Choria Network Protocols – Transport

The old MCollective protocols are now ancient and was quite Ruby slanted – full of Symbols and used YAML and quite language specific – in Choria I’d like to support other Programming Languages, REST gateways and so forth, so a rethink was needed.

I’ll look at the basic transport protocol used by the Choria NATS connector, usually it’s quite unusual to speak of Network Protocols when dealing with messages on a broker but really for MCollective it is exactly that – a Network Protocol.

The messages need enough information for strong AAA, they need to have an agreed on security structure and within them live things like RPC requests. So a formal specification is needed which is exactly what a Protocol is.

While creating Choria the entire protocol stack has been redesigned on every level except the core MCollective messages – Choria maintains a small compatibility layer to make things work. To really achieve my goal I’d need to downgrade MCollective to pure JSON data at which point multi language interop should be possible and easy.

Networks are Onions

Network protocols tend to come in layers, one protocol within another within another. The nearer you go to the transport the more generic it gets. This is true for HTTP within TCP within IP within Ethernet and likewise it’s true for MCollective.

Just like for TCP/IP and HTTP+FTP one MCollective network can carry many protocols like the usual RPC others can exist, a typical MCollective install uses 2 protocols at this inner most layer.

( middleware protocol
  ( transport packet that travels over the middleware
      ( security plugin internal representation
        ( mcollective core representation that becomes M::Message
          ( MCollective Core Message )
          ( RPC Request, RPC Reply )
          ( Other Protocols, .... )

Here you can see when you do mco rpc puppet status you’ll be creating a RPC Request wrapped in a MCollective Message, wrapped in a structure the Security Plugin dictates, wrapped in a structure the Connector Plugin dictates and from there to your middleware like NATS.

Today I’ll look at the Transport Packet since that is where Network Federation lives which I spoke about yesterday.

Transport Layer

The Transport Layer packets are unauthenticated and unsigned, for MCollective security happens in the packet carried within the transport so this is fine. It’s not inconceivable that a Federation might only want to route signed messages and it’s quite easy to add later if needed.

Of course the NATS daemons will only accept TLS connections from certificates signed by the CA so these network packets are encrypted and access to the transport medium is restricted, but the JSON data you’ll see below is sent as is.

In all the messages shown below you’ll see a seen-by header, this is a feature of the NATS Connector Plugin that records the connected NATS broker, we’ll soon expose this information to MCollective API clients so we can make a traceroute tool for Federations. This header is optional and off by default though.

I’ll show messages in Ruby format here but it’s all JSON on the wire.

Message Targets

First it’s worth knowing where things are sent on the NATS clusters. The targets used by the NATS connector is pretty simple stuff, there will no doubt be scope for improvement once I look to support NATS Streaming but for now this is adequate.

  • Broadcast Request for agent puppet in the mycorp sub collective – mycorp.broadcast.agent.puppet
  • Directed Request to a node for any agent in the mycorp sub collective –
  • Reply to a node identity with pid 9999 and a message sequence of

As the Federation Brokers are independent of Sub Collectives they are not prefixed with any collective specific token:

  • Requests from a Federation Client to a Federation Broker Cluster called productionchoria.federation.production.federation queue group production_federation
  • Replies from the Collective to a Federation Broker Cluster called productionchoria.federation.production.collective queue group production_collective
  • production cluster Federation Broker Instances publishes statistics – choria.federation.production.stats

These names are designed so that in smaller setups or in development you could use a single NATS cluster with Federation Brokers between standalone collectives. Not really a recommended thing but it helps in development.

Unfederated Messages

Your basic Unfederated Message is pretty simple:

  "data" => "... any text ...",
  "headers" => {
    "mc_sender" => "",
    "seen-by" => ["", ""],
    "reply-to" => "",
  • it’s is a discovery request within the sub collective mcollective and would be published to mcollective.broadcast.agent.discovery.
  • it is sent from a machine identifying as
  • we know it’s traveled via a NATS broker called
  • responses to this message needs to travel via NATS using the target

The data is completely unstructured as far as this message is concerned it just needs to be some text, so base64 encoded is common. All the transport care for is getting this data to its destination with metadata attached, it does not care what’s in the data.

The reply to this message is almost identical but even simpler:

  "data" => "... any text ...",
  "headers" => {
    "mc_sender" => "",
    "seen-by" => ["", "", "", ""],

This reply will travel via, we know that the node is connected to

We can create a full traceroute like output with this which would show -> -> ->

Federated Messages

Federation is possible because MCollective will just store whatever Headers are in the message and put them back on the way out in any new replies. Given this we can embed all the federation metadata and this metadata travels along with each individual message – so the Federation Brokers can be entirely stateless, all the needed state lives with the messages.

