) · 9.25 KB

README.md

File metadata and controls

245 lines (163 loc) · 9.25 KB
Raw

Summary

rhubarbe is an asyncio/python module for loading images on a bunch of nodes using frisbee. It can be installed from pypi (see also INSTALLING.md for more details):

pip3 install rhubarbe

This is connected to an authorization system; most of the functions are rather intrusive, and for this reason they require the user to have obtained a lease (reservation), applicable to the current time and day, before the tool can be used. Note that the root user is unconditionnally granted permission.

Purpose

This is the software that runs the R2lab testbed in Inria Sophia Antipolis.

It is written in python3.5 on top of asyncio, and covers the following features :

  • on, off, and the like : managing nodes status through their CMC cards: status of the motherboard and of the USRP extension;
  • load, save , images : loading and saving images on nodes;
  • leases : displaying, and to some extent modifying, reservations on the testbed
  • monitornodes and monitorphones: feeding the sidecar server live info about the global testbed status (phones excluded)
  • accountsmanager : propagates the changes in the PLCAPI into actual Unix accounts and related authorized keys.

With these additional benefits:

  • single configuration file in /etc/rhubarbe/rhubarbe.conf, individual settings can be overridden at either

    • system-level in /etc/rhubarbe/rhubarbe.conf.local,
    • user-level in ~/.rhubarbe.conf, or
    • directory-level in ./rhubarbe.conf;

  • all commands accept a timeout

  • all commands return a reliable code. When everything goes fine for all subject nodes they return 0, and 1 otherwise

  • load and wait come with a curses mode that lets you visualize every node individually; very helpful when something goes wrong with a large number of nodes, so you can pinpoint which node is not behaving as expected.

Primer

Image loading

To load your fedora-23 image on 18 nodes simultaneously:

rhubarbe load -i fedora-23 1-18 &

You can safely load another batch of nodes at the same time, maybe with a different bandwidth

rhubarbe load -i ubuntu-16.04 -b 200 19-36 &

Controlling nodes

To turn nodes on, or off, or to reset (send Ctrl-Alt-Del) a set of nodes, use the following commands.

# turn on nodes 1 to 5
rhubarbe on 1-5
# turn off node 6
rhubarbe off 6
# reset node 7 (needs to be already on)
rhubarbe reset 7
# see if node 8 is on or off
rhubarbe status 8

Controlling USRP extensions

Nodes that have a USRP hardware can have the USRP extension handled in a similar way

# turn on the USRP extension on node 11
rhubarbe usrpon 11
# turn off the USRP extension on node 12
rhubarbe usrpoff 12
# get status of the USRP extension on node 13
rhubarbe usrpstatus 13

Waiting for a node

To wait for nodes 1, 3, 5, 7 and 9 to be reachable through ssh :

rhubarbe wait 1-9-2

Image saving

To save the image of node 10, just do this

rhubarbe save 10 -o image-name

or rather, if you'd like to specify a comment for bookkeeping :

 rhubarbe save 10 -o image-name -c 'this will end up in /etc/rhubarbe-image right in the image'

How to use

List of subcommands

The python entry point is named rhubarbe but it should be called with an additional subcommand.

root@etourdi ~ # rhubarbe
Unknown subcommand help - use one among {nodes,status,on,off,reset,info,usrpstatus,usrpon,usrpoff,load,save,wait,monitor,leases,images,share,inventory,config,version}

root@etourdi ~ # rhubarbe load --help

Invoking : node scope

Most commands expect a list of nodes as its arguments

$ rhubarbe load [-i filename] 1 4 5

The arguments, known as a node_spec can be individual nodes, ranges, or even steps, like e.g.

  • individual nodes

    • $ rhubarbe load 1 3 fit8 reboot12 for working on nodes, well
    • 1, 3, 8 12

  • ranges

    • $ rhubarbe load 1-3 8-10 on nodes
    • 1, 2, 3, 8, 9, 10

  • steps : in a python-like manner, from-to-step:

    • $ rhubarbe load 1-10-2 on nodes
    • 1 3 5 7 9

  • all nodes

    • $ rhubarbe load -a on nodes
    • 1 through 37 on faraday.inria.fr (exact scope being defined in the configuration)

  • negation

    • $ rhubarbe load -a ~10-20 on nodes
    • 1 to 9, and 21 to 37

  • if no node argument is provided on the command line, the value of the NODES environment variable is used. So you can select your set of nodes once, and just use the commands without arguments

    • $ export NODES="12-15 18-24"
    • $ rhubarbe load
    • $ rhubarbe wait

The leases subcommand

As of version 0.7.3:

  • rhubarbe leases is offered to all users to inspect the current leases in the system
  • rhubarbe leases -i is a utiliy offered to root only, unfortunately, that can create, update and delete leases in a simple terminal application.

