Langue: en

Autres versions - même langue

Version: 2010-01-14 (fedora - 01/12/10)

Section: 5 (Format de fichier)


collectd-python - Documentation of collectd's "python plugin"


   <LoadPlugin python>
     Globals true
   # ...
   <Plugin python>
     ModulePath "/path/to/your/python/modules"
     LogTraces true
     Interactive true
     Import "spam"
     <Module spam>
       spam "wonderful" "lovely"


The "python plugin" embeds a Python-interpreter into collectd and provides an interface to collectd's plugin system. This makes it possible to write plugins for collectd in Python. This is a lot more efficient than executing a Python-script every time you want to read a value with the "exec plugin" (see collectd-exec(5)) and provides a lot more functionality, too.

Currently only Python 2 is supported and at least version 2.3 is required.


LoadPlugin Plugin
Loads the Python plugin Plugin. Unlike most other LoadPlugin lines, this one should be a block containing the line ``Globals true''. This will cause collectd to export the name of all objects in the python interpreter for all plugins to see. If you don't do this or your platform does not support it, the embeded interpreter will start anywa but you won't be able to load certain python modules, e.g. ``time''.
Encoding Name
The default encoding for Unicode objects you pass to collectd. If you omit this option it will default to ascii on Python 2 and utf-8 on Python 3. This is hardcoded in Python and will ignore everything else, including your locale.
ModulePath Name
Appends Name to sys.path. You won't be able to import any scripts you wrote unless they are located in one of the directories in this list. Please note that it only has effect on plugins loaded after this option. You can use multiple ModulePath lines to add more than one directory.
LogTraces bool
If a python script throws an exception it will be logged by collectd with the name of the exception and the message. If you set this option to true it will also log the full stacktrace just like the default output of an interactive python interpreter. This should probably be set to false most of the time but is very useful for development and debugging of new modules.
Interactive bool
This option will cause the module to launch an interactive python interpreter that reads from and writes to the terminal. Note that collectd will terminate right after starting up if you try to run it as a daemon while this option is enabled to make sure to start collectd with the -f option.

The collectd module is not imported into the interpreter's globals. You have to do it manually. Be sure to read the help text of the module, it can be used as a reference guide during coding.

This interactive session will behave slightly differently from a daemonized collectd script as well as from a normal python interpreter:

1. collectd will try to import the readline module to give you a decent way of entering your commands. The daemonized collectd won't do that.
2. collectd will block SIGINT. Pressing Ctrl+C will usually cause collectd to shut down. This would be problematic in an interactive session, therefore this signal will be blocked. You can still use it to interrupt syscalls like sleep and pause but it won't generate a KeyboardInterrupt exception either.

To quit collectd send EOF (press Ctrl+D at the beginning of a new line).

3. collectd handles SIGCHLD. This means that python won't be able to determine the return code of spawned processes with system(), popen() and subprocess. This will result in python not using external programs like less to display help texts. You can override this behavior with the PAGER environment variable, e.g. export PAGER=less before starting collectd. Depending on your version of python this might or might not result in an OSError exception which can be ignored.
<Module Name> block
This block may be used to pass on configuration settings to a Python module. The configuration is converted into an instance of the Config class which is passed to the registered configuration callback. See below for details about the Config class and how to register callbacks.

The name identifies the callback.


Writing your own plugins is quite simple. collectd manages plugins by means of dispatch functions which call the appropriate callback functions registered by the plugins. Any plugin basically consists of the implementation of these callback functions and initializing code which registers the functions with collectd. See the section ``EXAMPLES'' below for a really basic example. The following types of callback functions are known to collectd (all of them are optional):
configuration functions
This type of functions is called during configuration if an appropriate Module block has been encountered. It is called once for each Module block which matches the name of the callback as provided with the register_config method - see below.

Python thread support has not been initialized at this point so do not use any threading functions here!

init functions
This type of functions is called once after loading the module and before any calls to the read and write functions. It should be used to initialize the internal state of the plugin (e. g. open sockets, ...). This is the earliest point where you may use threads.
read functions
This type of function is used to collect the actual data. It is called once per interval (see the Interval configuration option of collectd). Usually it will call plugin_dispatch_values to dispatch the values to collectd which will pass them on to all registered write functions. If this function throws any kind of exception the plugin will be skipped for an increasing amount of time until it returns normally again.
write functions
This type of function is used to write the dispatched values. It is called once for every value that was dispatched by any plugin.
flush functions
This type of function is used to flush internal caches of plugins. It is usually triggered by the user only. Any plugin which caches data before writing it to disk should provide this kind of callback function.
log functions
This type of function is used to pass messages of plugins or the daemon itself to the user.
notification function
This type of function is used to act upon notifications. In general, a notification is a status message that may be associated with a data instance. Usually, a notification is generated by the daemon if a configured threshold has been exceeded (see the section ``THRESHOLD CONFIGURATION'' in collectd.conf(5) for more details), but any plugin may dispatch notifications as well.
shutdown functions
This type of function is called once before the daemon shuts down. It should be used to clean up the plugin (e.g. close sockets, ...).

