You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe.
nomad-pot-driver calls "pot get-rss" once per second per container. When running many containers, this causes a lot of forking.
Describe the feature you'd like to have
Not entirely sure how to approach this best - doing some actual profiling with very cheap containers would be a first step.
One idea would be some sort of daemon that writes rss information of all processes controlled by a pot instance to a file that can be used as a source to make "pot get-rss" cheaper? In general this feels like it should be solved by reducing the amount of script work involved (maybe potnet could be extended to include more tools that optimize certain things and/or make them more secure, or improve inside of nomad-pot-driver). Keeping as much as possible in shell is a good thing though in terms of iterating over the code and being able to fix things in production easily.
Describe potential alternatives or workaround you've considered (if any)
As a first step, reduce the amount of forking required when calling "pot get-rss".
The text was updated successfully, but these errors were encountered:
Is your feature request related to a problem? Please describe.
nomad-pot-driver calls "pot get-rss" once per second per container. When running many containers, this causes a lot of forking.
Example on a host running 6 containers:
# while sleep 1; do ps -ax | grep -i "ps -ax"; done 73060 0 R+ 0:00.00 ps -ax 73061 0 S+ 0:00.00 grep -i ps -ax 73303 0 R+ 0:00.00 ps -ax 73304 0 S+ 0:00.00 grep -i ps -ax 73546 0 R+ 0:00.00 ps -ax 73547 0 S+ 0:00.00 grep -i ps -ax 73789 0 R+ 0:00.00 ps -ax 73790 0 S+ 0:00.00 grep -i ps -ax 74032 0 R+ 0:00.00 ps -ax 74033 0 S+ 0:00.00 grep -i ps -ax
Describe the feature you'd like to have
Not entirely sure how to approach this best - doing some actual profiling with very cheap containers would be a first step.
One idea would be some sort of daemon that writes rss information of all processes controlled by a pot instance to a file that can be used as a source to make "pot get-rss" cheaper? In general this feels like it should be solved by reducing the amount of script work involved (maybe potnet could be extended to include more tools that optimize certain things and/or make them more secure, or improve inside of nomad-pot-driver). Keeping as much as possible in shell is a good thing though in terms of iterating over the code and being able to fix things in production easily.
Describe potential alternatives or workaround you've considered (if any)
As a first step, reduce the amount of forking required when calling "pot get-rss".
The text was updated successfully, but these errors were encountered: