[JCLOUDS-1230] azurecompute-arm: VM deletion takes very long time (many extra api calls?)

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.0.0
• Component/s: jclouds-compute
• Labels: azurecompute-arm
• Assignee: Andrea Turli
• Reporter: Aled Sage
• Created: 25/Jan/17 17:12
• Updated: 22/Apr/17 20:01

Description

Deleting a VM in azurecompute-arm took 5mins 55seconds. Looking at the logging for that thread, I suspect there are a lot of optimisations that could be done.

Here's a rough breakdown of the destroy steps, and how long they take:

• Get details of the VM and other pieces (e.g. network interface, etc
  3 seconds
• List all the images
  11 seconds
• List all the storageaccounts
  3mins 39secs
• List providers/locations
  1 second
• Delete the VM, and poll for completion .
  1min 51secs
• Delete nic
  3 seconds
• Delete VM disk storage
  1 second
• delete storage account
  4 seconds

Can we avoid listing all the images and all the storage?

Below are some log snippets that illustrate this (all taken from a single thread):

  
2017-01-25 16:01:44,266 DEBUG 106 j.compute [ger-M9fVKLUG-247] >> destroying node(southeastasia/qa-entity-id-ca)

... 
2017-01-25 16:01:44,267 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking GetVirtualMachine

... 
2017-01-25 16:01:47,859 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking sku:list

... 
2017-01-25 16:01:58,598 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking resourcegroup:list

... 
2017-01-25 16:01:58,628 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking storageaccount:list

... 
2017-01-25 16:05:37,543 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking providers:get

...
2017-01-25 16:05:37,592 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking location:list
...

...
2017-01-25 16:05:38,916 DEBUG 106 j.compute [ger-M9fVKLUG-247] >> destroying southeastasia/qa-entity-id-ca ...
2017-01-25 16:05:38,917 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking DeleteVirtualMachine

...
2017-01-25 16:07:29,170 DEBUG 106 j.compute [ger-M9fVKLUG-247] >> destroying nic jc-nic-qa-entity-id-ca...
2017-01-25 16:07:29,171 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking networkinterfacecard:delete

...
2017-01-25 16:07:32,091 DEBUG 106 j.compute [ger-M9fVKLUG-247] >> deleting public ip nic public-address-qa-entity-id-ca...
2017-01-25 16:07:32,091 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking publicipaddress:delete

...
2017-01-25 16:07:34,107 DEBUG 106 j.compute [ger-M9fVKLUG-247] >> deleting virtual machine disk storage...
2017-01-25 16:07:34,121 DEBUG 101 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking DeleteContainer

...
2017-01-25 16:07:35,141 DEBUG 106 j.compute [ger-M9fVKLUG-247] >> deleting storage account qaentityid524b...
2017-01-25 16:07:35,142 DEBUG 106 o.j.r.i.InvokeHttpMethod [ger-M9fVKLUG-247] >> invoking storageaccount:delete

...
2017-01-25 16:07:39,855 DEBUG 106 j.compute [ger-M9fVKLUG-247] << destroyed node(southeastasia/qa-entity-id-ca) success(true)

Also, combined with https://issues.apache.org/jira/browse/JCLOUDS-1229 this could be very dangerous: we might well be rate-limited while trying to list all the images and thus the VM deletion / cleanup would be aborted.