Commonly asked Service Fabric Mesh questions

Azure Service Fabric Mesh is a fully managed service that enables developers to deploy microservices applications without managing virtual machines, storage, or networking. This article has answers to commonly asked questions.

How do I report an issue or ask a question?

Ask questions, get answers from Microsoft engineers, and report issues in the service-fabric-mesh-preview GitHub repo.

Quota and Cost

What is the cost of participating in the preview?

There are no charges for deploying applications or containers to the Mesh preview currently. However, we encourage you to delete the resources you deploy and not leave them running unless you're actively testing them.

Is there a quota limit of the number of cores and RAM?

Yes. The quotas for each subscription are:

  • Number of applications: 5
  • Cores per application: 12
  • Total RAM per application: 48 GB
  • Network and Ingress end points: 5
  • Azure Volumes that you can attach: 10
  • Number of Service replicas: 3
  • The largest container you can deploy is limited to 4 cores and 16GB RAM.
  • You can allocate partial cores to your containers in increments of 0.5 cores, up to a maximum of 6 cores.

How long can I leave my application deployed?

We have currently limited the lifetime of an application to two days. This is in order to maximize the use of the free cores allocated to the preview. As a result, you are only allowed to run a given deployment continuously for 48 hours, after which time it will be shut down.

If you see this happen, you can validate that the system shut it down by running the az mesh app show command in the Azure CLI. Check to see if it returns "status": "Failed", "statusDetails": "Stopped resource due to max lifetime policies for an application during preview. Delete the resource to continue."

For example:

~$ az mesh app show --resource-group myResourceGroup --name helloWorldApp
{
  "debugParams": null,
  "description": "Service Fabric Mesh HelloWorld Application!",
  "diagnostics": null,
  "healthState": "Ok",
  "id": "/subscriptions/1134234-b756-4979-84re-09d671c0c345/resourcegroups/myResourceGroup/providers/Microsoft.ServiceFabricMesh/applications/helloWorldApp",
  "location": "eastus",
  "name": "helloWorldApp",
  "provisioningState": "Succeeded",
  "resourceGroup": "myResourceGroup",
  "serviceNames": [
    "helloWorldService"
  ],
  "services": null,
  "status": "Failed",
  "statusDetails": "Stopped resource due to max lifetime policies for an application during preview. Delete the resource to continue.",
  "tags": {},
  "type": "Microsoft.ServiceFabricMesh/applications",
  "unhealthyEvaluation": null
}

To delete the resource group, use the az group delete <nameOfResourceGroup> command.

Supported container OS images

If you are developing on a Windows Fall Creators Update (version 1709) machine, you can only use Windows version 1709 docker images.

If you are developing on a Windows 10 April 2018 update (version 1803) machine, you can use either Windows version 1709 or Windows version 1803 docker images.

The following container OS images can be used to deploy services:

  • Windows - windowsservercore and nanoserver
    • Windows Server version 1709
    • Windows Server version 1803
  • Linux
    • No known limitations

Developer experience issues

DNS resolution from a container doesn't work

Outgoing DNS queries from a container to the Service Fabric DNS service may fail under certain circumstances. This is being investigated. To mitigate:

  • Use Windows Fall Creators update (version 1709) or higher as your base container image.
  • If the service name alone doesn't work, try the fully qualified name: ServiceName.ApplicationName.
  • In the Docker file for your service, add EXPOSE <port> where port is the port you are exposing your service on. For example:
EXPOSE 80

DNS does not work the same as it does for Service Fabric development clusters and in Mesh

You may need to reference services differently in your local development cluster than in Azure Mesh.

In your local development cluster use {serviceName}.{applicationName}. In Azure Service Fabric Mesh, use {servicename}.

Azure Mesh does not currently support DNS resolution across applications.

For other known DNS issues with running a Service Fabric development cluster on Windows 10, see: Debug Windows containers and known DNS issues.

Networking

The ServiceFabric network NAT may disappear while using running your app on your local machine. To diagnose whether this has happened, run the following from a command prompt:

docker network ls and note whether servicefabric_nat is listed. If not, then run the following command: docker network create -d=nat --subnet 10.128.0.0/24 --gateway 10.128.0.1 servicefabric_nat

This will address the issue even if app is already deployed locally and in an unhealthy state.

Issues running multiple apps

You might encounter CPU availability and limits being fixed across all applications. To mitigate:

  • Create a five node cluster.
  • Reduce CPU usage in services across the app that is deployed. For example, in your service's service.yaml file, change cpu: 1.0 to cpu: 0.5

Multiple applications can't be deployed to a one-node cluster. To mitigate:

  • Use a five node cluster when deploying multiple apps to a local cluster.
  • Remove apps that you are not currently testing.

Feature gaps and other known issues

After deploying my application, the network resource associated with it does not have an IP address

There is a known issue in which the IP address doesn't become available immediately. Check the status of the network resource in a few minutes to see the associated IP address.

My application fails to access the right network/volume resource

In your application model, use the full resource ID for networks and volumes to be able to access the associated resource. Here is an example from the quickstart sample:

"networkRefs": [
    {
    "name":  "[resourceId('Microsoft.ServiceFabric/networks', 'SbzVotingNetwork')]" 
    }
]

When I scale out, all of my containers are affected, including running ones

This is a bug and a fix is being implemented.

Next steps

To learn more about Service Fabric Mesh, read the overview.