With Federation Brokers being clusters this means your message request might flow over a cluster member a but the reply can come via b – and if it’s a stream of replies they will be load balanced by the members. The Federation Broker Instances do not need something like Consul or shared store since all the data needed is in the messages.

Lets look at the same Request as earlier if the client was configured to belong to a Federation with a network called production as one of its members. It’s identical to before except the federation structure was added:

  "data" => "... any text ...",
  "headers" => {
    "mc_sender" => "",
    "seen-by" => ["", ""],
    "reply-to" => "",
    "federation" => {
       "req" => "68b329da9893e34099c7d8ad5cb9c940",
       "target" => ["mcollective.broadcast.agent.discovery"]
  • it’s is a discovery request within the sub collective mcollective and would be published via a Federation Broker Cluster called production via NATS choria.federation.production.federation.
  • it is sent from a machine identifying as
  • it’s traveled via a NATS broker called
  • responses to this message needs to travel via NATS using the target
  • it’s federated and the client wants the Federation Broker to deliver it to it’s connected Member Collective on mcollective.broadcast.agent.discovery

The Federation Broker receives this and creates a new message that it publishes on it’s Member Collective:

  "data" => "... any text ...",
  "headers" => {
    "mc_sender" => "",
    "seen-by" => [
    "reply-to" => "choria.federation.production.collective",
    "federation" => {
       "req" => "68b329da9893e34099c7d8ad5cb9c940",
       "reply-to" => ""

This is the same message as above, the Federation Broker recorded itself and it’s connected NATS server and produced a message, but in this message it intercepts the replies and tell the nodes to send them to choria.federation.production.collective and it records the original reply destination in the federation header.

A node that replies produce a reply, again this is very similar to the earlier reply except the federation header is coming back exactly as it was sent:

  "data" => "... any text ...",
  "headers" => {
    "mc_sender" => "",
    "seen-by" => [
    "federation" => {
       "req" => "68b329da9893e34099c7d8ad5cb9c940",
       "reply-to" => ""

We know this node was connected to and you can see the Federation Broker would know how to publish this to the client – the reply-to is exactly what the Client initially requested, so it creates:

  "data" => "... any text ...",
  "headers" => {
    "mc_sender" => "",
    "seen-by" => [

Which gets published to

Route Records

You noticed above there’s a seen-by header, this is something entirely new and never before done in MCollective – and entirely optional and off by default. I anticipate you’d want to run with this off most of the time once your setup is done, it’s a debugging aid.

As NATS is a full mesh your message probably only goes one hop within the Mesh. So if you record the connected server you publish into and the connected server your message entered it’s destination from you have a full route recorded.

The Federation Broker logs and MCollective Client and Server logs all include the message ID so you can do a full trace in message packets and logs.

There’s a PR against MCollective to expose this header to the client code so I will add something like mco federation trace which would send a round trip to that node and tell you exactly how the packet travelled. This should help a lot in debugging your setups as they will now become quite complex.

The structure here is kind of meh and I will probably improve on it once the PR in MCollective lands and I can see what is the minimum needed to do a full trace.

By default I’ll probably record the identities of the MCollective bits when Federated and not at all when not Federated. But if you enable the setting to record the full route it will produce a record of MCollective bits and the NATS nodes involved.

In the end though from the Federation example we can infer a network like this:

Federation NATS Cluster

  • Federation Broker production_a ->
  • Federation Broker production_b ->
  • Client ->

Production NATS Cluster:

  • Federation Broker production_a ->
  • Federation Broker production_b ->
  • Server ->

We don’t know the details of all the individual NATS nodes that makes up the entire NATS mesh but this is good enough.

Of course this sample is the pathological case where nothing is connected to the same NATS instances anywhere. In my tests with a setup like this the overhead added across 10 000 round trips against 3 nodes – so 30 000 replies through 2 x Federation Brokers – was only 2 seconds, I couldn’t reliably measure a per message overhead as it was just too small.

The NATS gem do expose the details of the full mesh though since NATS will announce it’s cluster members to clients, I might do something with that not sure. Either way, auto generated network maps should be totally possible.


So this is how Network Federation works in Choria. It’s particularly nice that I was able to do this without needing any state on the cluster thanks to past self making good design decisions in MCollective.

Once the seen-by thing is figured out I’ll publish JSON Schemas for these messages and declare protocol versions.

I can probably make future posts about the other message formats but they’re a bit nasty as MCollective itself is not yet JSON safe, the plan is it would become JSON safe one day and the whole thing will become a lot more elegant. If someone pings me for this I’ll post it otherwise I’ll probably stop here.

Choria Update

Recently at Config Management Camp I’ve had many discussions about Orchestration, Playbooks and Choria, I thought it’s time for another update on it’s status.

