gestalt
Gestalt is a library for managing configuration information for
node.js applications. The main premise of gestalt is that the
underlying configuration for an application may change while the
application is still running. Gestalt provides a framework detecting
and reacting to these changes without having to completely restart
your application.
There are a couple of motivations for gestalt: Configuration of a
large software system is often complicated - there are of course many
tools out there for gathering configuration information from a bunch
of different sources. nconf for node is a good one, and gestalt is to
some extent based upon it. Gestalt is also influenced by configliere
for ruby and the configuration node structure of chef. However, there
are a couple of things that many of these tools do not do well. First,
configuration files (and other sources) can change, and it would be
nice to be able to react to these changes on-the-fly. Second, for a
sufficiently complicated system of default and override configuration
sources, it can become difficult to figure out exactly where a
particular setting came from.
Gestalt solves both of these problems. It has a per-value event change
tracking system so that you can track changes to individual settings
to your configuration. It also rigorously keeps track of where the
values for particular settings came from.
Basics
The basic object is a Configuration:
var gestalt = require('gestalt');
var options = {source: 'Source'};
var config = new gestalt.Configuration( options );
config.set("x", 'stuff');
config.set("y", 6, 'HERE');
var x = config.get("x");
var y = config.get("y");
Configuration names are hierarchical - use a ':' to delimit namespaces.
The namespaces become nested Configuration objects.
config.set("owner:name", 'Joe');
config.set("owner:phone", '5551234');
var ownerConfig = config.get('owner');
var owner = ownerConfig.toObject();
Values can be primative values (numbers, strings, booleans,
etc.). By default, assignments of structured objects get destructured into nested
Configuration objects.
config.set("neighbor", {name: 'Fred', phone: '5559876'} );
var fred_phone = config.get( 'neighbor:phone' );
( The default destructuring of assignments can be disabled. See
the destructure_assignments and destructure_arrays options, below. )
It is possible to turn a configuration object back into a regular
object. Also, if a configuration object looks like an array (all
integer keys...) toObject will in fact return an array.
Events
Configuration objects are EventEmitters. When a value of a
configuration object changes, it emits a 'change' event.
config.on('change', function( change) {
console.log("name: %s, value: %s, old_value: %s, source: %s ",
change.name, change.value, change.old_value, change.source);
});
config.set("owner:phone", "5554444", "phone book");
The nested configuration objects are also event emitters. Note that
the configuration names in the events are reported relative to the
configuration object being listening to.
Configuration objects also have a ready/invalid/other state. When
everything about the configuration is loaded the way it expects that
it should be, it enters a 'ready' state. If something has gone wrong
in a way that might require some attention, it enters an 'invalid'
state. Other states are possible - most configuration objects start in
a 'not ready' state, but particular implementations might add some other
special states.
State propagates from children to parents and from contained objects
to containers. The parent/container will become invalid if any of its
children/contents become 'invalid'. Also, a parent/container will only become
'ready' if all of its children/contents are 'ready'. If no sub-configuration
is invalid, but not all of them are ready, parents and containers become
'not ready'.
On state transitions (and whenever further 'invalid' states are
encountered) configuration objects will emit a 'state' message. The
payload object is of the form:
{ state: 'ready',
old_state: 'not ready',
data: 'arbitrary data that might help explain the transition'
}
To reiterate, there are three main states a Configuration object can
be in:
-
'ready' This means that there were no problems getting to the
configuration data. You can read data from it, and set up
listeners. Note that a configuration can only be in a ready
state if all of its dependencies are also ready.
-
'invalid' Something has gone wrong. We may not be able to get in
contact with the data source, or there may have been a problem
parsing the data. In any case, the data contained in the
configuration object may be corrupt, out of date, or wrong. When a
configuration object goes into an 'invalid' state, the program should
stop consuming configuration data from the object and seek a way to remedy
the invalid state. If any of an objects depencencies is in an invalid
state, the object will also be invalid.
-
'not ready', (and others). The configuration object is not ready for reads.
It will send out another 'state' event when it is ready or when something
actually goes wrong. 'not ready' states are not an error state, but while
a configuration is in a 'not ready' state, you may not be able to rely on
the contents to be accurate, up to date, or even present.
