Syer Monitoring Integration And Batch

Chicago, October 19 - 22, 2010Chicago, October 19 - 22, 2010
Deploying and Monitoring Spring
Integration and Spring Batch
Applications
Dave Syer - SpringSource

SpringOne 2GX 2010. All rights reserved. Do not distribute without permission.
Agenda
• Spring Batch and Spring Integration
• Spring and JMX
• Spring Integration
– Metrics
– Notifications and alerts
– Control
• Metametrics and Metaoperations: e.g. deployment and provisioning
• Spring Batch
– Control
– Metrics
– Notifications and alerts
• Using SpringSource Hyperic
• More Future Directions

Spring Batch
Application
Batch Core
Batch Infrastructure
Re-usable low level
stuff: flat files, XML
files, database keys
Quality of service,
auditability,
management
information, Job,
Step, status
Business logic

Job Configuration
<job>
<step id="load" next="analyse"/>
<step id="analyse">
<next on="COMPLETED" to="summarize"/>
<next on="SKIPS" to="logSkips"/>
<end on="FAILED"/>
</step>
<step id="logSkips" next="summarize"/>
<step id="summarize" />
</job>
<job>
<step id="load" next="analyse"/>
<step id="analyse">
<next on="COMPLETED" to="summarize"/>
<next on="SKIPS" to="logSkips"/>
<end on="FAILED"/>
</step>
<step id="logSkips" next="summarize"/>
<step id="summarize" />
</job>

Quality of Service
• Stuff happens:
– Item fails
– Job fails
• Failures can be
– Transient – try again and see if you succeed
– Skippable – ignore it and maybe come back to it later
– Fatal – need manual intervention
• Mark a job execution as FAILED
• When it restarts, pick up where you left off
• Monitor status of running and historic executions
• All framework concerns: not business logic

Batch Domain Diagram
Job
Step
locates
Infrastructure dependencies
executes
JobLauncher
starts and stops
JobParameters
JobExecution
StepExecution
*
JobRepository
stored in
used to identify and manage jobs
executes
Batch
Operator
uses
Application
Developer
configures

Item-Oriented Processing
• Input-output can be grouped together = Item-Oriented Processing
ItemReader
read()
ExitStatus
ItemWriter
write(item)
Step
item
execute()
repeat,
retry,
etc.

Demo

Spring Integration
Application
Integration Adapters
Integration Core
Enterprise
Integration Patterns,
message, poller,
router, splitter,
aggregator, etc.
File, JMS, JDBC,
JMX, TCP, etc.
Business logic

Message
• Payload can be any object
• Header values are stored in a Map
Headers
Payload

Message Channel
• Decouples Producers from Consumers
• Provides extension point for interceptors
• May be Point-to-Point
• Or Publish/Subscribe
Producer
Consumer

Low-level Components
Message
Source
Message
Handler
Endpoint
• Passive
• Poller
• Lifecycle
Message
Channel

The Big Picture

Spring Batch and Spring Integration
• Message triggers Job
• Step wraps gateway (e.g. background processing)
• Remote chunking
• Partitioning
• Automatic retry
• Flat file and XML processing
• QoS: once only processing, restart
• Audit: track status of execution
• http://www.springsource.org/spring-batch-admin/spring-
batch-integration

JMX
Application
Client
MBeanServer
MBean MBean MBean
Connector Connector
Client

JMX
• MBeanServer provides services to MBeans and clients
(through connectors)
– Registration (ObjectName -> MBean)
– Attributes
– Operations
– Notifications
– Monitors (thresholds, value changes etc.)