I am nearing version 1.0.0, there are a few things to deal with but it’s getting close. Foremost I wanted to get the project it’s own space on all the various locations like GitHub, Forge, etc.

Inevitably this means getting a logo, it’s been a bit of a slog but after working through loads of feedback on Twitter and offers for assistance from various companies I decided to go to a private designer called Isaac Durazo and the outcome can be seen below:


The process of getting the logo was quite interesting and I am really pleased with the outcome, I’ll blog about that separately.

Other than the logo the project now has it’s own GitHub organisation at and I have moved all the forge modules to it’s own space as well

There are various other places the logo show up like in the Slack notifications and so forth.

On the project front there’s a few improvements:

  • There is now a registration plugin that records a bunch of internal stats on disk, the aim is for them to be read by Collectd and Sensu
  • A new Auditing plugin that emits JSON structured data
  • Several new Data Stores for Playbooks – files, environment.
  • Bug fixes on Windows
  • All the modules, plugins etc have moved to the Choria Forge and GitHub
  • Quite extensive documentation site updates including branding with the logo and logo colors.

There is now very few things left to do to get 1.0.0 out but I guess another release or two will be done before then.

So from now to update to coming versions you need to use the choria/mcollective_choria module which will pull in all it’s dependencies from the Choria project rather than my own Forge.

Still no progress on moving the actual MCollective project forward but I’ve discussed a way to deal with forking the various projects in a way that seems to work for what I want to achieve. In reality I’ll only have time to do that in a couple of months so hopefully something positive will happen in the mean time.

Head over to to take a look.

Choria Playbooks – Data Sources

About a month ago I blogged about Choria Playbooks – a way to write series of actions like MCollective, Shell, Slack, Web Hooks and others – contained within a YAML script with inputs, node sets and more.

Since then I added quite a few tweaks, features and docs, it’s well worth a visit to to check it out.

Today I want to blog about a major new integration I did into them and a major step towards version 1 for Choria.


In the context of a playbook or even a script calling out to other system there’s many reasons to have a Data Source. In the context of a playbook designed to manage distributed systems the Data Source needed has some special needs. Needs that tools like Consul and etcd fulfil specifically.

So today I released version 0.0.20 of Choria that includes a Memory and a Consul Data Source, below I will show how these integrate into the Playbooks.

I think using a distributed data store is important in this context rather than expecting to pass variables from the Playbook around like on the CLI since the business of dealing with the consistency, locking and so forth are handled and I can’t know all the systems you wish to interact with, but if those can speak to Consul you can prepare an execution environment for them.

For those who don’t agree there is a memory Data Store that exists within the memory of the Playbook. Your playbook should remain the same apart from declaring the Data Source.

Using Consul

Defining a Data Source

Like with Node Sets you can have multiple Data Sources and they are identified by name:

    type: consul
    timeout: 360
    ttl: 20

This creates a Consul Data Source called pb_data, you need to have a local Consul Agent already set up. I’ll cover the timeout and ttl a bit later.

Playbook Locks

You can create locks in Consul and by their nature they are distributed across the Consul network. This means you can ensure a playbook can only be executed once per Consul DC or by giving a custom lock name any group of related playbooks or even other systems that can make Consul locks.

  - pb_data
  - pb_data/custom_lock

This will create 2 locks in the pb_data Data Store – one called custom_lock and another called choria/locks/playbook/pb_name where pb_name is the name from the metadata.

It will try to acquire a lock for up to timeout seconds – 360 here, if it can’t the playbook run fails. The associated session has a TTL of 20 seconds and Choria will renew the sessions around 5 seconds before the TTL expires.

The TTL will ensure that should the playbook die, crash, machine die or whatever, the lock will release after 20 seconds.

Binding Variables

Playbooks already have a way to bind CLI arguments to variables called Inputs. Data Sources extend inputs with extra capabilities.

We now have two types of Input. A static input is one where you give the data on the CLI and the data stays static for the life of the playbook. A dynamic input is one bound against a Data Source and the value of it is fetched every time you reference the variable.

    description: "Cluster to deploy"
    type: "String"
    required: true
    data: "pb_data/choria/kv/cluster"
    default: "alpha"

Here we have a input called cluster bound to the choria/kv/cluster key in Consul. This starts life as a static input and if you give this value on the CLI it will never use the Data Source.

If however you do not specify a CLI value it becomes dynamic and will consult Consul. If there’s no such key in Consul the default is used, but the input remains dynamic and will continue to consult Consul on every access.

You can force an input to be dynamic which will mean it will not show up on the CLI and will only speak to a data source using the dynamic: true property on the Input.

