As pointed out in the SimpleRPCIntroduction page you can use the
mco rpc CLI
to call agents and it will do it’s best to print results in a sane way. When
this is not enough you can write your own clients.
Simple RPC clients can do most of what a normal client can do but it makes a lot of things much easier if you stick to the Simple RPC conventions.
This guide shows how to write standalone scripts to interact with your collective. There is a single executable system. You can apply most of the techniques documented here to writing plugins for that application system. See the full reference for the plugin system here. You should try to write your general agent CLIs using this plugin system rather than the stand alone scripts detailed below as that promote a unified interface that behave in a consistent manner.
We’ll walk through building a ever more complex example of Hello World here.
The client is mostly a bunch of helper methods that you use as a Ruby Mixin in your own code, it provides:
MCollective::Clientif you need the additional feature sets
We’ll write a client for the
Helloworld agent that you saw in the
A basic hello world client can be seen below:
Save this into
hello.rb and run it with
--help, you should see the standard basic help including filters for discovery.
If you’ve set up the Agent and run the client you should see something along these lines:
While it ran you would have seen a little progress bar and then just the summary line. The idea is that if you’re talking to a 1000 machine there’s no point in seeing a thousand
OK, you only want to see what failed and this is exactly what happens here, you’re only seeing errors.
If you run it with
--verbose you’ll see a line of text for every host and also a larger summary of results.
I’ll explain each major line in the code below then add some more features from there:
The first line pulls in the various helper functions that we provide, this is the Mixin we mentioned earlier.
We then create a new client to the agent “helloworld” that you access through the
To call a specific action you simply have to do
mc.echo this calls the
echo action, we pass a
:msg parameter into it with the string we want echo’d back. The parameters will differ from action to action. It returns a simple array of the results that you can print any way you want, we’ll show that later.
printrpcstats are functions used to print the results and stats respectively.
This cleanly disconnects the client from the middleware, some middleware tools like ActiveMQ will log confusing exceptions if you do not do this. It’s good form to always disconnect but isn’t strictly required.
As you see there’s no indication that discovery is happening and as pointed out we do not display results that are ok, you can force verbosity as below on individual requests:
Here we’ve added a
:verbose flag and we’ve specifically called the discover method. Usually you don’t need to call discover it will do it on demand. Doing it this way you’ll always see the line:
Passing verbose to
printrpc forces it to print all the results, failures or not.
If you just wanted to force verbose on for all client interactions, do:
In this case everything will be verbose, regardless of command line options.
You can disable the twirling progress indicator easily:
Now whenever you call an action you will not see the progress indicator.
You can retrieve the stats from the clients and also get text of reports without printing them:
report will now have the text that would have been displayed by ‘printrpcstats’ you can also use
no_response_report to get report text for just the list of hosts that didnt respond.
If you didn’t want to just print the results out to STDOUT you can also get them back as just text:
You can pass filters on the command line using the normal
--with-* options but you can also do it programatically. Here’s a new version of the client that only calls machines with the configuration management class
/dev_server/ and the fact
You can set other filters like
fact_filter method supports a few other forms in adition to above:
This will limit it to all machines in the UK with more than 3 processors.
If while using the client you wish to reset the filters to an empty set of filters - containing only the agent name that you’re busy addressing you can do it as follows:
After this code snippet the filter will only have an agent filter of
By default the client will communicate with all machines at the same time. This might not be desired as you might affect a DOS on related components.
You can instruct the client to communicate with remote agents in batches and sleep between each batch.
Any client application has this capability using the
command line options.
You can also enable this programatically either per client or per request:
By default batching is disabled and sleep time is 1
batch_size to 0 will disable batch mode in both examples above,
effectively overriding what was supplied on the command line.
By default it will only do discovery once per script and then re-use the results, you can though force rediscovery if you had to adjust filters mid run for example.
Here we make one
echo call - which would do a discovery - we then reset the client, adjust filters and call it again. The 2nd call would do a new discovery and have new client lists etc.
A new core messaging mode has been introduced that enables direct non filtered communicatin to specific nodes. This has enabled us to provide an discovery-optional mode but only if the collective is configured to support direct messaging.
This will immediately, without doing discovery, communicate just with these 3 hosts. It will do normal failure reporting as with normal discovery based requests but will just be much faster as the 2 second discovery overhead is avoided.
The goal with this feature is for cases such as deployment tools where you have a known expectation of which machines to deploy to and you always want to know if that fails. In that use case a discovery based approach is not 100% suitable as you won’t know about down machines. This way you can provide your own source of truth.
