Skip to content

Latest commit

 

History

History
331 lines (243 loc) · 19.2 KB

CREATING_MONITORS.md

File metadata and controls

331 lines (243 loc) · 19.2 KB
Raw

Creating a Scalyr Monitor Plugin

Introduction

Monitor Plugins are a feature of Scalyr Agent 2, the new Python-based agent framework created by Scalyr. These plugins are Python modules that execute within the Scalyr Agent 2 processes run on our user’s servers. Their primary purpose is to collect metrics, such as CPU usage, from the local server and report them to Scalyr. By default, Scalyr includes several builtin monitors that perform such tasks as collecting overall system metrics, as well as per-process metrics.

In this document, we will describe how you can write and run your own Monitor Plugin. We have strived to create a fairly simple interface for the plugins so that users will have a lot of freedom to do what they would like with them.

Setting up your environment:

To develop a Monitor Plugin, we strongly recommend you use pip and virtualenv to set up your development environment on a machine that does not already have Scalyr Agent 2 installed. If you would prefer to either use git to clone the Scalyr Agent 2 source code or develop against the Scalyr Agent 2 package already installed on your machine, please see the Alternate Setup Instructions at the bottom of this page.

If you do not have pip and virtualenv installed on your machine, please see the pip install instructions and the virtualenv install instructions.

First, create a directory to hold your new Monitor Plugin (for now):

mkdir ~/my-plugin
cd ~/my-plugin

Next, use virtualenv to create a development environment where you can install the Scalyr Agent 2 Python package locally:

virtualenv ENV
source ENV/bin/activate

Install the Scalyr Agent 2 Python package:

pip install scalyr-agent-2

You may optionally test that your environment is set up by running a test plugin:

python -m scalyr_agent.run_monitor scalyr_agent.builtin_monitors.test_monitor

If everything has been set up correctly, you should see lines being written to stdout containing the results of picking numbers from uniform and gaussian distributions.

Defining the Monitor Plugin

Next, you need to actually create your plugin. You can use test_monitor.py as a starting point. This file should be located in the builtin_monitors directory in the scalyr_agent package installed in your local environment. Generally, this is ENV/lib/python*/site-packages/scalyr_agent/builtin_monitors.

There are a few important considerations when creating your plugin:

  • Each Python module can only contain one Scalyr Monitor Plugin
  • Your plugin must be implemented as a class that derives from ScalyrMonitor
  • An instance of your class is created for each reference to your plugin in the config.
  • Each Monitor Plugin instance will run in its own thread.
  • All output must be recorded using the provided Logger instance, not stdout or stderr.
  • All metrics values reported by your Monitor Plugin will be written to a metric log filed shared by all instances of your Monitor Plugin.
  • All diagnostic and error messages reported by your Monitor Plugin will be written to the main agent log file.
  • The instance will be used until the config changes (by the user editing it).

To better illustrate how a plugin is written, let’s give an example of a fully working plugin and then go into details about the ScalyrMonitor class and its interface down below. For this example, we will create a file ~/my-plugin/random_coin_monitor.py with the following contents:

import random

from scalyr_agent import ScalyrMonitor

class RandomCoinMonitor(ScalyrMonitor):
    def _initialize(self):
        self.__counter = 0
        # Read two optional config fields.  You may also create a required
        # configuration  field by supplying the argument
        # ‘require_field=True’.  Then, if the user does not supply the
        # field in the monitor’s configuration, an exception will be raised.
        self.__gauss_mean = self._config.get('gauss_mean', 
                                             default=0.5,
                                             convert_to=float,
                                             min_value=0,
                                             max_value=10)
        self.__gauss_stddev = self._config.get('gauss_stddev',
                                               default=0.25,
                                               convert_to=float,
                                               min_value=0,
                                               max_value=5)


    def gather_sample(self):
        self.__counter += 1
        self._logger.emit_value('uniform', random.random(),
                                extra_fields={'count': self.__counter})
        self._logger.emit_value('gauss', random.gauss(self.__gauss_mean,
                                                      self.__gauss_stddev),
                                extra_fields={'count': self.__counter})

This plugin, when run, will record two metrics every sample interval, one based a uniform coin flip between 0 and 1, and another based on a gaussian distribution with an expected mean of 0.5. It will also keep a count of how many samples it has recorded, including that count in the reported metrics.

The ScalyrMonitor class

As you can probably guess, the ScalyrMonitor class is responsible for invoking the gather_sample method once per sample interval. It also provides several important instance variables that derived classes can use.

You can read the full documentation for ScalyrMonitor in the ENV/lib/python*/site-packages/scalyr_agent/scalyr_monitor.py file, but let’s go over a few of the important methods here.

Important ScalyrMonitor methods:

  • _initialize: Invoked during instance initialization. Derived classes may optionally override this method to initialize instance variables. Additionally, derived classes may verify the configuration parameters for the plugin as read from the config file is valid (stored in self._config). This method should throw an Exception if the config is not valid. Overriding this method is suggested as an alternative to overriding the base __init__ method.
  • run: Invoked to run the plugin. The default implementation will invoke self.gather_sample once every sample interval. Most plugins will not need to override this method, but they may do so. This method is invoked on a separate thread for each Monitor Plugin instance.
  • gather_sample: The default implementation of self.run will invoke this method once every sample interval. The derived classes are expected to override this method and perform any metric reporting required by the plugin.