Spring and JMX
@ManagedResource
public class TransferServiceImpl implements TransferService {
@ManagedOperation
public void transfer(String source, String target, double amount) {
…
}
@ManagedAttribute
public double getTotalBalance() {
…
}
}
@ManagedResource
@ManagedOperation
public void transfer(String source, String target, double amount) {
…
}
@ManagedAttribute
…
}
}

Spring and JMX
<context:mbean-server id="mbeanServer" />
<context:mbean-export server="mbeanServer" />
<bean id="transferService" class="com.foo.TransferServiceImpl"/>
<context:mbean-server id="mbeanServer" />
<context:mbean-export server="mbeanServer" />
<bean id="transferService" class="com.foo.TransferServiceImpl"/>
MBean
com.foo:type=TransferServiceImpl,name=transferService
ObjectName
Automatically unregistered when
ApplicationContext closes

Spring 3.0 and JMX: Hyperic Metrics
@ManagedResource
…
@ManagedMetric(metricType = MetricType.GAUGE)
…
}
}
<context:mbean-export server="mbeanServer" default-
domain="spring.application"/>
@ManagedResource
…
@ManagedMetric(metricType = MetricType.GAUGE)
…
}
}
<context:mbean-export server="mbeanServer" default-
domain="spring.application"/>

Spring Integration Metric Use Cases
• How many messages have been processed?
• How long did it take?
• Are things getting any worse?
• Were there any errors?
• Can I shutdown the system now without losing any
messages?

Spring Integration Metrics
• MessageChannel
– send count, receive count (if polling)
– execution times
– error rates and ratios
• MessageHandler
– handle count
– execution times
– error rates
• MessageSource
– message count
• Global
– active handler count
– queued message count

MessageHandler: Naming Problems
• MessageHandler is end user API (usually POJO) and
location of interesting metrics
• Endpoint is the thing with a name (bean id)
• Many endpoints are anonymous
• But channels are not
widgets

Spring Integration JMX
<int-jmx:mbean-export server="mbeanServer“
default-domain="spring.application"/>
<int:channel id="widgets"/>
<int:service-activator input-channel="widgets"
expression="payload.length()" />
<int-jmx:mbean-export server="mbeanServer“
default-domain="spring.application"/>
<int:channel id="widgets"/>
<int:service-activator input-channel="widgets"
expression="payload.length()" />
MBean
spring.application:type=MessageChannel,name=widgets
MBean
spring.application:type=MessageHandler,name=widgets,bean=anonymous

Spring Integration Metric Types
ExponentialMovingAverageHandleDurationMessageHandler
IntegerHandleCountMessageHandler
ExponentialMovingAverageRatioErrorRatioMessageChannel
ExponentialMovingAverageRateSendErrorRateMessageChannel
ExponentialMovingAverageSendDurationMessageChannel
IntegerQueueSizeMessageChannel
IntegerSendCountMessageChannel
AlgorithmMetricComponent

Spring Integration Control
• Alerts on hotspots: slow handling, high error rates
• Lifecycle
– start, stop, running
• Throttle messages on downstream error
• Graceful shutdown

Metametrics and Metaoperations
• Deployment and provisioning: get me one instance of this
application.
• What have I got deployed?
• Is it working?
• Scale and elastic scale: increase the number of workers.
• Dynamic load: re-balance thread pools.
• Highlight 4 approaches
– Command line
– Local service
– Build and deployment tool
– Private or public cloud

Command Line
• Write a Java main() method
• Use one off the shelf, e.g. CommandLineJobRunner
• Maybe use Maven, Ant, Gradle etc.
• Pros:
• Lowest common denominator
• High granularity:
– Tweak JVM parameters (heap size etc.)
– Tweak global transaction settings
• Cons:
• Setting up classpath and dependencies
• Inventory: tracing and managing versions

Local Service: Servlet Container
• Write a servlet / web service for the business features
• Use one off the shelf, e.g. Spring Batch Admin
• Pros:
• Readily available container
• Standard packaging and dependency management
• Tools for monitoring and management, e.g. Hyperic
• Cons:
• Some overhead creating container instance
• Low granularity: module boundaries have to be drawn

Remote Build and Deployment Tool
• Write a remote build and deployment tool
• Use one off the shelf, e.g. Capistrano, Hudson
• Pros:
• Granularity of command line
• Can be used to manage remote server as well
• Handles versioning
• Cons:
• New technology for some users
• Still quite low level: requires script maintainence