Writing and Deleting Data

Of course if you can read data you should be able to write and delete it, I’ve added tasks to let you do this:

  - pb_data
    description: "Cluster to deploy"
    type: "String"
    required: true
    data: "pb_data/choria/kv/cluster"
    default: "alpha"
    validation: ":shellsafe"
    - data:
        action: "delete"
        key: "pb_data/choria/kv/cluster"
  - shell:
      description: Deploy to cluster {{{ inputs.cluster }}}
      command: /path/to/script --cluster {{{ inputs.cluster }}}
  - data:
      action: "write"
      value: "bravo"
      key: "pb_data/choria/kv/cluster"
  - shell:
      description: Deploy to cluster {{{ inputs.cluster }}}
      command: /path/to/script --cluster {{{ inputs.cluster }}}

Here I have a pre_book task list that ensures there is no stale data, the lock ensures no other Playbook will mess around with the data while we run.

I then run a shell command that uses the cluster input, with nothing there it uses the default and so deploys cluster alpha, it then writes a new value and deploys cluster brova.

This is a bit verbose I hope to add the ability to have arbitrarily named tasks lists that you can branch to, then you can have 1 deploy task list and use the main task list to set up variables for it and call it repeatedly.


That’s quite a mouthful, the possibilities of this is quite amazing. On one hand we have a really versatile data store in the Playbooks but more significantly we have expanded the integration possibilities by quite a bit, you can now have other systems manage the environment your playbooks run in.

I will soon add task level locks and of course Node Set integration.

For now only Consul and Memory is supported, I can add others if there is demand.

Choria Playbooks

Today I am very pleased to release something I’ve been thinking about for years and actively working on since August.

After many POCs and thrown away attempts at this over the years I am finally releasing a Playbook system that lets you run work flows on your MCollective network – it can integrate with a near endless set of remote services in addition to your MCollective to create a multi service playbook system.

This is a early release with only a few integrations but I think it’s already useful and I’m looking for feedback and integrations to build this into something really powerful for the Puppet eco system.

The full docs can be found on the Choria Website, but below you can get some details.


Today playbooks are basic YAML files. They do not have a pseudo programming language in them though I am not against the idea. Eventually I envision a Service to execute playbooks on your behalf, but today you just run them in your shell. I do not anticipate YAML to be the end format of playbooks but it’s good enough for today.

Playbooks have a basic flow that is more or less like this:

  1. Discover named Node Sets
  2. Validate the named Node Sets meet expectations such as reachability and versions of software available on them
  3. Run a pre_book task list that lets you do prep work
  4. Run the main tasks task list where you do your work, around every task certain hook lists can be run
  5. Run either the on_success or on_fail task list for notification of Slacks etc
  6. Run the post_book task list for cleanups etc

Today a task can be a MCollective request, a shell script or a Slack notification. I imagine this list will grow huge, I am thinking you will want to ping webhooks, or interact with Razor to provision machines and wait for them to finish building, run Terraform or make EC2 API requests. This list of potential integrations is endless and you can use any task in any of the above task lists.

A Node Set is simply a named set of nodes, in MCollective that would be certnames of nodes but the playbook system itself is not limited to that. Today Node Sets can be resolved from MCollective Discovery, PQL Queries (PuppetDB), YAML files with groups of nodes in them or a shell command. Again the list of integrations that make sense here is huge. I imagine querying PE or Foreman for node groups, querying etcd or Consul for service members. Talking to random REST services that return node lists or DB queries. Imagine using Terraform outputs as Node Set sources or EC2 API queries.

In cases where you wish to manage nodes via MCollective but you are using a cached discovery source you can ask node sets to be tested for reachability over MCollective. And node sets that need certain MCollective agents can express this desire as SemVer version ranges and the valid network state will be asserted before any playbook is run.


I’ll show an example here of what I think you will be able to achieve using these Playbooks.

Here we have a web stack and we want to do Blue/Green deploys against it, sub clusters have a fact cluster. The deploy process for a cluster is:

  • Gather input from the user such as cluster to deploy and revision of the app to deploy
  • Discover the Haproxy node using Node Set discovery from PQL queries
  • Discover the Web Servers in a particular cluster using Node Set discovery from PQL queries
  • Verify the Haproxy nodes and Web Servers are reachable and running the versions of agents we need
  • Upgrade the specific web tier using:
    1. Tell the ops room on slack we are about to upgrade the cluster
    2. Disable puppet on the webservers
    3. Wait for any running puppet runs to stop
    4. Disable the nodes on a particular haproxy backend
    5. Upgrade the apps on the servers using appmgr#upgrade to the input revision
    6. Do up to 10 NRPE checks post upgrade with 30 seconds between checks to ensure the load average is GREEN, you’d use a better check here something app specific
    7. Enable the nodes in haproxy once NRPE checks pass
    8. Fetch and display the status of the deployed app – like what version is there now
    9. Enable Puppet

Should the task list all FAIL we run these tasks:

  1. Call a webhook on AWS Lambda
  2. Tell the ops room on slack
  3. Run a whole other playbook called deploy_failure_handler with the same parameters

Should the task list PASS we run these tasks:

  1. Call a webhook on AWS Lambda
  2. Tell the ops room on slack

This example and sample playbooks etc can be found on the Choria Site.