As the example above illustrated, your plugin implementation will heavily rely on the instance variables defined by the ScalyrMonitor class.

Listed below are important ScalyrMonitor instance variables that can be used by derived classes:

  • _config: A MonitorConfig instance containing the entry for this Monitor Plugin instance in the “monitors” section of the configuration. This object works much like a dict but has extra functionality to help Monitor developers validate configuration options. See documentation for the MonitorConfig.get method in scalyr_monitor.py for more details or see the example above.

    The standard configuration fields for each monitor instance are “module” and “id”, but you may use any other field to allow users to provide configuration options specific to your plugin. Generally, _config should be read during initialization and if it is invalid, throw an Exception explaining the error. The get method can perform validation and will throw an appropriate Exception if needed.

  • _logger: The logging.Logger instance to use to report errors and, more importantly, metrics to Scalyr. All records created with INFO or higher will be sent to Scalyr. This _logger is specific to this Monitor Plugin instance and should be used only for the metrics and messages generated by it.

  • _sample_interval_secs: The number of seconds between calls to gather_sample. This may be overridden by the derived class during initialization. This defaults to 30 seconds, but when run by the run_monitor script (for testing) it is set to 5 seconds.

  • log_config: A dict containing the entry that determines how the metric log generated by this Monitor Plugin instance will be copied to Scalyr. This has the same fields as the entries in the “logs” section in the configuration. This may be modified by the derived class during initialization to change the configuration. For example, this can be used to add in additional attributes to every metric line copied to Scalyr, such as the parser to use. You may also use this to change the path of where the metric log will be written if the default does not suit your needs.

  • disabled: Whether or not the plugin should be run. This can be modified during initialization by the derived class to disable plugins that should not be run.

The AgentLogger class

We have defined our own logging class derived from logging.Logger named AgentLogger. This class has several important features that you should be aware of as developer. First, it provides an additional method for reporting metric values. You should use this method if possible because it will emit your metric in a standard format that can be parsed by the Scalyr servers. Otherwise, you will have to define your own parser on the Scalyr servers.

The method for reporting metrics is emit_value and has the following arguments:

  • metric_name: The name of the metric to report. It must start with a letter and can only contain alphanumeric characters, periods, and underscores.
  • metric_value: The value for the metric. Only int, long, float, boolean, str, and unicode are allowed.
  • extra_fields: An optional dict that if specified, will be included as extra fields on the logged line. These fields can be used in future searches/graphs expressions to restrict which specific instances of the metric are matched/aggregated together. The keys for the dict must be str and the only allowed value types are int, long, float, str, bool, and unicode.
  • monitor: The monitor instance reporting the metric. This defaults to the monitor the logger belongs to, so you typically will not need to supply it yourself.

Testing the Monitor Plugin

You must use the run_monitor.py tool to run your Monitor Plugin by itself for testing and debugging purposes. It creates and initializes the plugin instance in the same way Scalyr Agent 2 will, and sends all output (both metric and diagnostic) to stdout rather than to a log file. To speed up testing, it changes the sample interval time from 30 secs to 5 secs. It also has options to control the configuration to pass into the plugin instance as well as control the sample interval.

Here’s how we would use it to test our plugin:

python -m scalyr_agent.run_monitor random_coin_monitor

The output will look something like this:

2014-07-29 20:22:06.789Z INFO [monitor:random_coin_monitor()] [scalyr_monitor.py:158] Starting monitor
2014-07-29 20:22:06.789Z [random_coin_monitor()] uniform 0.982530747431 count=1
2014-07-29 20:22:06.789Z [random_coin_monitor()] gauss 1.07020127799 count=1
2014-07-29 20:22:11.790Z [random_coin_monitor()] uniform 0.730115799546 count=2
2014-07-29 20:22:11.790Z [random_coin_monitor()] gauss 0.340942612923 count=2

You can use Control-C to stop the process.

If you wish to change the Monitor configuration passed to your instance, you may supply a string containing a JSON object with the desired configuration. Here’s an example where we set the monitor’s id and supply a custom foo option:

python -m scalyr_agent.run_monitor -c "{ gauss_mean:8.5 }" random_coin_monitor

There also are options to change the sampling interval as well as the Python path use to locate plugins.
Run python -m scalyr_agent.run_monitor -h for more details.

Deploying your Plugin

To deploy your Monitor Plugin, you simply have to install it into a location where Scalyr Agent 2 will look for it and then add appropriate entries to the “monitors” section of your configuration file.

You may place your module in /usr/share/scalyr-agent-2/local/monitors. This path is always included in the Python search path when locating plugins.