Logging

At this point:

  • all logging goes into a file named rhubarbe.log,
  • except for the monitoring tool that logs into /var/log/monitor.log

Installation

Core

You need python-3.4 or higher, and installation can be achieved simply with

pip3 install rhubarbe

Update

pip3 install --upgrade rhubarbe

Other libraries

The following will be automatically installed by pip3 if not yet installed:

  • telnetlib3 for invoking frisbee on the nodes
  • aiohttp for talking to the CMC cards
  • asyncssh for talking to ssh (rhubarbe wait mostly for now);
    • ubuntu: you may need to run apt-get install libffi-dev before pip3 install asyncssh
  • progressbar33 is used in rendering progress bar in the regular monitor (i.e. without the -c option).

Setting up

Inventory

In short: see /etc/rhubarbe/inventory.json

Unfortunately the tool needs a mapping between hostnames and MAC addresses - essentially for messing with pxelinux nextboot symlinks. This is why the tool needs to find an inventory in a file named /etc/rhubarbe/inventory.json. See an extract below; note that the 'data' entry is not needed by the tools, we have them in place over here at r2lab for convenience only. The CMC mac address is not needed either, strictly speaking, as of this writing.

On R2LAB: this is taken care of by inventory/configure.py and its Makefile. Note that like for the OMF JSON inventory file, configure.py creates 2 different files for faraday and bemol - to account for any replacement node on faraday, like when e.g. node 41 actually sits in slot 15.

FYI an inventory files just looks like below; the data field is not needed

# less /etc/rhubarbe/inventory.json
 [
  {
    "cmc": {
      "hostname": "reboot01",
      "ip": "192.168.1.1",
      "mac": "02:00:00:00:00:01"
    },
    "control": {
      "hostname": "fit01",
      "ip": "192.168.3.1",
      "mac": "00:03:1d:0e:03:19"
    },
    "data": {
      "hostname": "data01",
      "ip": "192.168.2.1",
      "mac": "00:03:1d:0e:03:18"
    }
  },
  ... etc
  ]

Configuration

In short: see /etc/rhubarbe/rhubarbe.conf

Configuration is done through a collection of files, which are loaded in this order if they exist:

  • /etc/rhubarbe/rhubarbe.conf
  • /etc/rhubarbe/rhubarbe.conf.local
  • ~/.rhubarbe.conf
  • ./rhubarbe.conf

So in essence, there is

  • a (mandatory) system-wide config /etc/rhubarbe/rhubarbe.conf, that ships with the pip3 package and that contains all variable definitions;
  • given that /etc/rhubarbe/rhubarbe.conf is likely to be overwritten at anytime by pip3 install, you can store your own system-wide changes in the (optional) file /etc/rhubarbe.conf.local;
  • then each user may override any value she likes,
  • and finally one can be even more specific and configure things at a directory level.

The first file will come with all the settings defined, but any of the other 3 does not need to be complete and can just redefine one variable if needed.

Format is known as a .ini file, should be straightforward. Just beware to not mention quotes around strings, as such quotes end up in the python string verbatim.

Authorization system

Among things to be configured are, in the plcapi section:

  • plcapi_url : the URL for a leases server, and
  • leases_hostname : the hostname of the PLCAPI node that is used to record leases on the testbed.

In other words, all this for now assumes the following:

  • You run an instance of a PLCAPI service at that URL, and
  • this API knows about one node whose hostname is specified above.

This is an admittedly specific policy for R2Lab, since we want the reservations to be made on the testbed as a whole (to ensure reproducibility).

A word on the asyncio module

As of September 2016, we only use the syntax of 3.5's asyncio, based on async def and await.

We use python 3.5's asyncio library. python3.5 can be taken as granted on the ubuntus we use on both faraday and bemol. This is consistent with python-3.5 being part of both Fedora-24 and Ubuntu-16.04

TODO

  • add option rshare --clean that would clean up the other/older candidates for a name

  • some option in rhubarbe leases to skip the nightly leases

  • in terms of the cleanup regarding display.stop / forever jobs

    • issue was: how to properly terminate a loop that has a infinite task going on
    • synciojobs provides a nice replacement with forever jobs
    • this has been deployed almost everywhere, except for monitor for now