Above is the eventual goal. Today the major missing piece here that I think MCollective needs to be extended with the ability for Agent plugins to deliver a Macro plugin. A macro might be something like Puppet.wait_till_idle(:timeout => 600), this would be something you call after disabling the nodes and you want to be sure Puppet is making no more changes, you can see the workflow above needs this.

There is no such Macros today, I will add a stop gap solution as a task that waits for a certain condition but adding Macros to MCollective is high on my todo list.

Other than that it works, there is no web service yet so you run them from the CLI and the integrations listed above is all that exist, they are quite easy to write so hoping some early adopters will either give me ideas or send PRs!

This is available today if you upgrade to version 0.0.12 of the ripienaar-mcollective_choria module.

Again see the Choria Website for much more details on this feature.

An update on my Choria project

Some time ago I mentioned that I am working on improving the MCollective Deployment story.

I started a project called Choria that aimed to massively improve the deployment UX and yield a secure and stable MCollective setup for those using Puppet 4.

The aim is to make installation quick and secure, towards that it seems a common end to end install from scratch by someone new to project using a clustered NATS setup can take less than a hour, this is a huge improvement.

Further I’ve had really good user feedback, especially around NATS. One user reports 2000 nodes on a single NATS server consuming 300MB RAM and it being very performant, much more so than the previous setup.

It’s been a few months, this is whats changed:

  • The module now supports every OS AIO Puppet supports, including Windows.
  • Documentation is available on, installation should take about a hour max.
  • The PQL language can now be used to do completely custom infrastructure discovery against PuppetDB.
  • Many bugs have been fixed, many things have been streamlined and made more easy to get going with better defaults.
  • Event Machine is not needed anymore.
  • A number of POC projects have been done to flesh out next steps, things like a very capable playbook system and a revisit to the generic RPC client, these are on GitHub issues.

Meanwhile I am still trying to get to a point where I can take over maintenance of MCollective again, at first Puppet Inc was very open to the idea but I am afraid it’s been 7 months and it’s getting nowhere, calls for cooperation are just being ignored. Unfortunately I think we’re getting pretty close to a fork being the only productive next step.

For now though, I’d say the Choria plugin set is production ready and stable any one using Puppet 4 AIO should consider using these – it’s about the only working way to get MCollective on FOSS Puppet now due to the state of the other installation options.

Introduction to MCollective deck

I’ve not had a good introduction to MCollective slide deck ever, I usually just give demos and talk through it. I was invited to talk in San Francisco about MCollective so made a new deck for this talk.

On the night I gave people the choice of talks between the new Introduction talk and the older Managing Puppet using MCollective and sadly the intro talk lost out.

Last night the excellent people at Workday flew me to Dublin to talk to the local DevOps group there and this group was predominantly Chef users who chose the Introduction talk so I finally had a chance to deliver it. This talk was recorded, hopefully it’ll be up soon and I’ll link to it once available.

This slide deck is a work in progress, it’s clear I need to add some more information about the non-cli orientated uses of MCollective but it’s good to finally have a deck that’s receiving good feedback.

We uploaded the slides back when I was in San Francisco to slideshare and those are the ones you see here.

Managing Puppet Using MCollective

I recently gave a talk titled “Managing Puppet Using MCollective” at the Puppet Camp in Ghent.

The talk introduces a complete rewrite of the MCollective plugin used to manage Puppet. The plugin can be found on our Github repo as usual. Significantly this is one of a new breed of plugin that we ship as native OS packages and practice continuous delivery on.

The packages can be found on and and are simply called mcollective-puppet-agent and mcollective-puppet-client.

This set of plugins show case a bunch of recent MCollective features including:

  • Data Plugins
  • Aggregation Functions
  • Custom Validators
  • Configurable enabling and disabling of the Agent
  • Direct Addressing and pluggable discovery to significantly improve the efficiency of the runall method
  • Utility classes shared amongst different types of plugin
  • Extensive testing using rspec and our mcollective specific rspec plugins

It’s a bit of a beast coming at a couple thousand lines but this was mostly because we had to invent a rather sizeable wrapper for Puppet to expose a nice API around Puppet 2.7 and 3.x for things like running them and obtaining their status.