Containers
The ConfigContainer class gives you a way to set up a system of
defaults and overrides of configuration information that comes from
different sources. For instance, if you allow configuration parameters
to be set at the command line, in environment variables, or from a
configuration file, this class can help you out.
var override = new gestalt.Configuration({source: "Override"});
var def = new gestalt.Configuration({source: "Default"});
var container = new gestalt.ConfigContainer({source: "Container"});
container.addOverride( override );
container.addDefault(def);
container.on( 'change',function(change) {
console.log("%j", change );
});
def.set("a",1);
container.set("a",2);
override.set("a",3);
container.set("a",4);
def.set("a",5);
container.remove("a");
override.remove("a");
Remap Configs
Configuration files are often structured, while environment variables
and command line arguments are usually not structured. However, it is
often the case that a program will use an environment variable or a
command line argument to override a setting in a structured
document. RemapConfig objects give you a formal way to show exactly
what part of the structure a given command line argument will override.
The constructor for RemapConfig expects to see a couple of options in
the options argument. First, it needs a reference to the original
configuration object that is being remapped. Second, it needs a function
that will map names from the original object into the new object space.
var gestalt = require('../lib/gestalt'),
Configuration = gestalt.Configuration,
RemapConfig = gestalt.RemapConfig;
function mapper(old) {
if( old.match(/^f/ )) {
return "new:" + old;
} else {
return undefined;
}
}
var c = new Configuration();
c.set("foo",1);
c.set("gak",4);
var r = new RemapConfig( { mapper: mapper, original: c } );
console.log( r.get('new:foo') );
The mapper can be a function as above, or it can be a simple
java object with old values as keys and new values as corresponding
values. Keys not present in the object will be mapped out of the
RemapConfig.
Not surprisingly, there are a couple of restrictions on this type of
configuration object. First, it is read only. Second, the mapper
function can show that it ignores part of the object space by
returning undefined for some values. For the rest of the values, it
must make sure to return unique new names for different old names.
Remapped objects do pass on events, and can be used as overrides or
defaults in a config container.
API
Configuration
A Configuration object is a container of name value pairs. The names
can include colon separated hierarchical namespaces.
- constructor Configuration( options )
Creates a configuration object. Options include
The default source for changes to this object. If no source is given
for a set operation on the Configuration, the default is used
instead. Note that only the initial value of this option is
important. If it is changed with the options() method, it will not
change the default source for an object.
By default, this option is on. When turned on, if you try to assign
an object or an array to a name in the configuration, it will
destructure the object into nested namespace structures.
This option is similar to destructure_assignments, only it only controls
whether or not to destructure array objects. It is true by default - when changed
to false, arrays will be valid configuration values.
Note, if you turn off destructure_arrays or destructure_assignments, the configuration
object will not react to internal changes in the structured values. In particular;
config.set("a",[1,2,3]);
congig.get("a")[0] = 7
State to transition to after initialization. Defaults to 'ready' for
many configuration objects. Defaults to 'not ready' for files,
zookeeper, and other objects that load asynchonously. Note that the
configuration object is created in a 'not ready' state, and the
transizion to its initial_state actually happens in the next tick after
creation.
Methods
Returns the value assigned to 'name', or undefined if not
present. Namespaces are separated by colons. If name is an array, it
is treated the same as if it were a single string joined together by
colons.
The same as get, only instead of the assigned value, it returns an
object that contains a value and a source field. The value is the
same as the value one would expect from get. The source field
contains the source of the value.
Sets the value of name. If no source is given, the default source for
the Configuration is used. If the configuration object does not have
a default source, then the source becomes a reference to the line
number and file of the function calling set.
- update(name, value, source)
For standard Configuration objects, update works the same as set,
except that if no source is provided in the update call, the
configuration's default source is not used. For ConfigContainers,
update and set have different behaviors.
Returns true if there is a value defined for the given name.
Returns an array containing all of the keys that have assigned values
in the configuration.
- each( function( value, name, configuration ) )
Iterates through the keys and values of the configuration, calling
the supplied function once for each setting.
Deletes the name from the configuration.
Generates a detailed report on the given stream (or stdout) of all of
the names and values in YAML format.
