- Introduction
- How does it work?
- James setup and configuration
- Information Point options
- Information Point scripts
- Contributions
James is a Java agent that may be attached to any JVM process and provide operational insight on what's going on under the hood of your service.
It may extract the state of execution of any running method, including method arguments, object state, stack trace, call duration, transform that state into meaningful information using plugin toolkits or custom, hot-pluggable scripts, and finally load the transformation result into destination data warehouse or a file.
James uses ByteBuddy code generation and manipulation library for injecting advices into underlying application bytecode.
You can wire up James to your service by providing JARs with the agent and its plugins and adding two parameters to the virtual machine. Once the service is running James provides a way to manage a set of Information Points through a Controller.
Information Point defines a point in your code where James hooks up to a running service. Think of it as an advice or an aspect wrapped around a method. When the method with an Information Point is called, James measures method execution time and invokes a Groovy script provided as a part of an Information Point definition. In that script you can transform the state of method parameters or method's object to anything you like and publish the result of mapping as an Event.
For example, if a method contains transaction ID in its arguments, you could call some kind of metadata service and retrieve metadata for that transaction ID, and publish it in your Event.
Beside a reference to method, at which Information Point has to be created and a script, that has to be executed when the method is executed, a Information Point may have a sampleRate parameter, which translates to a probability at which the advise will invoke the script. For example when sampleRate is set to 50, only half of method invocations will cause the script to execute. This is designed to enable throttling of outbound message throughput for methods that are executed too frequently.
Controllers enabled you to interact with James by listing, adding and removing Information Points and by querying diagnostic information.
There are two kinds of Controllers at the moment โ HTTP Web Service Controller and Consul Controller. The former enables managing Information Points on single agent-enabled service via REST interface, the latter uses Consul node as a source of Information Point configuration for a cluster of agent-enabled services. Controllers enable rapid deployment of a set of Information Points.
Event is a structured piece of information published by a Groovy script defined for Information Point that can be loaded into either text file in any format (JSON formatter is provided) or through Kinesis to ElasticSearch. That way all information gathered from Information Points is easily available in Kibana for analysis.
A Publisher is a component responsible for sending your Events to the outside world. It may be
Console Publisher that will simply write JSON-formated Events to stdout
, it may be File Publisher that will
put Events into a file, or it may be Kinesis Publisher that will send your Events to
AWS Kinesis for buffering and further processing.
James is designed with performance in mind, the execution time overhead it introduces to the method invocations is minimal and all I/O and data processing by Groovy script engine is done in the background. Our microbenchmarks let us believe that it can handle about 150k Information Point hits per second on a laptop, thus effectively for most use cases its performance is bound by a Publisher message throughput.
At TomTom Content Production Platform we use James with our services to gather runtime information for analysis and troubleshooting purposes. The configuration of the cluster of James-enabled services is managed using Consul Controller. Kinesis Publisher is used to send Events to AWS Kinesis stream from which they are picked by AWS Lambda, put into Elasticsearch cluster and visualized using Kibana. With this setup we gather more than 30 GB of valuable runtime information from our system each day and analyze it in near-realtime with Kibana.
To start working with James you need to add two command line arguments to Java Virtual Machine running your application:
-javaagent:/path/to/james-agent.jar -Djames.configurationPath=/path/to/james-configuration.yaml
The configuration of James is an YAML-formatted document made of several sections that contain settings of the agent's subsystems. Most of settings have default values, thus minimal configuration may look like this:
plugins:
includeDirectories:
- /opt/james/plugins/common # Look for plugins in this directory at start
- /opt/james/plugins/service-specific # ... and in this directory
controllers:
- id: james.controller.webservice # Use Webservice Controller for management with default settings
publishers:
- id: james.publisher.console # Use Console Publisher to write JSON-formatted events to stdout with
# default settings
Sections of controllers
, publishers
and toolkits
are described below, let's take a look at more complete
example of core settings configuration:
quiet: true # Don't print out ASCII banner at the start (default: false)
logLevel: trace # Level of James logs - trace|debug|info|warn|error|fatal
# (default: warn)
plugins:
includeDirectories:
- /opt/james/plugins # Look for plugins in this directory at start
includeFiles:
- /opt/james-extra/additional-plugin1.jar # ... and in this JAR file
- /opt/james-extra/additional-plugin2.jar # ... and in this JAR file
controllers:
- id: james.controller.webservice
properties:
