Build in queue assigned to cloud agent that is in the process of stopping

 Hi,

I'm encountering the following issue (in TeamCity 2017.1):

We have an agent pool that has a single cloud profile with a single cloud image configured to clone a VM when starting and delete the clone when stopping. The profile is set to terminate instances after the first build.

We have a configuration A with a snapshot dependency on configuration B. When we add A to the build queue, B is queued with it (as expected) and a new instance (let's call it cloud agent X) of the cloud profile starts up and runs the build for configuration B. Immediately afterward, cloud agent X begins the process of stopping, in preparation for deletion. This is supported by the TeamCity server cloud logs.

During this time, though, cloud agent X is still shown to be connected and enabled when we look at the agent pools, and soon afterward configuration A attempts to run on cloud agent X (though it's clearly stopping-- I can't access its agent logs, for example). After a minute of waiting, configuration A's build is cancelled with the message "Could not connect to build agent. Details: Agent didn't come for the command in 60 seconds." 

This isn't what we expected-- we wanted a new agent instance to start up and run configuration A. Is the scenario described above expected behaviour given the settings we defined? If so, how can we get the behaviour we want instead?

Thanks for any help you can give us.