- writeFile( filename, format )
Attempts to open a writeStream on the filename and writes the contents
of the configuration object to the file in the specified format. If format
is left out, 'json' is assumed. Valid options are 'json', 'yaml' and a
function that takes a plain javascript object (the results of toObject() )
and returns a string or a buffer object.
Builds a javascript object representing the current state of the
Configuration. Namespaces are converted to nested objects. If a
namespace has the appearance of an array in that its internal names
are sequential numbers starting with 0, it will be converted into an
array instead of a regular object. The resultant object will not
generate any events and will not have source attribution.
Returns the options object for the Configuration. If passed an
object, it will use it to override the existing options. Note that
changing the options of a Configuration does not change the options
of existing nested namespace configurations, but it will affect any
namespaces created after the change.
- addPatternListener(pattern, callback )
Calls the callback whenever there is a change to the configuration
that matches the pattern. Pattern could be a string, or a regex, in
which cases match means that the name of the changed value either
equals the string or matches the regex respectively. Pattern could
also be a function which return a truthy value if the change matches
its criteria.
- removePatternListener( callback )
Removes a previously added listener callback.
If no arguments are given, it returns the current state of the object.
If a state it given, it transitions the configuration to that
state. If the object's state has actually changed (or if we get
another 'invalid' when it was already in an invalid state) it will
emit a 'state' event, passing the data argument on in the state change
payload along with the new and old states.
Events
Change events are emitted whenever the Configuration object detects
that something has changed in the data. The handlers to these events
are passed an object describing the change:
change = { name: "a:b:c",
value: 5,
old_value: undefined,
source: "X"};
When the object's state changes between 'invalid', 'ready' and 'not
ready', it will emit 'state' events of the form:
state = { state: 'ready',
old_state: 'invalid',
data: 'good now'
};
ConfigContainer
gestalt's mechanism to deal with the notion of default and override
behavior is implemented in the ConfigContainer object. ConfigContainers
contain a list or priority ordered configuration objects. Calling 'get(name)'
on the ConfigContainer will return the value assigned to the highest priority
configuration that has a value set for that name.
- constructor ConfigContainer( options )
Takes the same options as a Configuration object, but will also accept
-config
A configuration object to act as the 'normal' object. See 'set' and
'update' below. If no config option is given, a new object will be
created as the 'normal' object;
Methods
All of the public methods of the Configuration object should work on a
ConfigContainer, however the semantics of a few of them are a little
different.
Returns the value associated with the name of the highest priority
object containing a defined value for the name. Actually, the real
contract is that the results of get will agree with the latest change
event for the object, which is effectively the highest priority
contained object's values, with the following caveat: unlike regular
Configuration objects, it is not well defined what will happen if you
get an intermediate Configuration value.
var container = new gestalt.ConfigContainer();
container.set("a:b:c",7);
container.get("a:b:c")
container.get("a:b")
Calls 'set' on the 'normal' priority configuration object.
- update(name, value, source)
Calls 'update' on the highest prior configuration that has a value
already set for 'name'. If none have defined 'name', then call 'update'
on the 'normal' priority object.
Returns the union of all of the keys in all of the contained configuration
objects.
Returns true if the name is in the result of calling keys()
Works the same as Configuration's each method, but the values are as
determined by the highest priority configuration object that has a
value for the given key.
Adds a configuration object to the priority list of configurations
as the highest priority object.
Adds a configuration object to the priority list as the lowest
priority config object.
RemapConfig
The RemapConfig object provides a way to change the names of a
configuration without changing the values. This useful for changing
the names that come from environment variables or command line
variables into names that match up with the configuration hierarchy
established in a configuraiton file - making it possible to use
the override and default behaviors from a ConfigContainer object.
- constructor RemapConfig( options )
Takes the same options as a Configuration object, but will also accept
-config (mandatory)
The configuration object to remap.
-mapper (mandatory)
Specifies how to remap configuration names. This can either be a
function of the form:
remap = function(old_value) {
var new_value = "a:b:c:" + old_value;
return new_value;
}
or it can be a flat javascript object with old names as keys and new
names as values.
Names that get mapped to 'undefined' by a remap function, or that are
not present in the remap object, will simply not be included in the
resultant object.