# See below for james.controller.webservice configuration properties.
- id: james.controller.consul
properties:
# See below for james.controller.consul configuration properties.
publishers:
- id: james.publisher.console
properties:
# See below for james.publisher.console configuration properties.
- id: james.publisher.file
properties:
# See below for james.publisher.file configuration properties.
- id: james.publisher.kinesis
asyncWorkers: 4 # Use 4 worker threads for consuming Events by this Publisher
# (default: 1).
maxAsyncJobQueueCapacity: 20000 # Maximum size of job queue for this Publisher is 20000 Events
# (default: 10000). If the queue is full when a script tries
# to publish an Event, the Event will be dropped.
properties:
# See below for james.publisher.kinesis configuration properties
toolkits:
yourOrganization.toolkit.yourToolkit:
# See below for toolkit configuration properties.
scriptEngine:
asyncWorkers: 4 # Run 4 instances of the Script Engine concurrently (default: # of cpus).
maxAsyncJobQueueCapacity: 10000 # Maximum size of job queue for the Script Engine (default: 10000).
informationPointStore:
persistenceEnabled: true # Should Information Points be persisted between restarts
# (default: false).
storeFilePath: /var/james/informationpoints.json # If Information Point persistence is enabled, use this file
# for storage.
Two plugins are built-in into James and do not require provisioning as separate plugin JARs:
james.publisher.console
and james.publisher.file
.
The role of Controller plugins is to provide an interface for interactions with James.
There may be any number of Controller plugins running at the same time in a single James instance.
Webservice Controller enables you to manage a single instance of James using HTTP REST interface. You can list, add, remove an Information Point or check the state of internal queues used for asynchronous communication between James' subsystems.
By default the controller listens for HTTP requests on TCP port 7007. You can configure the controller using the following configuration properties:
plugins:
includeFiles:
- /path/to/plugins/james-controller-webservice.jar
controllers:
- id: james.controller.webservice
properties:
port: 7007 # TCP port on which the controller is listening for connections (default: 7007)
minThreads: 1 # Minimal number of HTTP server worker threads (default: 1)
maxThreads: 8 # Maximal number of HTTP server worker threads (default: 8)
Following HTTP commands are supported:
Request Path | HTTP Verb | Request Media Type | Response Media Type | Description |
---|---|---|---|---|
/v1/information-point | GET | N/A | application/json | Get definitions of all Information Points. |
/v1/information-point/{className}/{methodName} | GET | N/A | application/json | Get a definition of specific Information Point. |
/v1/information-point | POST | application/json | N/A | Create an Information Point. See below for an example of payload. |
/v1/information-point/{className}/{methodName} | DELETE | N/A | N/A | Delete an Information Point. |
/v1/queue | GET | N/A | application/json | Get an information about all async queues. |
/v1/queue/script-engine | GET | N/A | application/json | Get an information about Script Engine async queue. |
/v1/queue/event-publisher | GET | N/A | application/json | Get an information about Event Publisher async queue. |
The body of "POST /v1/information-point" command that adds and Information Point at
com.github.me.myapp.MyClass#myMethodInMyClass
with 30% sampling rate and provided script may look like this:
{
"className": "com.github.me.myapp.MyClass",
"methodName": "myMethodInMyClass",
"script": [
"// First line of Information Point script",
"// Second line of Information Point script"
],
"sampleRate": 30
}
If you need a simple CLI manager for James that uses Webservice Controller take a look at james-m project. The tool is self-describing with its built-in help system. Some usage examples:
- Add a new Information Point at James running with Webservice Controller on
100.101.102.103:17007
with sample rate 20% andscript1.groovy
script:
james-m --host 100.101.102.103 --port 17007 ip add \
com.github.me.myapp.MyClass#myMethodInMyClass \
scripts/script1.groovy \
--sample-rate 20
- Remove an Information Point at James running with Webservice Controller on
localhost:7007
(default):
james-m ip remove com.github.me.myapp.MyClass#myMethodInMyClass
- Show all Information Points at James running with Webservice Controller on
localhost:7007
(default):
james-m ip show
James-m is available as Windows executable. Alternatively you can
build it yourself using go get github.com/pdebicki/james-m
if you have Go build
environment installed.
With Consul Controller you can add, remove or modify Information Points using Consul
key-value store. Each Information Point is stored as a key-value pair, where the key is a method reference
of Information Point (for example com.github.me.myapp.MyClass!myMethodInMyClass
) and the value is
a JSON-formatted definition of the Information Point.
