Jcliff configures a running instance of EAP6/JBoss7 using modular configuration files.
There are several problems with that:
- It is not easily maintainable, especially if you have many hosts with different configurations.
- It requires a bounce. JCliff uses jboss-cli.sh to invoke configuration commands, so many configuration modifications can be done without bouncing the server.
JCliff modifies the configuration incrementally. Only necessary parts of the configuration is modified, and if the required changes are already applied, nothing is done.
Use puppet to lay down configuration files, and then have puppet execute jcliff. This way, you can define datasources, logging, etc. using puppet code and templates, and configure only what's necessary.
There's nothing puppet specific in jcliff. You can use jcliff as a command line tool, or from any configuration management system than can install files, and execute other programs.
JCliff uses the same JSON-like language used by jboss-cli.sh (jboss-dmr library). A configuration file that sets the root level logging looks like
{ "logging" => { "root-logger" => { "ROOT" => { "level" => "debug" } } } }
EAP6 configuration is difficult if you intend to use puppet. The command line client provides an interactive front-end for configuration tasks, not easily called from puppet scripts. There is no easy way to retrieve what's already configured, find the differences between the current state and the desired state, and come up with a way to implement those. Jcliff does exactly that. User supplies the desired configuration, jcliff executes the jboss client to retrieve the current state, derive deltas, and apply them. What kind of delta results in what kind of action is defined using a property file based rule language.
The JBoss DMR library is used to represent jboss configuration in a hierarchical markup language.
Look at this logging configuration fragment as example:
> /subsystem=logging:read-resource(recursive=true)
{
"outcome" => "success",
"result" => {
"custom-handler" => undefined,
"async-handler" => {
"blah" => {
"encoding" => undefined,
"filter" => undefined,
"formatter" => "%d{HH:mm:ss,SSS} %-5p [%c] (%t) %s%E%n",
"level" => undefined,
"overflow-action" => "BLOCK",
"queue-length" => 1000,
"subhandlers" => undefined
},
"ASYNC" => {
"encoding" => undefined,
"filter" => undefined,
"formatter" => "%d{HH:mm:ss,SSS} %-5p [%c] (%t) %s%E%n",
"level" => undefined,
"overflow-action" => "BLOCK",
"queue-length" => 1000,
"subhandlers" => ["FILE"]
}
},
...
}
}
The command
/subsystem=logging:read-resource(recursive=true)
is parsed as follows:
- Select the subsytem whose name is logging. That's a node in the configuration tree
:read-resourceis an operation defined on that node. Invoke the operation with attributes(recursive=true)
As for the return result:
- Anything between
{and}is an object. So the return result is an object containing two items,outcomeandresult. resultis another object, containingcustom-handler,async-handler, etc. all of which are objectsasync-handlercontains objectsASYNCandblah.ASYNC/subhandlersis a list of strings. Anything between[and]are lists.
Every node in the configuration tree defines a set of operations that
can be run on that node. For instance, to add a new async-handler, you
have to invoke:
/subsystem=logging/async-handler=newHandler:add(queue-length=someNumber)
Here's the primary reasons why it is not easy to automate the configuration tasks:
- Every node defines a separate set of operations.
- Every operation has a different set of required parameters
That is, there is no standard way of adding/removing nodes to the
configuration tree. For instance, to add an element to the handlers
list of a logger, you have to call assign-subhandler on that node. To
assign a handler to root logger, you have to call
root-logger-assign-subhandler on that root logger.
/subsystem is not a universal prefix. That would be too easy. For
instance, to get system properties:
/core-service=platform-mbean/type=runtime:read-attribute(name=system-properties)
The idea is to have puppet lay configuration files to a given directory, and then run jcliff on those files. Jcliff loads the puppetized configuration files, talks to JBoss, determines what needs to be changed, and changes them. Each puppetized configuration file has to tell what it is configuring. For instance, a logging configuration file looks like:
{ "logging" =>
{
"async-handler" => {
"ASYNC" => {
"subhandlers" => [ "FILE" ],
"queue-length" => 1000,
},
},
"size-rotating-file-handler" => {
"SFILE" => { }
}
}
}
...
Similarly, jdbc drivers:
{ "jdbc-driver" =>
{ "oracle" => {
"driver-name" => "oracle",
"driver-module-name" => "oracle.jdbcx",
"driver-xa-datasource-class-name" => "oracle.jdbc.XADataSource"
}
}
}
Datasources:
{
"datasource" => {
"BSProduct" => {
"jndi-name" => "java:/BSProduct",
"connection-url" => "jdbc:oracle:oci@web",
"driver-name" => "oracle",
"user-name" => "web",
"password" => "web_dev0",
}
}
}
System properties:
{ "system-property" => {
"foo" => "bar",
"bah" => "gah"
}
}
Jcliff does not delete anything from the existing configuration unless
explicitly required. Therefore, not specifying certain properties of
objects will leave them untouched. If you want to delete them, assign
objects/values to "deleted", or undefine them, by assigning them to
undefined.
