Windows Azure BizTalk Services was recently released by Microsoft, and you can find a fair bit about this cloud service online. I wrote up a simple walkthrough, Sam Vanhoutte did a nice comparison of features between BizTalk Server and BizTalk Services, the Neudesic folks have an extensive series of blog posts about it, and the product documentation isn’t half bad.
However, I wanted to learn more about how the service itself works, so I reached out to Karthik Bharathy who is a senior PM on the BizTalk team and part of the team that shipped BizTalk Services. I threw a handful of technical questions at him, and I got back some fantastic answers. Hopefully you learn something new; I sure did!
Richard: Explain what happens after I deploy an app. Do you store the package in my storage account and add it to a new VM that’s in a BizTalk Unit?
Karthik: Let’s start with some background information – the app that you are referring today is the VS project with a combination of the bridge configuration and artifacts like maps, schemas and DLLs. When you deploy the project, each of the artifacts and the bridge configuration are uploaded into one by one. The same notion also applies through the BizTalk Portal when you are deploying an agreement and uploading artifacts to the resources.
The bridge configuration represents the flow of the message in the pipeline in XML format. Every time you build BizTalk Services project in Visual Studio, an <entityName>.Pipeline.atom is generated in the project bins folder. This atom file is the XML representation of the pipeline configuration. For example, under the <pipelines> section you can see the bridges configured along with the route information. You can also get a similar listing by issuing a GET operation on the bridge endpoint with the right ACS credentials.
Now let’s say the bridge is configured with Runtime URL <deploymentURL>/myBridge1. After you click deploy, the pipeline configuration gets published in the repository of the <deploymentURL> for handling /myBridge1. For every message sent to the <deploymentURL>, the role looks at the complete path including /myBridge1 and load the pipeline configuration from the repository. Once the pipeline configuration is loaded, the message is processed per the configuration. If the configuration does not exist, then an error is returned to the caller.
Richard: What about scaling an integration application? How does that work?
Karthik: Messages are processed by integration roles in the deployment. When the customer initiates a scale operation, we update the service configuration to add/remove instances based on the ask. The BizTalk Services deployment updates its state during scaling operation and new messages to the deployment is handled by one of the instances of the role. This is similar to how Web or Worker roles are scaled up/down in Azure today.
Richard: Is there any durability in the bridge? What if a downstream endpoint is offline?
Karthik: The concept of a bridge is close to a messaging channel if we may borrow the phrase from Enterprise Integration Patterns. It helps in bridging the impedance between two messaging systems. As such the bridge is a stateless system and does not have persistence built into it. Therefore bridges need to report any processing errors back to the sender. In the case where the downstream endpoint is offline, the bridge propagates the error back to the sender – the semantics are slightly different a) based on the bridge and b) based on the source from where the message has been picked up.
For EAI, bridges with HTTP the error code is sent back to the sender while with the same bridge using an FTP head, the system tries to pickup and process the message again from the source at regular intervals (and errors out eventually). In both cases you can see relevant track records in the portal.
For B2B, our customers rarely intend to send an HTTP error back to their partners. When the message cannot be sent to a downstream (success) endpoint, the message is routed to the suspend endpoint. You might argue that the suspend endpoint could be down as well – while this is generally a bad idea to put in a flaky target in success or suspend endpoints, we don’t rule out this possibility. In the worst case we deliver the error code back to the sender.
Richard: Bridges resemble BizTalk Server ESB Toolkit itineraries. Did you ever consider re-using that model?
Karthik: ESB is an architectural pattern and you should look at the concept of bridge as being part of the ESB model for BizTalk Services on Azure. The sources and destinations are similar to the on-ramp, off-ramp model and the core processing is part of the bridge. Of course, additional capabilities like exception management, governance, alerts will need to added to bring it closer to the ESB Toolkit.
Richard: How exactly does the high availability option work?
Karthik: Let’s revisit the scaling flow we talked about earlier. If we had a scale >=2, you essentially have a system that can process messages even when one of the machines can down in your configuration. If one of the machines are down, the load balancer in our system routes the message to the running instances. For example, this is taken care of during “refresh” when customers can restart their deployment after updating a user DLL. This ensures message processing is not impacted.
Richard: It looks like backup and restore is for the BizTalk Services configuration, not tracking data. What’s the recommended way to save/store an audit trail for messages?
Karthik: The purpose of backup and restore is for the deployment configuration including schemas, maps, bridges, agreements. The tracking data comes from the Azure SQL database provided by the customer. The customer can add the standard backup/restore tools directly on that storage. To save/store an audit of messages, you have couple of options at the moment – with B2B you can turn archiving on either AS2 or X12 processing and with EAI you can plug-in an IMessageInspector extension that can read the IMessage data and save it to an external store.
Richard: What part of the platform are you most excited about?
Karthik: Various aspects of the platform are exciting – we started off building capabilities with ‘AppFabric Connect’ to enable server customers leverage existing investments with the cloud. Today, we have built a richer set of functionality with BizTalk Adapter services to connect popular LOBs with Bridges. In the case of B2B, BizTalk Server traditionally exposed functionality for IT Operators to manage trading partner relationships using the Admin Console. Today, we have a rich TPM functionality in the BizTalk Portal and also have the OM API public for the developer community. In EAI we allow extending message processing using custom code If I should call out one I like, it has to be custom code enablement. The dedicated deployment model managed by Microsoft makes this possible. It is always a challenge to enable user DLL to execute without providing some sort of sandboxing. Then there are also requirements around performance guarantees. BizTalk Services dedicated deployment takes care of all these – if the code behaves in an unexpected way, only that deployment is affected. As the resources are isolated, there are also better guarantees about the performance. In a configuration driven experience this makes integration a whole lot simpler.
Thanks Karthik for an informative chat!