platoon
platoon copied to clipboard
mila-iqia
[Launcher] Check if the controller launched properly before starting the workers
We should check if the controller actually launched before starting the workers.
Imagine the case you have a controller already running on the same port you are trying to use, an old one or just somebody else controller. Then you would launch worker what would work on with the other controller which is bad.
Why not tell the controller how many worker in total it should have and after he accept that number of worker, any later worker will get an error?
On Mon, Feb 8, 2016 at 5:20 PM, Mathieu Germain [email protected] wrote:
We should check if the controller actually launched before starting the workers.
Imagine the case you have a controller already running on the same port you are trying to use, an old one or just somebody else controller. Then you would launch worker what would work on with the other controller which is bad.
— Reply to this email directly or view it on GitHub https://github.com/mila-udem/platoon/issues/57.
Let say you are running something that takes a long time and a GPU get freed up. You probably want to be able to add a new worker whenever you want.
On Tue, Feb 9, 2016 at 10:32 AM, Frédéric Bastien [email protected] wrote:
Why not tell the controller how many worker in total it should have and after he accept that number of worker, any later worker will get an error?
On Mon, Feb 8, 2016 at 5:20 PM, Mathieu Germain [email protected] wrote:
We should check if the controller actually launched before starting the workers.
Imagine the case you have a controller already running on the same port you are trying to use, an old one or just somebody else controller. Then you would launch worker what would work on with the other controller which is bad.
— Reply to this email directly or view it on GitHub https://github.com/mila-udem/platoon/issues/57.
— Reply to this email directly or view it on GitHub https://github.com/mila-udem/platoon/issues/57#issuecomment-181915236.
I don't mean it is mandatory to use it, but I think that frequently, you won't be able to add more GPU as you will already use all the usable GPU on the computer. But yes, we should not make what I descrive mandatory.
On Tue, Feb 9, 2016 at 10:42 AM, Mathieu Germain [email protected] wrote:
Let say you are running something that takes a long time and a GPU get freed up. You probably want to be able to add a new worker whenever you want.
On Tue, Feb 9, 2016 at 10:32 AM, Frédéric Bastien < [email protected]> wrote:
Why not tell the controller how many worker in total it should have and after he accept that number of worker, any later worker will get an error?
On Mon, Feb 8, 2016 at 5:20 PM, Mathieu Germain < [email protected]> wrote:
We should check if the controller actually launched before starting the workers.
Imagine the case you have a controller already running on the same port you are trying to use, an old one or just somebody else controller. Then you would launch worker what would work on with the other controller which is bad.
— Reply to this email directly or view it on GitHub https://github.com/mila-udem/platoon/issues/57.
— Reply to this email directly or view it on GitHub https://github.com/mila-udem/platoon/issues/57#issuecomment-181915236.
— Reply to this email directly or view it on GitHub https://github.com/mila-udem/platoon/issues/57#issuecomment-181919634.