v14

Breaking Changes:

• Change the Azure CPI job name from cpi to azure_cpi. You should update your bosh.yml.
  From

  jobs:
  - name: bosh
    templates:
    - {name: cpi, release: bosh-azure-cpi}
  
    properties:
      director:
        cpi_job: cpi
  
  cloud_provider:
    template: {name: cpi, release: bosh-azure-cpi}
  

  to

  jobs:
  - name: bosh
    templates:
    - {name: azure_cpi, release: bosh-azure-cpi}
  
    properties:
      director:
        cpi_job: azure_cpi
  
  cloud_provider:
    template: {name: azure_cpi, release: bosh-azure-cpi}
  

New Features:

• Allow to specify OS disk size by setting root_disk.size in cloud_properties of resource_pools.
  • Default OS disk size depends on the size of VHD in the stemcell. It is 3 GiB for now.
  • The size must be in the range [3 * 1024, 1023 * 1024]. Please always use N * 1024 as the size because Azure always uses GiB but not MiB.
• Allow to use OS disk to store ephemeral data by setting ephemeral_disk.use_root_disk to true in cloud_properties of resource_pools.
  • Compatible Stemcell Versions: v3262.7 or later
  • The OS disk will be atuomatically resized to 30 * 1024 MiB if you do not set root_disk.size.
  • No data disk will be created as the ephemeral disk.
• Allow to use env.bosh.group_name as the name of avaliability set.
  • You still can use availability_set in cloud_properties of resource_pools as the name of avaliability set.
  • If availability_set is not specified, Azure CPI will search env.bosh.group_name as the name of availability set. If neither availability_set nor env.bosh.group_name exists, VMs in this resource_pool won't be in any availability set.
• Support to pick an available storage account under the default resource group to create VMs by a pattern automatically.
  • You can use a pattern *keyword* as storage_account_name in cloud_properties of resource_pools.
  • The pattern must start with * and end with *.
  • The keyword must only contain numbers and lower-case letters because of the naming rule of storage account name.
  • The rule to select an available storage account is to check the number of disks under the container bosh does not exceed the limitation.
  • The default number of disks limitation is 30 but you can specify it in storage_account_max_disk_number.
  • storage_account_max_disk_number is the number of disks limitation in a storage account.
    1. Every storage account has a limitation to host disks. You may hit the performance issue if you create too many disks in one storage account.
    2. The maximum number of disks of a standard storage account is 40 because the maximum IOPS of a standard storage account is 20,000 and the maximum IOPS of a standard disk is 500.
    3. If you are using premium storage account, Azure maps the disk size (rounded up) to the nearest Premium Storage Disk option (P10, P20 and P30). For example, a disk of size 100 GiB is classified as a P10 option.
       • The maximum number of disks of a premium storage account is 280 if you are using P10 (128 GiB) as your disk type.
       • The maximum number of disks of a premium storage account is 70 if you are using P20 (512 GiB) as your disk type.
       • The maximum number of disks of a premium storage account is 35 if you are using P30 (1024 GiB) as your disk type.
    4. storage_account_max_disk_number should be less than the maximum number. Suggest you to use (MAX - 10) as the value because CPI always creates VMs in parallel.
    5. Please see more information about azure-subscription-service-limits.

Improvements:

• Log x-ms-request-id, x-ms-correlation-request-id and x-ms-routing-request-id in xRP responses and get_token response for trouble shooting.
• Use 2016-06-01 as RP API version for AzureChinaCloud.
• Update Azure API version for resource provider to 2016-06-01 and others to 2015-06-15.
• Using azure-storage instead of azure-sdk-for-ruby to access Azure storage.
• Check whether a property is nil but not only exist if it is required.
• Pick default ephemeral disk size based on instance type if ephemeral_disk.size in cloud_properties of resource_pools is not set.
  • If the Azure temporary disk size for the instance type is less than 30*1024 MiB, the default size is 30*1024 MiB because the space may not be enough.
  • If the Azure temporary disk size for the instance type is larger than 1023*1024 MiB, the default size is 1023*1024 MiB because max data disk size is 1023*1024 MiB on Azure.
  • Otherwise, the Azure temporary disk size will be used as the default size. See more information about Azure temporary disk size.
• CPI does not delete availability sets because it is difficult to know whether an availability set will be used or not.
• Allow to specify storage_account_type and storage_account_location in cloud_properties of resource_pools to use a non-exist storage account. CPI will create the storage account automatically.
  • storage_account_type is required if the storage account does not exist. It can be either Standard_LRS, Standard_ZRS, Standard_GRS, Standard_RAGRS or Premium_LRS. You can click HERE to learn more about the type of Azure storage account.
  • storage_account_location is needed if the storage account does not exist. If it is not set, the location of the default resource group will be used. You can use azure cli command azure location list to list all locations. For more information, see List all of the available geo-locations.
• Append error details in exceptions so that users can see the error details in the console when executing bosh-init/bosh command.

Fixes:

• Handle the race when multiple processes are creating a same storage account in parallel. Please see more details about the issue #176.
• CPI packaging should not fail on OS X with Homebrew-installed OpenSSL. Please see more details about the issue #184.
• RetryableError often happens when deleting a deployment. Please see more details about the issue #179.
• CPI should make sure to log ruby stacktrace to stderr when exception occurs. Please see more details about the issue #177.

Dev:

• Update CI pipeline after upgrading concourse to v1.6.0.
• Update docker image for CI pipeline.
  • Upgrade ruby version to v2.2.4
  • Use Ubuntu 15.04
  • Install nodejs 4.x and npm
  • Install Azure CLI 0.10.1

v13

New Features:

• Support multiple resource groups. Please reference this doc PR

Improvements:

• Improve the performance of uploading stemcells
• Filter credential in logs
• Catch BAD_REQUEST in get_token

Dev:

• Add debug_mode to log all raw HTTP requests/responses. Please reference this doc PR

Known Issues:

• bosh-init v0.0.92 has issue with connection timeouts; use bosh-init v0.0.94.

v12

• Change the default ephemeral size from 15GB to 30GB
• Fix invalid disk name error when failing to create VM

Dev:

• Avoid extra bash process by passing CPI input directly to exec'ed process

v11

Improvements:

• Reduce storage account queries
• Fix retry issue in asynchronous operations.

Fixes:

• Use LUN and host device id as the disk identifier
  • Compatible Stemcell Versions: v3181 or later
  • Stemcell v3232.5 or later is recommended with CPI v11. You may hit the issue #135 if you still use an older stemcell than v3232.5
• Use a data disk as the persistent ephemeral disk
  • The VM size should not be Standard_A0 if you want to use persistent data disks
  • The CPU cores of VM should be more than 1 if you plan to resize the persistent data disks in future
  • The premium storage is recommended for better performance
  • You can see more information about the 'Max data disks' and 'Max disk IOPS & bandwidth' of Azure Linux VMs here
• Fix an issue in calculating the sleep interval when copying blobs

Dev:

• Updated Ruby to 2.2.4
• Removed several unnecessary gem dependencies

v10

New Features:

• Add AzureChinaCloud support

Improvements:

• Auto retry when an internal error occurs in asynchronous operations
• Reduce retry interval to improve performance

v9

New Features:

• Allow specifying security_groups for a VM via resource pool's cloud_properties
  • Strongly recommded to be specified when using public IPs
• Allow specifying default_security_group as a global configuration

v8

Improvements:

• Assign public IPs directly to VMs
  • Previously public IPs were assigned by adding a LB to the VM
• Retry around DNS resolution errors
  • Typically shown as SocketError: Hostname not known

Dev:

• Set open_timeout explicitly for HTTP requests