When using the direct mode messages have a TTL associated with them that defaults to 60 seconds. Since 1.3.2 you can set the TTL globally in the configuration file but you can also set it on the client:
With the TTL set to 3600 if any of the hosts are down at the time of the request the request will wait on the middleware and should they come back up before 3600 has passed since request time they will then perform the requested action.
By default all nodes that get discovered will get the request. This isn’t always desirable maybe you want to deploy only to a random subset of hosts or maybe you have a service exposed over MCollective that you want to treat as a HA service and so only speak with one host that provides the functionality.
You can limit the hosts to talk to either using a number or a percentage, the code below shows both:
This will pick 10% of the discovered hosts - or 1 if 10% is less than 1 - and only target those nodes with your request. You can also set it to an integer.
There are two possible modes for choosing the targets. You can configure a global default method but also set it on your client:
The above code will force a
:random selection, you can also set it to
If you wanted to work with the Client directly as in WritingAgents after perhaps setting up some queries or gathering data first you can gain access to the client, you might also need access to the options array that was parsed out from the command line and any subsequent filters that you added.
The first call will set up the CLI option parsing, create clients etc, you can then just grab the client and options and go on as per WritingAgents. This is a much quicker way to write full power clients, by just by short-circuiting the options parsing etc.
The biggest reason that you’d write custom clients is probably if you wanted to do custom processing of the results, there are 2 options to do it.
Results have a set structure and depending on how you access the results you will either get Exceptions or result codes.
|Status Code||Description||Exception Class|
|1||OK, failed. All the data parsed ok, we have a action matching the request but the requested action could not be completed.||RPCAborted|
Just note these now, I’ll reference them later down.
Simple RPC provides a trimmed down version of results from the basic Client library. You’d choose to use this if you just want to do simple things or maybe you’re just learning Ruby. You’ll get to process the results after the call is either done or timed out completely.
This is an important difference between the two approaches, in one you can parse the results as it comes in, in the other you will only get results after processing is done. This would be the main driving facter for choosing one over the other.
Here’s an example that will print out results in a custom way.
This will produce a result something like this:
each in the above code just loops through the array of results. Results are an array of Hashes, the data you got for above has the following structure:
:statuscode matches the table above so you can make decisions based on each result’s status.
You can get access to each result in real time, in this case you will supply a block that will be invoked for each result as it comes in. The result set will be exactly as from the full blown client.
In this mode there will be no progress indicator, you’ll deal with results as and when they come in not after the fact as in the previous example.
The output will be the same as above
In this mode the results you get will be like this:
You can additionally gain access to a SimpleRPC style result in addition to the more complex native client results:
You can still use printrpc to print these style of results and gain advantage of the DDL and so forth:
This should give you a simpler result set to deal with
You can look at the
mco rpc script for a big sample, here I am just adding a simple
--msg option to our script so you can customize the message that will be sent and received.
This version of the code should be run like this:
Documentation for the Options Parser can be found in its code.
And finally if you add options as above rather than try to parse it yourself you will get help integration for free:
Sometimes, perhaps when embedding an MCollective client into another tool like Puppet, you do not want MCollective to do any command line parsing as there might be conflicting command line options etc.
This can be achieved by supplying an options hash to the SimpleRPC client:
This will create a RPC client for the agent test without any options parsing at all.
To set options like discovery timeout and so forth you will need use either the client utilities or manipulate the hash upfront, the client utility methods is the best. The code below sets the discovery timeout in a way that does not require you to know any internal structures or the content of the options hash.
Using this method of creating custom options hashes mean we can make internal changes to MCollective without affecting your code in the future.
Usually this section will not apply to you. The client libraries support sending a request without waiting for a reply. This could be useful if you want to clean yum caches but don’t really care if it actually happens everywhere.
You will loose these abilities:
The above should make it clear already that this is a limited use case, it’s essentially a broadcast request with no feedback loop.
The code below will send a request to the
runonce action for an agent
puppetd, once the request is dispatched I will have no idea if it got handled etc, my code will just continue onwards.
This will honor any attached filters set either programatically or through the command line, it will send the request but will just not handle any responses. All it will do is return the request id.
For web applications you’d probably use cached copied of Registration data to do discovery rather than wait for MC to do discovery between each request.
To do this, you’ll need to construct a filter and a list of expected hosts and then do a custom call:
This will do a call with exactly the same stats, block and other semantics as a normal call like:
But instead of doing any discovery it will use the host list and filter you supplied in the call.
A helper agent called
rpcutil is included that helps you gather stats, inventory etc about the running daemon.