The slides from the talk can be seen below, hopefully a video will be up soon else I’ll turn it into a screencast.

Graphing on the CLI

I’ve recently been thinking about ways to do graphs on the CLI. We’ve written a new Puppet Agent for MCollective that can gather all sorts of interesting data from your server estate and I’d really like to be able to show this data on the CLI. This post isn’t really about MCollective though the ideas applies to any data.

I already have sparklines in MCollective, here’s the distribution of ping times:

This shows you that most of the nodes responded quickly with a bit of a tail at the end being my machines in the US.

Sparklines are quite nice for a quick overview so I looked at adding some more of this to the UI and came up with this:

Which is quite nice – these are the nodes in my infrastructure stuck into buckets and the node counts for each bucket is shown. We can immediately tell something is not quite right – the config retrieval time shows a bunch of slow machines and the slowness does not correspond to resource counts etc. On investigation I found these are my dev machines – KVM nodes hosted on HP Micro Servers so that’s to be expected.

I am not particularly happy with these graphs though so am still exploring other options, one other option is GNU Plot.

GNU Plot can target its graphs for different terminals like PNG and also line printers – since the Unix terminal is essentially a line printer we can use this.

Here are 2 graphs of config retrieval time produced by MCollective using the same data source that produced the spark line above – though obviously from a different time period. Note that the axis titles and graph title is supplied automatically using the MCollective DDL:

$ mco plot resource config_retrieval_time
                   Information about Puppet managed resources
    6 ++-*****----+----------+-----------+----------+----------+----------++
      +      *    +          +           +          +          +           +
      |       *                                                            |
    5 ++      *                                                           ++
      |       *                                                            |
      |        *                                                           |
    4 ++       *      *                                                   ++
      |        *      *                                                    |
      |         *    * *                                                   |
    3 ++        *    * *                                                  ++
      |          *  *  *                                                   |
      |           * *   *                                                  |
    2 ++           *    *                         *        *              ++
      |                 *                         **       **              |
      |                  *                       * *      *  *             |
    1 ++                 *               *       *  *     *   **        * ++
      |                  *              * *     *   *     *     **    **   |
      +           +       *  +         * + *    *   +*   *     +     *     +
    0 ++----------+-------*************--+--****----+*****-----+--***-----++
      0           10         20          30         40         50          60
                              Config Retrieval Time

So this is pretty serviceable for showing this data on the console! It wouldn’t scale to many lines but for just visualizing some arbitrary series of numbers it’s quite nice. Here’s the GNU Plot script that made the text graph:

set title "Information about Puppet managed resources"
set terminal dumb 78 24
set key off
set ylabel "Nodes"
set xlabel "Config Retrieval Time"
plot '-' with lines
3 6
6 6
9 3
11 2
14 4
17 0
20 0
22 0
25 0
28 0
30 1
33 0
36 038 2
41 0
44 0
46 2
49 1
52 0
54 0
57 1

The magic here comes from the second line that sets the output terminal to dump and supplies some dimensions. Very handy, worth exploring some more and adding to your toolset for the CLI. I’ll look at writing a gem or something that supports both these modes.

There are a few other players in this space, I definitely recall coming across a Python tool to do graphs but cannot find it now, shout out in the comments if you know other approaches and I’ll add them to the post!

Updated: some links to related projects: sparkler, Graphite Spark

Scaling Nagios NRPE checks

Most Nagios systems does a lot of forking especially those built around something like NRPE where each check is a connection to be made to a remote system. On one hand I like NRPE in that it puts the check logic on the nodes using a standard plugin format and provides a fairly re-usable configuration file but on the other hand the fact that the Nagios machine has to do all this forking has never been good for me.

In the past I’ve shown one way to scale checks by aggregate all results for a specific check into one result but this is not always a good fit as pointed out in the post. I’ve now built a system that use the same underlying MCollective infrastructure as in the previous post but without the aggregation.

I have a pair of Nagios nodes – one in the UK and one in France – and they are on quite low spec VMs doing around 400 checks each. The problems I have are:

  • The machines are constantly loaded under all the forking, one would sit on 1.5 Load Average almost all the time
  • They use a lot of RAM and it’s quite spikey, if something is wrong especially I’d have a lot of checks concurrently so the machines have to be bigger than I want them
  • The check frequency is quite low in the usual Nagios manner, sometimes 10 minutes can go by without a check
  • The check results do not represent a point in time, I have no idea how the check results of node1 relate to those on node2 as they can be taken anywhere in the last 10 minutes

These are standard Nagios complaints though and there are many more but these ones specifically is what I wanted to address right now with the system I am showing here.

