Change vm.destroy gracefulness for performance when discarding vms

[pevogam]

During environment setup we discard vms that are not going to participate in a test but doing so gracefully results in a large timeout (60 seconds by default) for each one and slows down the test and overall test run significantly. As the discarded vm will either be fully reconstructed or restored from a clean state when it is needed again by a test, graceful shutdown is not needed and ridding ourselves of it will improve performance for free.

[pevogam]

@luckyh Hopefully this is an easy diff and it seems reasonable to you.

[luckyh]

Hi @pevogam, thanks for bringing this up! I just have a concern to the following point.

As the discarded vm will either be fully reconstructed or restored from a clean state when it is needed again by a test ...

However, the guest os images might continue to be used for the rest tests. As unclean shutdowns may trigger the guest encountering into the rescue mode (especially for Windows guests), without any protection being taken (e.g. using snapshots) those tests may fail to run into the end which would be not expected by the users.

So I'd suggest that, if we exactly want such a tweak to be applied, we'd better either

• make the behavior configurable by a param and tell the users about the risk

or

• do ungraceful shutdowns only if there is any protection been activated to the os images

Let me know your thoughts.

[pevogam]

Hi @pevogam, thanks for bringing this up! I just have a concern to the following point.

As the discarded vm will either be fully reconstructed or restored from a clean state when it is needed again by a test ...

However, the guest os images might continue to be used for the rest tests. As unclean shutdowns may trigger the guest encountering into the rescue mode (especially for Windows guests), without any protection being taken (e.g. using snapshots) those tests may fail to run into the end which would be not expected by the users.

So I'd suggest that, if we exactly want such a tweak to be applied, we'd better either

* make the behavior configurable by a param and tell the users about the risk

Indeed I guess you are right about the images, I originally thought the vms are always reconstructed on top of black images since anything else might not isolate tests well enough but I was wrong here. So I pushed changes making this configurable, let me know what you think.

[pevogam]

@YongxueHong @PaulYuuu Would you like to review this?

[luckyh]

Thanks all.