Jcliff can be used to deploy applications. After applying all the configuration changes, Jcliff attempts to process deployments of the form:
{ "deployments" => {
"myApp-v2.1.ear" => {
"NAME" => "myApp-v2.1.ear",
"path" => "/var/lib/redhat/deploy/myApp-v2.1.ear",
"replace-name-regex" => "\\QmyApp-v\\E\\..*",
}
}
}
Jcliff runs deploy -l to retrieve the list of deployed packages. The
idea is to have Jcliff first undeploy older versions of the same
application, and then to redeploy the new version. The
replace-name-regex and replace-runtime-name-regex regular expressions
are used to locate applications that the new application will
replace. Any applications whose name matches replace-name-regex, or
whose runtime name matches replace-runtime-name-regex will be
undeployed before redeploying the new application. In the above
example, myApp-v2.1.ear would replace an existing myApp-v2.0.ear.
If myApp-v2.1.ear already is deployed, Jcliff will not attempt to
redeploy it, unless --redeploy flag is passed. So, after deploying all
the applications, running Jcliff without the --redeploy flag with the
existing deployment list will not alter any of the deployments.
Jcliff reads all configuration files, and builds a list of paths. Every value in the configuration tree is represented by a path containing all the object names up to that value. For instance:
{ "system-property" => {
"foo" => "bar",
"bah" => "gah"
}
}
Paths:
system-property
system-property/foo
system-property/bah
For the datasource example:
{
"datasource" => {
"BSProduct" => {
"jndi-name" => "java:/BSProduct",
"connection-url" => "jdbc:oracle:oci@web",
"driver-name" => "oracle",
"user-name" => "web",
"password" => "web_dev0",
}
}
}
Paths:
/datasource
/datasource/BSProduct
/datasource/BSProduct/jndi-name
/datasource/BSProduct/connection-url
/datasource/BSProduct/driver-name
/datasource/BSProduct/user-name
/datasource/BSProduct/password
Same thing is also done for the configurations retrieved from JBoss. For instance, when datasource configuration is read,
"result" => {
"ExampleDS" => {
"allocation-retry" => undefined,
"allocation-retry-wait-millis" => undefined,
"allow-multiple-users" => undefined,
"background-validation" => undefined,
"background-validation-millis" => undefined,
"blocking-timeout-wait-millis" => undefined,
"check-valid-connection-sql" => undefined,
"connection-properties" => undefined,
"connection-url" => "jdbc:h2:mem:test;DB_CLOSE_DELAY=-1",
"datasource-class" => undefined,
"driver-class" => undefined,
"driver-name" => "h2",
"enabled" => true,
"exception-sorter-class-name" => undefined,
"exception-sorter-properties" => undefined,
"flush-strategy" => undefined,
"idle-timeout-minutes" => undefined,
"jndi-name" => "java:jboss/datasources/ExampleDS",
"jta" => true,
"max-pool-size" => undefined,
"min-pool-size" => undefined,
"new-connection-sql" => undefined,
"password" => "sa",
"pool-prefill" => undefined,
"pool-use-strict-min" => undefined,
"prepared-statements-cache-size" => undefined,
"query-timeout" => undefined,
"reauth-plugin-class-name" => undefined,
"reauth-plugin-properties" => undefined,
"security-domain" => undefined,
"set-tx-query-timeout" => false,
"share-prepared-statements" => false,
"spy" => false,
"stale-connection-checker-class-name" => undefined,
"stale-connection-checker-properties" => undefined,
"track-statements" => "NOWARN",
"transaction-isolation" => undefined,
"url-delimiter" => undefined,
"url-selector-strategy-class-name" => undefined,
"use-ccm" => true,
"use-fast-fail" => false,
"use-java-context" => true,
"use-try-lock" => undefined,
"user-name" => "sa",
"valid-connection-checker-class-name" => undefined,
"valid-connection-checker-properties" => undefined,
"validate-on-match" => false,
"statistics" => {
"jdbc" => undefined,
"pool" => undefined
}
}
}
Jcliff uses the value of result, that is, the object containing the
ExampleDS. Some hacking is required here, because the configuration
tree starts with datasource, but the JBoss tree does not. We either
have to remove datasource from the configuration tree, or add
datasource to JBoss tree. The definition of datasource rules has
this property:
server.preprocess.prepend=/datasource
That is, once loaded the jboss configuration is converted into:
"datasource" => {
"ExampleDS" => {...}
}
All four options are possible:
server.preprocess.strip
server.preprocess.prepend
client.preprocess.strip
client.preprocess.prepend
These directives either insert, or remove levels from the client or server configuration tree.