Private or Public Cloud (PaaS)
• Make deployment and provisioning a service
• Pros:
• Commoditisation / standardisation
• Public services easily accessible
• Private cloud products emerging (e.g. Napa)
• Very efficient once running: “blueprint” applies to all
deployments
• Elastic and dynamic scale are built in features
• Cons:
• Extra layer in process and implementation
• Some restrictions on application, especially in public
clouds

Spring Batch Control
• Basic use cases
– Stop a job execution
– Abort a job execution
– Restart a job execution*
– Start new or next execution*
• Exposed through Spring Batch Admin and
CommandLineJobRunner
• SBA can be deployed as a standalone service (just needs
a JobRepository)
• *WithSBA only for Jobs that are configured inside SBA
application

Spring Batch Metrics
• Latest JobExecution status, duration
• StepExecution status and history, items processed,
skipped, failure analysis
• Implementation: Spring Batch Admin
– web application
– services (REST and Java)
– JMX endpoints, e.g. consume with Hyperic

Notifications and Alerts
• Spring Batch SLA: step overrun
– When a step execution takes too long, downstream
processing can fail
– Use MBean Monitor features to send notifications
– JMX client can enable and consume the notifications as
desired

Step Execution SLA
<bean id="serviceLevel"
class="org...admin.jmx.StepExecutionServiceLevelMonitor">
<constructor-arg ref="batchMBeanExporter"/>
<property name="timeout" value=“60000" />
<property name="jobName" value="job1" />
<property name="stepName" value="j1.s1" />
<property name="enabled" value="true" />
</bean>
<bean id="serviceLevel"
class="org...admin.jmx.StepExecutionServiceLevelMonitor">
<constructor-arg ref="batchMBeanExporter"/>
<property name="timeout" value=“60000" />
<property name="jobName" value="job1" />
<property name="stepName" value="j1.s1" />
<property name="enabled" value="true" />
</bean>
MBean
spring.application:type=StepExecutionMonitor/WARN,name=job1/j1.s1
spring.application:type=StepExecutionMonitor/ERROR,name=job1/j1.s1

JMX Notifications and Spring Integration
• Spring Integration provides adapters for JMX notifications:
– inbound: notification -> Message
– outbound: Message -> notification
<jmx:notification-listening-channel-adapter
channel="input"
object-name="com.foo:type=TestBean,name=testBean"/>
<jmx:notification-publishing-channel-adapter
channel="input"
object-name="com.foo:type=TestBean,name=testBean“
default-notification-type="com.foo.test"/>
<jmx:notification-listening-channel-adapter
channel="input"
object-name="com.foo:type=TestBean,name=testBean"/>
<jmx:notification-publishing-channel-adapter
channel="input"
object-name="com.foo:type=TestBean,name=testBean“
default-notification-type="com.foo.test"/>

Spring Integration ControlBus
• MBean operations on Message* components triggered by
Message
<jmx:control-bus mbean-exporter="mbeanExporter"
channel="control"/>
<int-jmx:mbean-export server="mbeanServer"/>
<jmx:control-bus mbean-exporter="mbeanExporter"
channel="control"/>
<int-jmx:mbean-export server="mbeanServer"/>
control
ControlBus.TARGET_BEAN_NAME=“myEndpoint"
JmxHeaders.OPERATION_NAME="stop"

JMX Notifications and Hyperic
• Hyperic can also consume JMX notifications
– treated like log files (regex pattern on message)
– generate alerts (email, SMS etc.)
– trigger control actions (e.g. switch off upstream messages)
• Some bugfixes not yet released support auto-detection
with no XML (should be in 4.5.x)

Demo?

More Future Directions
• Tracing and Business Transaction Monitoring
• Replay failed transaction
• Automated scheduling and restart
• Performance summary, test harness

Q&A
Source Code for demos:
http://git.springsource.org/s2gx-2010/monitoring-spring-batch-integration

Syer Monitoring Integration And Batch

More Related Content

What's hot

Similar to Syer Monitoring Integration And Batch

Recently uploaded

Syer Monitoring Integration And Batch