Alternatively, you may make your own directory to hold your custom plugins, and then include the path to your directory in the additional_monitor_module_paths in your configuration file. This is a string that specifies additional paths to search for modules beyond the default locations. It defaults to empty string. It can contain multiple paths, separated by the system specific path separator (colon for Unix, semi-colon for Windows). Note, the PYTHONPATH, the Scalyr Agent 2 package, the local monitor path, and the contrib monitor path are always searched.

Be sure to set the permissions on your files such that they can be read by the user running the Scalyr Agent 2 process.

To actually run the Monitor Plugin, you simply have to add a new entry to the “monitors” section in your config. In our example, we would add the following:

{
…
   monitors: [ { 
       module: "random_coin_test",
       gauss_mean: 8.5,
       gauss_stddev: 5.0
   }]
…
}

You should not have to restart the agent to have the new monitor to begin running. The change should be noticed in 30 seconds. However, if you have changed the contents of your Python module, you may wish to restart the agent to ensure the changes the Python files are picked up.

Suggestions for writing Monitor Plugins

Here are are a few tips on writing high quality plugins that can be easily reused by other Scalyr customers.

  • Minimize your dependencies on non-standard libraries. Scalyr customers should not need to use 'pip' to install other Python packages to run your plugin. If you require a pure Python library, see if it can be included in the third-party directory in the Scalyr Agent package.
  • Make sure your plugin can run using Python 2.4 or greater. Many Scalyr customers monitor older systems that only support 2.4. Of course, if your plugin is monitoring something that would not be one these older systems, you may chose to not follow this rule.
  • Perform as much verification as you can during the _initialize method. This is the only plugin method that will be run before the daemon process is forked in the background (when starting the agent from the commandline). If you need to communicate an error message, this is the best time to do it since it will be shown to the user on stdout and it will prevent the agent from starting. Typically, you should be checking for errors that will not correct themselves without the user modifying the configuration file, such as an invalid configuration option or that your plugin cannot run on the current platform. Errors that may correct over time (such as a server being temporarily unavailable) do not need to be reported.
  • Anticipate common error scenarios and report clear error messages that also suggest what the user may do to fix the problem. For example, if your plugin is connecting to a local server to fetch pages and it returns a permission denied message, your error message should recommend that the user needs to change their server configuration to allow access to that page.

Important design considerations

Here are some items you should consider as you design and implement your Monitor Plugin.

Security

All plugins are run in the Scalyr Agent 2 process, which means that, by default, the plugin code will be executing commands as root. Care should be taken to only run plugins that you trust. Since you must explicitly reference a plugin in your config to run it (except for a few Scalyr-provided builtin in plugins), you should have clear visibility as to what you are trusting.

Log volume

Every log line and metric generated by your plugin will be sent to the Scalyr servers. You should take care to not log excessive, unimportant information since it does consume your allowed log volume. The AgentLogger class does have a rate limiter built into it to prevent excessive log, especially in the case of improperly behaving monitors.

Agent stability

Since the Monitor Plugins are running in the same process as Scalyr Agent 2, any problems with the monitors that cause the entire Python process to crash or exit will stop your agent from working. Scalyr Agent 2 does try to isolate the effects of monitors from one another by executing them in their own threads and catching thrown exceptions, but it cannot isolate all things.

Alternate Setup Instructions

Using git

If you plan on submitting your plugin to Scalyr for other customers to use or plan on building your own Scalyr Agent 2 RPM or Debian packages, you may wish to develop directly out of the Scalyr Agent 2 source code. You can do this by cloning the public scalyr-agent-2 repository and adding your own plugin source code.

First, clone the scalyr-agent-2 repository:

cd ~/
git clone https://github.com/scalyr/scalyr-agent-2.git --branch release

Next, add the source tree to your PYTHONPATH. The instructions to do this will be platform and shell dependent, but for Linux running bash, you just need to execute:

export PYTHONPATH=$PYTHONPATH:~/scalyr-agent-2/

You can then add your plugin to either the ~/scalyr-agent-2/monitors/contrib or ~/scalyr-agent-2/monitors/local directory. The contrib directory should be used for plugins you plan on submitting to Scalyr. The local directory should be used if the plugin is just for your own use. If you use the build_package.py script, any plugin you place in the local directory will be included in your built RPM or Debian package.

You may follow the rest of the instructions on how to build your plugin from above, just substituting the contrib or local directory for ~/my-plugin in the instructions.

Using already installed Scalyr Agent 2 package

If you are developing on a machine that has the Scalyr Agent 2 package already install on it (because you installed the RPM or Debian packages) then you can directly use that package instead of installing it again.

You simply have to set the PYTHONPATH to include the scalyr-agent-2/py directory, where ever that is installed on your machine.

For packages installed using RPM or Debian, it will be located in /usr/share/scalyr-agent-2/py:

export PYTHONPATH=$PYTHONPATH:/usr/share/scalyr-agent-2/py

For packages installed using the tarball method, it will be located in ~/scalyr-agent-2/py:

export PYTHONPATH=$PYTHONPATH:~/scalyr-agent-2/py

You may follow the rest of the instructions from above, including creating a ~/my-plugin directory to hold your plugin.