When the KV store entry is added in Consul, all James instances with Consul Controller automatically add an Information Point defined in that KV entry. When the entry is modified, all instances replace the Information Point and when the entry is removed, all instances remove the Information Point.
Note that unlike other use cases, the type name and the method name in method reference string are separated
with !
character instead of common #
character. That's because Consul does not allow #
character in its keys.
You can configure Consul Controller using the following configuration properties:
plugins:
includeFiles:
- /path/to/plugins/james-controller-consul.jar
controllers:
- id: james.controller.consul
properties:
host: localhost # Host of the Consul node (default: localhost)
port: 8500 # Port of the Consul node (default: 8500)
folderPath: james/test/information-points # A path to the Consul KV folder containing Information Points
An example of Consul KV store value containing Information Point definition could be:
{
"version": 1,
"script": [
"// First line of Information Point script",
"// Second line of Information Point script"
],
"sampleRate": 30
}
For the above example, the key value could be com.github.me.myapp.MyClass!myMethodInMyClass
. The version
property is the version of the JSON schema, intended to maintain backward compatibility of existing
Information Point definitions.
The role of a Publisher is to marshall and send out an Event generated by an Information Point script. All Publishers, same way as the Script Engine are asynchronous; the advise on Information Point's method only captures the context of method execution and schedules execution of the script in the Script Engine thread pool. Same way Events created by Information Point scripts are scheduled for future processing by one or more Publishers in the background. That way the impact of James on the running application is minimal.
A single instance of James may have any number of Publishers.
Console Publisher enables writing of Events generated by James to stdout
. It is built-in into James agent
JAR, thus plugin provisioning is not required to use it.
The configuration is very simple:
publishers:
- id: james.publisher.console
properties:
prettifyJSON: true # Whether the output JSON should be pretty-printed (default: false).
File Publisher enables writing of Events generated by James to a file. It is built-in into James agent JAR, thus plugin provisioning is not required to use it.
The configuration is also very simple:
publishers:
- id: james.publisher.file
properties:
prettifyJSON: true # Whether the output JSON should be pretty-printed (default: false).
path: /var/james/file-publisher.out # Output file path (default: $HOME/james-publisher-file-output.json).
Kinesis Publisher enables sending Events generated by James to AWS Kinesis stream. The stream may serve as a buffer, for example between James instances and Elasticsearch cluster.
Published Events are marshalled as JSON documents, with additional fields like @timestamp
, @version
,
host
, jvmName
, type
, and environment
that enable identification of Event source and sending Events to
Elasticsearch cluster without additinal data structure transformations. Fields type
and environment
are
configurable and @timestamp
, @version
, host
, jvmName
are auto-generated.
The structure of the configuration is as in the following example:
plugins:
includeFiles:
- /path/to/plugins/james-controller-kinesis.jar
publishers:
- id: james.publisher.kinesis
asyncWorkers: 4 # Use 4 worker threads for consuming Events by this Publisher
# (default: 1).
maxAsyncJobQueueCapacity: 20000 # Maximum size of job queue for this Publisher is 20000 Events
# (default: 10000). If the queue is full when a script tries
# to publish an Event, the Event will be dropped.
properties:
stream: stream # Name of the target Kinesis stream (default: james-kinesis).
partitionKey: partitionkey # Stream partition key (default: random UUID).
elasticSearch:
eventType: james # Type of the Event for Elasticsearch (default: james).
environment: dev-env # Name of the environment for Elasticsearch (only sent if present).
producer: # Optional. See below.
aggregationEnabled: true
aggregationMaxCount: 4294967295
aggregationMaxSize: 51200
cloudwatchEndpoint: cloudwatch-endpoint
cloudwatchPort: 443
collectionMaxCount: 500
collectionMaxSize: 5242880
connectTimeout: 6000
enableCoreDumps: false
failIfThrottled: false
kinesisEndpoint: kinesis-endpoint
kinesisPort: 443
logLevel: info
maxConnections: 10
minConnections: 1
nativeExecutable: native-exec
rateLimit: 150
recordMaxBufferedTime: 100
recordTtl: 30000
region: region
requestTimeout: 6000
tempDirectory: /tmp
verifyCertificate: true
All the properties in producer:
section of the configuration enable overriding of default settings in
amazon-kinesis-producer library. Check default_config.properties
for description of each property and default value.
