rest interface goes awol while new instances are transfered

[echofs]

 

I have a vanilla seller installation with no running instances (#1047). It is an 8 core x 3.2GHz box with 8G free with a dedicated 1GB uplink to the internet. It is running the free ECP cloud edition, and it is really fast.

 

I notice that when a buyer tries to provision an instance, and that instance as never been provisioned before, the instance is pushed to the server. When that push happens, the server does not respond to polls on the rest interface while the PUT is happening. They simply time out. What this means is that the first buyer for each instance type generates failures in the logs for the few minutes it takes to push the instance down (see below).

 

The only process chewing up CPU while this happens is the single python process (py) responsible for receiving the file. Once the download is done and the qemu vm starts up, things return to normal and it happily accepts connections again.

 

Is this typical for a server of this capacity? Does the i/o or the cpu for the download really overwhelm the process responsible for the rest interface? Is there something I can do to debug this and/or increase the number of threads/processes that are available to respond during the download?

 

eg:

 

June 8, 2011, 9:22 p.m.	GET	/utilization	OK
June 8, 2011, 9:21 p.m.	GET	/vm/list	Could not contact the server
June 8, 2011, 9:20 p.m.	GET	/htemplate/list	Could not contact the server
June 8, 2011, 9:19 p.m.	PUT	/vm	OK

[Adam Boduch]

Hi,

That is very interesting. We've never had someone point out this
problem. The scenario you're describing shouldn't halt your system due
to a file download. I'll see if I can get a better answer for you with
regard to diagnosing this problem.

Regards,

Adam