Any function (except log functions) may set throw an exception in case of any errors. The exception will be passed on to the user using collectd's logging mechanism. If a log callback throws an exception it will be printed to standard error instead.

See the documentation of the various register_ methods in the section ``FUNCTIONS'' below for the number and types of arguments passed to each callback function. This section also explains how to register callback functions with collectd.

To enable a module, copy it to a place where Python can find it (i. e. a directory listed in sys.path) just as any other Python plugin and add an appropriate Import option to the configuration file. After restarting collectd you're done.


The following complex types are used to pass values between the Python plugin and collectd:


The Config class is an object which keeps the informations provided in the configuration file. The sequence of children keeps one entry for each configuration option. Each such entry is another Config instance, which may nest further if nested blocks are used.
  class Config(object)

This represents a piece of collectd's config file. It is passed to scripts with config callbacks (see register_config) and is of little use if created somewhere else.

It has no methods beyond the bare minimum and only exists for its data members.

Data descriptors defined here:

This represents the parent of this node. On the root node of the config tree it will be None.
This is the keyword of this item, i.e. the first word of any given line in the config file. It will always be a string.
This is a tuple (which might be empty) of all value, i.e. words following the keyword in any given line in the config file.

Every item in this tuple will be either a string or a float or a boolean, depending on the contents of the configuration file.

This is a tuple of child nodes. For most nodes this will be empty. If this node represents a block instead of a single line of the config file it will contain all nodes in this block.


This should not be used directly but it is the base class for both Values and Notification. It is used to identify the source of a value or notification.
  class PluginData(object)

This is an internal class that is the base for Values and Notification. It is pretty useless by itself and was therefore not exported to the collectd module.

Data descriptors defined here:

The hostname of the host this value was read from. For dispatching this can be set to an empty string which means the local hostname as defined in collectd.conf.
The name of the plugin that read the data. Setting this member to an empty string will insert ``python'' upon dispatching.
Plugin instance string. May be empty.
This is the Unix timestamp of the time this value was read. For dispatching values this can be set to zero which means ``now''. This means the time the value is actually dispatched, not the time it was set to 0.
The type of this value. This type has to be defined in your types.db. Attempting to set it to any other value will raise a TypeError exception. Assigning a type is mandatory, calling dispatch without doing so will raise a RuntimeError exception.
Type instance string. May be empty.


A Value is an object which features a sequence of values. It is based on then PluginData type and uses its members to identify the values.
  class Values(PluginData)

A Values object used for dispatching values to collectd and receiving values from write callbacks.

Method resolution order:


Methods defined here:

dispatch([type][, values][, plugin_instance][, type_instance][, plugin][, host][, time][, interval]) -> None.
Dispatch this instance to the collectd process. The object has members for each of the possible arguments for this method. For a detailed explanation of these parameters see the member of the same same.

If you do not submit a parameter the value saved in its member will be submitted. If you do provide a parameter it will be used instead, without altering the member.

write([destination][, type][, values][, plugin_instance][, type_instance][, plugin][, host][, time][, interval]) -> None.
Write this instance to a single plugin or all plugins if ``destination'' is omitted. This will bypass the main collectd process and all filtering and caching. Other than that it works similar to ``dispatch''. In most cases ``dispatch'' should be used instead of ``write''.

Data descriptors defined here:

The interval is the timespan in seconds between two submits for the same data source. This value has to be a positive integer, so you can't submit more than one value per second. If this member is set to a non-positive value, the default value as specified in the config file will be used (default: 10).

If you submit values more often than the specified interval, the average will be used. If you submit less values, your graphs will have gaps.

These are the actual values that get dispatched to collectd. It has to be a sequence (a tuple or list) of numbers. The size of the sequence and the type of its content depend on the type member your types.db file. For more information on this read the types.db(5) manual page.

If the sequence does not have the correct size upon dispatch a RuntimeError exception will be raised. If the content of the sequence is not a number, a TypeError exception will be raised.


A notification is an object defining the severity and message of the status message as well as an identification of a data instance by means of the members of PluginData on which it is based.

class Notification(PluginData) The Notification class is a wrapper around the collectd notification. It can be used to notify other plugins about bad stuff happening. It works similar to Values but has a severity and a message instead of interval and time. Notifications can be dispatched at any time and can be received with register_notification.

Method resolution order:


Methods defined here:

dispatch([type][, values][, plugin_instance][, type_instance][, plugin][, host][, time][, interval]) -> None. Dispatch a value list.
Dispatch this instance to the collectd process. The object has members for each of the possible arguments for this method. For a detailed explanation of these parameters see the member of the same same.

If you do not submit a parameter the value saved in its member will be submitted. If you do provide a parameter it will be used instead, without altering the member.