Toolkits enable provisioning of external libraries to James for use in Information Point scripts. Toolkit may come with it's own set of third party dependencies, if bundled together into a single JAR.
Let's consider an example. You might want to extract a transactionID from method arguments of your application, call a metadata service to fetch some kind of metadata for that transaction, perform some additional processing of the data and publish an Event containing the results. To do that you will either need to implement metadata service client in Groovy or use existing client. Using Toolkit you can wrap an existing client in James plugin and call it from Information Point script code.
To create a Toolkit you need to implement com.tomtom.james.common.api.toolkit.Toolkit
from james-agent-common
module.
public class MetadataServiceToolkit implements Toolkit {
private RealMetadataClient realMetadataClient;
@Override
public String getId() {
return "yourOrganization.toolkit.metadataService";
}
@Override
public void initialize(ToolkitConfiguration toolkitConfiguration) {
StructuredConfiguration configuration = toolkitConfiguration.getProperties()
.orElseGet(StructuredConfiguration.Empty::new);
String serviceURL = configuration.get("serviceURL")
.map(StructuredConfiguration::asString)
.orElseThrow(() -> new ConfigurationStructureException("Missing serviceURL property"));
realMetadataClient = new RealMetadataClient(serviceURL);
}
public Map<String, String> getMetadataForTransaction(UUID transactionID) {
return realMetadataClient.getMetadataForTransaction(transactionID);
}
}
In the example above the initialize()
method expects the serviceURL
property to be present in property
set of your tookit. In that case James configuration might look like this:
plugins:
includeFiles:
- /path/to/metadata-toolkit.jar
toolkits:
yourOrganization.toolkit.metadataService:
serviceURL: http://metadata-service.somewhere.org/
To call a method of a Toolkit you have to retrieve the Toolkit object from the Script Engine framework using Toolkit's ID, and then simply invoke the method using Groovy's dynamic nature.
import com.tomtom.james.script.SuccessHandlerContext
def onSuccess(SuccessHandlerContext context) {
def txnID = context.runtimeParameters[0].value
def toolkit = getToolkitById("yourOrganization.toolkit.metadataService")
def txnMetadata = toolkit.getMetadataForTransaction(txnID)
def txnCreator = txnMetadata["creator"]
def eventMap = [
transaction_id: txnID,
transaction_creator: txnCreator
]
publishEvent(new Event(eventMap))
}
Information point can contains following information
Property | Type | Description |
---|---|---|
className |
java.lang.String |
Name of class information point should be applied |
methodName |
java.lang.String |
Method of class information point should be applied |
baseScript |
java.lang.String |
Optional base script for this information point |
script |
java.lang.String |
Script for this information point |
sampleRate |
java.lang.Integer |
Sample rate (it is deprecated in favor of successSampleRate and errorSampleRate ) |
successSampleRate |
java.lang.Double |
Success sample rate |
errorSampleRate |
java.lang.Double |
Error sample rate |
successExecutionThreshold |
java.lang.Long |
Success threshold in milliseconds (onSuccess will be executed only for values with higher value of executionTime ) |
Information Point script is a piece of Groovy code that is invoked by the Script Engine after a method with defined Information Point is executed. It has to contain at least two functions:
import com.tomtom.james.script.SuccessHandlerContext
import com.tomtom.james.script.ErrorHandlerContext
def onSuccess(SuccessHandlerContext context) {
}
def onError(ErrorHandlerContext context) {
}
Function onSuccess
is an event handler that is called by the Script Engine when the method returns a result.
Function onError
is called when the method throws an exception.
Both SuccessHandlerContext
and ErrorHandlerContext
are derived from the common type -
InformationPointHandlerContext
and share a set of runtime information captured by the Information Point during
execution of the method.
Context property | Type | Available in |
---|---|---|
informationPointClassName |
java.lang.String |
InformationPointHandlerContext |
informationPointMethodName |
java.lang.String |
InformationPointHandlerContext |
origin |
java.lang.reflect.Method |
InformationPointHandlerContext |
runtimeInstance |
java.lang.Object |
InformationPointHandlerContext |
runtimeParameters |
List<RuntimeInformationPointParameter> |
InformationPointHandlerContext |
currentThread |
java.lang.Thread |
InformationPointHandlerContext |
eventTime |
java.time.Instant |
InformationPointHandlerContext |
executionTime |
java.time.Duration |
InformationPointHandlerContext |
callStack |
java.lang.String[] |
InformationPointHandlerContext |
returnValue |
java.lang.Object |
SuccessHandlerContext |
errorCause |
java.lang.Throwable |
ErrorHandlerContext |
Properties informationPointClassName
and informationPointMethodName
contain class and method name of the
Information Point that triggered the script.