After the preprocessing, the paths are built, and a difference is computed. The differences are:
add: Add a new configuration item to JBoss treemodify: Modify an existing item in JBoss tree. Only the leaf modified paths appear in the delta. That is, if a datasource attribute is modified, the datasource node itself is not in the delta, only the modified attribute is.remove: Remove an item from JBoss tree. This is done by assigning the value of an object to"deleted".undefine: Undefine item in JBoss tree.listAdd: Add a new element in a list in JBoss tree.listRemove: Remove an element from a JBoss list.
Hack: you can't add a nontrivial object, and set its attributes. The attributes of the nontrivial object don't exist until the object is created, which results in attribute add operations. Attributes cannot be added, only modified. So,
- You have to run the rules that add objects before the ones that modify the objects properties
- Once you add a nontrivial object, you have to refresh the JBoss configuration tree, so that you have a representation of the existing configuration with all the default values of the newly added object.
So, after adding a new object, you have to re-read the relevant
configuration from JBoss. This is done by the refresh directive. For
instance:
match.addConsoleHandler=add:/console-handler/*
addConsoleHandler.precedence=50
addConsoleHandler.rule=/subsystem=logging/console-handler:${name(.)}:add
addConsoleHandler.refresh=true
match.modifyConsoleHandler=modify:/console-handler/*/*
modifyConsoleHandler.precedence=55
modifyConsoleHandler.rule=/subsystem=logging/console-handler=${name(..)}:write-attribute(name=${name(.)},value=${value(.)})
addConsoleHandler will add a new console handler by only specifying
its name. Once added, all the attributes of the new handler will be
set to their default values. Jcliff will re-read the configuration,
retrieving all the attributes. Then, attribute modification rules are
run. Rules with lower precedence value run before rules with higher
precedence value.
Now the rules themselves:
-
The configurable subsytems are defined in the rules file:
configurable.1=system-properties configurable.2=logging configurable.3=jdbc-driver configurable.4=datasource configurable.5=xadatasource
Each configurable defines a rule file defining the rules to deal with that particular configurable. The explicit ordering defines the order in which the subsystems will be configured.
Each rule file contains at least the following:
name=xadatasource
getContents=/subsystem=datasources:read-children-resources(child-type=xa-data-source)
server.preprocess.prepend=/xadatasource
name: name of the configurablegetContents: The statement to run to retrieve the contents of this configurable from JBoss- preprocessing directives: optional
server/client.preprocess.prepend/strip, defining what to add/remove from the server/client configuration so that a meaningfull delta can be computed. One server and one client operation can be specified, but you can't specify two server or client operations.
Each rule is defined using a match property:
match.addDatasource=add:/datasource/*
This match property defines a rule name addDatasource. It matches
the node on the delta where a path of the form /datasource/<Name> is
added to the JBoss configuration. That is, a datasource exists in the
puppetized configuration, but not in JBoss. So, you define how to add
a new datasource:
addDatasource.rule.1=data-source add --name=${name(.)} --jndi-name=${value(jndi-name)} --driver-name=${value(driver-name)} --connection-url=${value(connection-url)}
addDatasource.rule.2=data-source enable --name=${name(.)}
addDatasource.refresh=true
The construct ${name(<path>)} evaluates to the name of the last
element in <path>. Path can be relative. A relative path is evaluated
with respect to the matched node. Above, if the matching node is
/datasource/BSProduct, ${name(.)} evaluates to BSProduct.
The construct ${value(<path>)} evaluates to the value of the node
denoted by <path>. In the above example, the expression
${value(jndi-name)} evaluates to the content of the path
/datasource/BSProduct/jndi-name, which should give the JNDI name of
the datasource in the puppetized configuration.
Multiple commands can be provided in a rule. In the above example,
addDatasource.rule.1 will create the datasource, and
addDatasource.rule.2 will enable it.
addDatasource.refresh will reload the JBoss configuration for
datasources. This is required after an add operation. The refresh
configuration will have all the datasource attributes initialized to
their default values, and any datasource attribute modification rule
will match after a refresh.
It is inefficient and error prone to keep writing rules of the form:
match.modifyProp=modify:/webservices/* modifyProp.rule.1=/subsystem=webservices:write-attribute(name=${name(.)},value=${value(.)})
Changing an attribute is done the same way for all configuration levels. So, instead of repeating this rule for every configuration level, it can be written as a prefix rule that applies to a set of nodes:
prefix.modifyProp=modify:/webservices
The above rule declaration will make modifyProp match any modifications of a path with prefix '/webservices'.
modifyProp.rule.1=/subsystem=webservices${cmdpath(${path(..)})}:write-attribute(name=${name(.)},value=${value(.)})
${cmdpath($path(..))} writes the path in the form /componen1=component2/component3=component4/...
${cmdpath(=$path(..))} writes the path in the form =component1/component2=component3/component4=component5...