e76c1ae972
* Initial work on rapl-daemon. Initial server set up. API to read max power per zone and API to write new power cap have both been written. * Removing python script since this has been ported to go code now. * Adding test for happy path retrieving max power. * Change some data types around to avoid too much conversion. * Add happy path test for cap zone * Removing uncessary print statement. * Change cap node to use a temporary setup. * Renaming arguments to be more descriptive. * Changing todo message. * Changing test structure to only set up mock subsystem once and allowing functions to test on it later. * Adding some more coverage for unhappy paths and fixing some values to reflect they are no longer floats. * Keeping the old script around as it should be removed in a different PR. * Delegating percentage check to capNode function. * Fixing typo. * Fixing typos. * Changing shortWindow to longWindow as constraint_0 actually points to the long window. * Renaming variable in test. * capping funciton now returns which zones were sucessfully capped and which zones could not be capped. This information is now returned to the caller of the HTTP api.
782 B
782 B
RAPL Daemon
This runs a server that is capable of changing the percentage at which a node is being throttled to using RAPL. This daemon should be installed on all worker nodes.
Sample payload for testing:
curl --header "Content-Type: application/json" \
--request POST \
--data '{"percentage":75}' \
http://localhost:9090/powercap
Payload
{
"percentage":75
}
Response
The daemon will respond with a json payload containing zones that were successfully capped as well as the zones that were not capped.
{
"cappedZones": null,
"failedZones": [
"intel-rapl:0",
"intel-rapl:1"
],
"error": "some zones were not able to be powercapped"
}
Field error will not exist if failed zones is empty.