Monitoring Camel applications on the Cloud

Apache Camel is the swiss army knife of application integration. If your application is processing data from different systems, sooner or later you will end up using some of the 130 (and growing number of) connectors available. Camel also has excellent cloud support. You can use either jCoulds component that provides portable service abstraction layer for Amazon, GoGrid, VMWare, Azure, and Rackspace or use Amazon specific components for SQS, SNS, S3, SES, SDB, DDB (some of which contributed by me).

In this tutorial I'm going to show you how to monitor Camel applications with JMX and Amazon CloudWatch service. The need for such a monitoring arose after I created livephotostream - the photo streaming application I created while playing with Camel. This application retrieves real time photos from Twitter streaming api, and not so real time data from Instagram and Tumblr, then display them using Bootstrap and sometimes Websocket. And I wanted to see how many photos it is receiving, how many of them are failing, how long it takes to process a photo and so on. So I end up creating a camel-cloudwatch component and a simple application demonstrating how to use it to display JMX metrics.

Amazon CloudWatch
Amazon CloudWatch provides monitoring for AWS cloud resources by tracking various metrics and activating alarms based on these metrics. It enables you to monitor AWS resources in real-time such as EC2 instances, EBS volumes, Load Balancers, RDS instances… But for my message-oriented application in addition to CPU utilisation, memory usage, SQS messages, etc... I want to see also metrics specific to my Camel application, such as ExchangesCompleted, ExchangesFailed, MaxProcessingTime, MeanProcessingTime, or number of exchanges processed by a given route or even by a specific processor in a route. Luckily Camel tracks and exposes these metrics over JMX. You can read about how to access these metric with jConsole or FuseIDE on MichaƂ Warecki excellent blog post. But connection to a java application running on AWS instance over JMX, using jConsonle or other application is not a joy. Instead I'd rather see these metrics on the nice graphs CloudWatch provides and even get notified if any of the metrics reaches a threshold value.

This is when the new cloudwatch producer becomes useful. It lets you send custom metrics to amazon CloudWatch service. All you have to do is give a namespace to your metrics, and send each metric by specifying its name, value and optionally the unit and timestamp. If you don't specify the unit and timestamp it will use count as unit and the current time as the metric time.

Let's say we want to tract the number of failed exchanges in a Camel route. This metric can be retrieved using the jmx component, which lets you subscribe to an mbean's Notifications. The jmx consumer below will cause a new MonitorBean to be created and deployed to the local mbean server that monitors the "ExchangesFailed" attribute on the "processingRoute" bean. And if that attribute value changes, this consumer will send a new message containing the change - i.e. the number of failures so far. After that all you have to do is set that value as CamelAwsCwMetricValue message header and give it a name with the CamelAwsCwMetricName header. The rest will be handled by the cloudwatch producer.

The route above doesn't require any additional code. It will send any changes to the ExchangesFailed attribute to CloudWatch service, and you can see the error count not changing in the CloudWatch graph:

The downside of using jmx consumer to monitor a bean is that you will need to create separate consumer and route for each bean and attribute monitored. Another approach is to have a processor that can access the beans from the JMX server, read all the attributes of interest and send all of them at once again using the cloudwatch component.

In this scenario you will need also a timer to trigger the polling. The cloudwatch producer can generate a metric based on message headers as in the first example, or simply send the metric objects from the message body if present as in the second example. The processor for retrieving data from camel jmx is also straightforward and can be seen in the demo app on my guthub account.

The result from the last route is couple of metrics, all of which sent to CloudWatch every minute. As you can see from the next image the number of exchanges completed successfully is increasing all the time, but with a different rate:

From the other hand, the mean processing time (in millis) is different for each message but in close interval and not growing drastically.

If the new metrics doesn't exist in CloudWatch, it will be created for you, but if you want to set up alarms and trigger automatic actions, you still have to log into amazon console and to it manually. And finally don't forget to check your "cloud bill" after adding all these metrics, they are not for free ;)


Nice blog Bilgin.

I took the liberty to add a link to our from our link collection at:

Thanks Claus. I wasn't sure whether it is good enough to be added to Camel articles :)

Post a Comment

About Me