-reverse (optional)
Specifies how to map new names back into old names. The reverse
function is calculated automatically if a structure mapper was
used. Reverse and Mapper should be inverses of each other over the
range of kkeys that are not mapped out of the model. If the reverse
function is not given, set and update functions will only be able to
modify existing values in the structure. With the reverse function,
they will also be able to set new values and have them map back to
the original structure.
Methods
All of the Configuration public methods are supported, with the following
additions and modifications:
Returns a reference to the unmapped configuration object.
Does nothing - remap objects are read only at this time.
Does nothing - remap objects are read only at this time.
Does nothing - remap objects are read only at this time.
ConfigArgs
This is a standard Configuration object that pulls its name and value
pairs from parsing the command line arguments with the optimist
library. In addition to the standard configuration options, it will also
accept the following:
Use the array instead of the arguments in process.argv
String to pass on to optimist as a usage string. ( Uses optimist's
"usage" method. )
Object to pass on to optimist as a configuration options. ( Uses
optimist's "usage" method. )
ConfigEnv
This is a standard Configuration object that pulls its name and value
pairs from parsing the environmental variables. In addition to the
standard configuration options, it will also accept the following:
Use this set of name value pairs instead of process.env.
ConfigFile
This is a standard Configuration object that draws its names and values
from a configuration file. In addition to the standard options, ConfigFiles
accept the following:
Tells what format the file is in. Current options are 'json', 'yaml'
and 'ini'. By default, the format will be 'auto', which will try to guess
the format of the file based on its file extension. If you specify 'raw'
as a format, the contents of the file will be added as a string to the
'contents' name of the configuration object.
Normally, the format of the file determines what to use as a parser. This option
can be used to override exactly how to turn the contents of a file into
configuration. It should be a function that can accept the data from a file read
and convert it into a raw javascript object.
Tells what file to read - gets passed to fs.readFile.
Boolean. If set to true, the constructed ConfigFile object will set up
a watch for changes to the underlying file. If it changes on disk,
ConfigFile will reload the file and update any changed values.
ZookeeperConfig
ZooKeeper (http://zookeeper.apache.org/) is "a centralized service for
maintaining configuration information, naming, providing distributed
synchronization, and providing group services". The ZookeeperConfig
object assists in integrating zookeeper services into an overall configuration
package.
ZookeeperConfig does not really assist with writing information to zookeeper,
only to reacting to information as it changes on the zookeeper servers.
Much like the ConfigFile object, the ZookeeperConfig object is just a standard
Configuration object with a couple of extra options and methods.
This configuration object relies on the 'zookeeper' npm package. This dependency
is not listed in the gestalt package dependencies, and is only required when
a ZookeeperConfig object is first instantiated.
Zookeeper has a slightly different idea about hierarchy from other
configuration systems: in general every node can have both a value and
children. ZookeeperConfig manages this by adding two sub-configuration
objects to a configuration representing a a zookeeper node - one
called 'data' and the other called 'children'. 'data' will contain
whatever comes back from parsing the data in the zookeeper
node. 'children' contains names corresponding to the relative names of the
zookeeper nodes children and values of more ZookeeperConfig objects
corresponding to the zookeeper child nodes.
- constructor ZookeeperConfig( options )
The options are the same as for Configuration with the following additions:
The source should be a string of the form 'zk://host1:port1,host2:port2/path/to/config'.
What format is the data stored in on zookeeper nodes. 'raw' means that
the data for a given node will be placed in a javascript string object and
stored under the name "data" for that node. Other options are 'json', 'ini',
and 'yaml'.
The same as for ConfigFile - you can provide your own parser.
Use this option to hand off an existing zookeeper connection
If the ZookeeperConfig object is to create its own zookeeper
connection, these options will be passed to the constructor for
ZooKeeper (from the npm package).
Boolean. If true, once connected to zookeeper, ZookeeperConfig will
create the path of the zookeeper node it is trying to listen to, if
it is not already there.
Boolean. If true, zookeeper nodes will include a 'stat' name in addition
to 'data' and 'children'. The stat object will contain the stats reported
by the node's data callback.
Methods
- zookeeper( function(zk) )
The callback will be called when a zookeeper connection becomes
available, or immediately if it is already available. This method can
be used to make zookeeper calls on the same connection that is being
used by the ZookeeperConfig object.