Probably not a surprise but the solution is built on MCollective, it uses the existing MCollective NRPE agent and the existing queueing infrastructure to push the forking to each individual node – they would do this anyway for every NRPE check – and read the results off a queue and spool it into the Nagios command file as Passive results. Internally it splits the traditional MCollective request-response system into a async processing system using the technique I blogged about before.

As you can see the system is made up of a few components:

  • The Scheduler takes care of publishing requests for checks
  • MCollective and the middleware provides AAA and transport
  • The nodes all run the MCollective NRPE agent which put their replies on the Queue
  • The Receiver reads the results from the Queue and write them to the Nagios command file

The Scheduler

The scheduler daemon is written using the excellent Rufus Scheduler gem – if you do not know it you totally should check it out, it solves many many problems. Rufus allows me to create simple checks on intervals like 60s and I combine these checks with MCollective filters to create a simple check configuration as below:

nrpe 'check_bacula_main', '6h', 'bacula::node monitored_by=monitor1'
nrpe 'check_disks', '60s', 'monitored_by=monitor1'
nrpe 'check_greylistd', '60s', 'greylistd monitored_by=monitor1'
nrpe 'check_load', '60s', 'monitored_by=monitor1'
nrpe 'check_mailq', '60s', 'monitored_by=monitor1'
nrpe 'check_mongodb', '60s', 'mongodb monitored_by=monitor1'
nrpe 'check_mysql', '60s', 'mysql::server monitored_by=monitor1'
nrpe 'check_pki', '60m', 'monitored_by=monitor1'
nrpe 'check_swap', '60s', 'monitored_by=monitor1'
nrpe 'check_totalprocs', '60s', 'monitored_by=monitor1'
nrpe 'check_zombieprocs', '60s', 'monitored_by=monitor1'

Taking the first line it says: Run the check_bacula_main NRPE check every 6 hours on machines with the bacula::node Puppet Class and with the fact monitored_by=monitor1. I had the monitored_by fact already to assist in building my Nagios configs using a simple search based approach in Puppet.

When the scheduler starts it will log:

W, [2012-12-31T22:10:12.186789 #32043]  WARN -- : activemq.rb:96:in `on_connecting' TCP Connection attempt 0 to stomp://
W, [2012-12-31T22:10:12.193405 #32043]  WARN -- : activemq.rb:101:in `on_connected' Conncted to stomp://
I, [2012-12-31T22:10:12.196387 #32043]  INFO -- : scheduler.rb:23:in `nrpe' Adding a job for check_bacula_main every 6h matching 'bacula::node monitored_by=monitor1', first in 19709s
I, [2012-12-31T22:10:12.196632 #32043]  INFO -- : scheduler.rb:23:in `nrpe' Adding a job for check_disks every 60s matching 'monitored_by=monitor1', first in 57s
I, [2012-12-31T22:10:12.197173 #32043]  INFO -- : scheduler.rb:23:in `nrpe' Adding a job for check_load every 60s matching 'monitored_by=monitor1', first in 23s
I, [2012-12-31T22:10:35.326301 #32043]  INFO -- : scheduler.rb:26:in `nrpe' Publishing request for check_load with filter 'monitored_by=monitor1'

You can see it reads the file and schedule the first check a random interval between now and the interval window this spread out the checks.

The Receiver

The receiver has almost no config, it just need to know what queue to read and where your Nagios command file lives, it logs:

I, [2013-01-01T11:49:38.295661 #23628]  INFO -- : mnrpes.rb:35:in `daemonize' Starting in the background
W, [2013-01-01T11:49:38.302045 #23631]  WARN -- : activemq.rb:96:in `on_connecting' TCP Connection attempt 0 to stomp://
W, [2013-01-01T11:49:38.310853 #23631]  WARN -- : activemq.rb:101:in `on_connected' Conncted to stomp://
I, [2013-01-01T11:49:38.310980 #23631]  INFO -- : receiver.rb:16:in `subscribe' Subscribing to /queue/mcollective.nagios_passive_results_monitor1
I, [2013-01-01T11:49:41.572362 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357040981] PROCESS_SERVICE_CHECK_RESULT;;mongodb;0;OK: connected, databases admin local my_db puppet mcollective
I, [2013-01-01T11:49:42.509061 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357040982] PROCESS_SERVICE_CHECK_RESULT;;zombieprocs;0;PROCS OK: 0 processes with STATE = Z
I, [2013-01-01T11:49:42.510574 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357040982] PROCESS_SERVICE_CHECK_RESULT;;zombieprocs;0;PROCS OK: 1 process with STATE = Z

As the results get pushed to Nagios I see the following in its logs:

[1357042122] EXTERNAL COMMAND: PROCESS_SERVICE_CHECK_RESULT;;zombieprocs;0;PROCS OK: 0 processes with STATE = Z
[1357042124] PASSIVE SERVICE CHECK:;zombieprocs;0;PROCS OK: 0 processes with STATE = Z

Did it solve my problems?