Property origin
contains a reference to the executed method itself. Note the difference - if an
Information Point was set on an abstract method or an overriden method of a base class then
informationPointClassName
will contain the name of superclass, and origin.declaringClass.name
will contain
the name of actual subclass of which the method was executed.
Property runtimeInstance
contains a reference to the object of which the method was invoked. You can use it
to access fields or call methods on the object.
Property runtimeParameters
has a list of all parameters of the method, elements of which contain parameter value,
type and name.
Property currentThread
contains a reference to a thead that executed the method.
Property eventTime
contains a time when event happened with precision of nanoseconds.
Property executionTime
contains a time of method's execution, with precision of nanoseconds.
Property callStack
is an array of all method references in call stack of the thread that executed the method.
Property returnValue
contains the return value of the method after successful execution.
Property errorCause
contains the reference to Throwable that was thrown by the method.
Beside runtime properties passed to event handlers via a context there are global functions defined for Information Point scripts.
Function Event createEvent(InformationPointHandlerContext context)
or Event createEvent(Map<String, Object> content, InformationPointHandlerContext context)
creates an Event for given context, optionally with content. It has advantage over simple
new Event()
by setting up event timestamp based on eventTime
.
Function void publishEvent(Event event)
sends an Event to Publisher's queue for processing in the future.
Function Toolkit getToolkitById(String toolkitId)
returns a reference to a Toolkit instance by given ID.
import com.tomtom.james.common.api.publisher.Event
import com.tomtom.james.script.ErrorHandlerContext
import com.tomtom.james.script.SuccessHandlerContext
def onSuccess(SuccessHandlerContext context) {
def eventMap = [
informationPointClassName : context.informationPointClassName,
informationPointMethodName: context.informationPointMethodName,
originDeclaringClassName : context.origin.declaringClass.name,
originName : context.origin.name,
instanceFieldValue : context.instance.field,
executionTimeNanos : context.executionTime.toNanos(),
callStack : context.callStack,
currentThreadName : context.currentThread.name,
returnValue : context.returnValue,
]
context.parameters.each {
eventMap["arg(${it.name})"] = it.value
}
publishEvent(new Event(eventMap))
}
def onError(ErrorHandlerContext context) {
def eventMap = [
informationPointClassName : context.informationPointClassName,
informationPointMethodName: context.informationPointMethodName,
originDeclaringClassName : context.origin.declaringClass.name,
originName : context.origin.name,
instanceFieldValue : context.instance.field,
executionTimeNanos : context.executionTime.toNanos(),
callStack : context.callStack,
currentThreadName : context.currentThread.name,
errorCauseMessage : context.errorCause.message,
]
context.parameters.each {
eventMap["arg(${it.name})"] = it.value
}
publishEvent(new Event(eventMap))
}
In complex deployment it could be beneficial to share common base implementation for you script, it can be done using concept of base script present in groovy.
Base script need to extend InformationPointHandler - if not specified InformationPointHandler is used as default
abstract class CustomInformationPointHandler extends InformationPointHandler {
Event createEvent(Map<String, Object> content, InformationPointHandlerContext context) {
content << [
result : context instanceof ErrorHandlerContext ? "error" : "success",
className : context.informationPointClassName,
methodName : context.informationPointMethodName,
]
context.parameters.each {
content["arg(${it.name})"] = it.value
}
return super.createEvent(content, context);
}
}
Using above base you can reuse part from base
import com.tomtom.james.common.api.publisher.Event
import com.tomtom.james.script.SuccessHandlerContext
def onSuccess(SuccessHandlerContext context) {
publishEvent(createEvent(context))
}
James is a young and small project created inside one of TomTom's development teams to give us more insight on what's going on inside our services. It was meant to deliver information where big guys like AppDynamics failed to do so and give us lots of flexibility when we need to analyze and troubleshoot our running code.
Although the codebase is very young, James is running successfully on hundreds of EC2 machines and delivers value to us for several months.
We hope you will find James useful and welcome all contributions from the Open Source community.