Data descriptors defined here:

Some kind of description what's going on and why this Notification was generated.
The severity of this notification. Assign or compare to NOTIF_FAILURE, NOTIF_WARNING or NOTIF_OKAY.


The following functions provide the C-interface to Python-modules.
register_*(callback[, data][, name]) -> identifier
There are eight different register functions to get callback for eight different events. With one exception all of them are called as shown above.
callback is a callable object that will be called every time the event is triggered.
data is an optional object that will be passed back to the callback function every time it is called. If you omit this parameter no object is passed back to your callback, not even None.
name is an optional identifier for this callback. The default name is python.module. module is taken from the __module__ attribute of your callback function. Every callback needs a unique identifier, so if you want to register the same callback multiple time in the same module you need to specify a name here. Otherwise it's save to ignore this parameter identifier is the full identifier assigned to this callback.

These functions are called in the various stages of the daemon (see the section ``WRITING YOUR OWN PLUGINS'' above) and are passed the following arguments:
The only argument passed is a Config object. See above for the layout of this data type. Note that you can not receive the whole config files this way, only Module blocks inside the Python configuration block. Additionally you will only receive blocks where your callback identifier matches python.blockname.
The callback will be called without arguments.
register_read(callback[, interval][, data][, name]) -> identifier
This function takes an additional parameter: interval. It specifies the time between calls to the callback function.

The callback will be called without arguments.

The callback will be called without arguments.
The callback function will be called with one arguments passed, which will be a Values object. For the layout of Values see above. If this callback function throws an exception the next call will be delayed by an increasing interval.
Like register_config is important for this callback because it determines what flush requests the plugin will receive.

The arguments passed are timeout and identifier. timeout indicates that only data older than timeout seconds is to be flushed. identifier specifies which values are to be flushed.

The arguments are severity and message. The severity is an integer and small for important messages and high for less important messages. The least important level is LOG_DEBUG, the most important level is LOG_ERR. In between there are (from least to most important): LOG_INFO, LOG_NOTICE, and LOG_WARNING. message is simply a string without a newline at the end.

If this callback throws an exception it will not be logged. It will just be printed to sys.stderr which usually means silently ignored.

The only argument passed is a Notification object. See above for the layout of this data type.
unregister_*(identifier) -> None
Removes a callback or data-set from collectd's internal list of callback functions. Every register_* function has an unregister_* function. identifier is either the string that was returned by the register function or a callback function. The identifier will be constructed in the same way as for the register functions.
flush(plugin[, timeout][, identifier]) - None
Flush one or all plugins. timeout and the specified identifiers are passed on to the registered flush-callbacks. If omitted, the timeout defaults to "-1". The identifier defaults to None. If the plugin argument has been specified, only named plugin will be flushed.
error, warning, notice, info, debug(message)
Log a message with the specified severity.


Any Python module will start similar to:
   import collectd

A very simple read function might look like:

   def read(data=None):
     vl = collectd.Values(type='gauge')
     vl.dispatch(values=[random.random() * 100])

A very simple write function might look like:

   def write(vl, data=None):
     for i in vl.values:
       print "%s (%s): %f" % (vl.plugin, vl.type, i)

To register those functions with collectd:


See the section ``CLASSES'' above for a complete documentation of the data types used by the read, write and match functions.


Please feel free to send in new plugins to collectd's mailinglist at <collectd at> for review and, possibly, inclusion in the main distribution. In the latter case, we will take care of keeping the plugin up to date and adapting it to new versions of collectd.

Before submitting your plugin, please take a look at <>.


collectd is heavily multi-threaded. Each collectd thread accessing the python plugin will be mapped to a Python interpreter thread. Any such thread will be created and destroyed transparently and on-the-fly.

Hence, any plugin has to be thread-safe if it provides several entry points from collectd (i. e. if it registers more than one callback or if a registered callback may be called more than once in parallel).

The Python thread module is initialized just before calling the init callbacks. This means you must not use Python's threading module prior to this point. This includes all config and possibly other callback as well.
The python plugin exports the internal API of collectd which is considered unstable and subject to change at any time. We try hard to not break backwards compatibility in the Python API during the life cycle of one major release. However, this cannot be guaranteed at all times. Watch out for warnings dispatched by the python plugin after upgrades.


This plugin is not compatible with python3. Trying to compile it with python3 will fail because of the ways string, unicode and bytearray bahavior was changed.
Not all aspects of the collectd API are accessible from python. This includes but is not limited to meta-data, filters and data sets.


collectd(1), collectd.conf(5), collectd-perl(5), collectd-exec(5), types.db(5), python(1),


The "python plugin" has been written by Sven Trenkel <collectd at>.

This manpage has been written by Sven Trenkel <collectd at>. It is based on the collectd-perl(5) manual page by Florian Forster <octo at> and Sebastian Harl <sh at>.