I listed the set of problems I wanted to solve so it’s worth evaluating if I did solve them properly.

Less load and RAM use on the Nagios nodes

My Nagios nodes have gone from load averages of 1.5 to 0.1 or 0.0, they are doing nothing, they use a lot less RAM and I have removed some of the RAM from the one and given it to my Jenkins VM instead, it was a huge win. The sender and receiver is quite light on resources as you can see below:

nagios    9757  0.4  1.8 130132 36060 ?        S     2012   3:41 ruby /usr/bin/mnrpes-receiver --pid=/var/run/mnrpes/ --config=/etc/mnrpes/mnrpes-receiver.cfg
nagios    9902  0.3  1.4 120056 27612 ?        Sl    2012   2:22 ruby /usr/bin/mnrpes-scheduler --pid=/var/run/mnrpes/ --config=/etc/mnrpes/mnrpes-scheduler.cfg

On the RAM side I now never get a pile up of many checks. I do have the stale detection enabled on my Nagios template so if something breaks in the scheduler/receiver/broker triplet Nagios will still try to do a traditional check to see what’s going on but that’s bearable.

Check frequency too low

With this system I could do my checks every 10 seconds without any problems, I settled on 60 seconds as that’s perfect for me. Rufus scheduler does a great job of managing that and the requests from the scheduler are effectively fire and forget as long as the broker is up.

Results are spread over 10 minutes

The problem with the results for load on node1 and node2 having no temporal correlation is gone too now, because I use MCollectives parallel nature all the load checks happen at the same time:

Here is the publisher:

I, [2013-01-01T12:00:14.296455 #20661]  INFO -- : scheduler.rb:26:in `nrpe' Publishing request for check_load with filter 'monitored_by=monitor1'

…and the receiver:

I, [2013-01-01T12:00:14.380981 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357041614] PROCESS_SERVICE_CHECK_RESULT;;load;0;OK - load average: 0.92, 0.54, 0.42|load1=0.920;9.000;10.000;0; load5=0.540;8.000;9.000;0; load15=0.420;7.000;8.000;0; 
I, [2013-01-01T12:00:14.383875 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357041614] PROCESS_SERVICE_CHECK_RESULT;;load;0;OK - load average: 0.00, 0.00, 0.00|load1=0.000;1.500;2.000;0; load5=0.000;1.500;2.000;0; load15=0.000;1.500;2.000;0; 
I, [2013-01-01T12:00:14.387427 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357041614] PROCESS_SERVICE_CHECK_RESULT;;load;0;OK - load average: 0.02, 0.07, 0.07|load1=0.020;1.500;2.000;0; load5=0.070;1.500;2.000;0; load15=0.070;1.500;2.000;0; 
I, [2013-01-01T12:00:14.388754 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357041614] PROCESS_SERVICE_CHECK_RESULT;;load;0;OK - load average: 0.07, 0.02, 0.00|load1=0.070;1.500;2.000;0; load5=0.020;1.500;2.000;0; load15=0.000;1.500;2.000;0; 
I, [2013-01-01T12:00:14.404650 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357041614] PROCESS_SERVICE_CHECK_RESULT;;load;0;OK - load average: 0.03, 0.09, 0.04|load1=0.030;1.500;2.000;0; load5=0.090;1.500;2.000;0; load15=0.040;1.500;2.000;0; 
I, [2013-01-01T12:00:14.405689 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357041614] PROCESS_SERVICE_CHECK_RESULT;;load;0;OK - load average: 0.06, 0.06, 0.07|load1=0.060;3.000;4.000;0; load5=0.060;3.000;4.000;0; load15=0.070;3.000;4.000;0; 
I, [2013-01-01T12:00:14.489590 #23631]  INFO -- : receiver.rb:34:in `receive_and_submit' Submitting passive data to nagios: [1357041614] PROCESS_SERVICE_CHECK_RESULT;;load;0;OK - load average: 0.06, 0.14, 0.14|load1=0.060;1.500;2.000;0; load5=0.140;1.500;2.000;0; load15=0.140;1.500;2.000;0;

All the results are from the same second, win.


So my scaling issues on my small site is solved and I think the way this is built will work for many people. The code is on GitHub and requires MCollective 2.2.0 or newer.

Having reused the MCollective and Rufus libraries for all the legwork including logging, daemonizing, broker connectivity, addressing and security I was able to build this in a very short time, the total code base is only 237 lines excluding packaging etc. which is a really low number of lines for what it does.