0.20.0

[machine-controller-manager]

Most notable changes

• [USER] MCM now needs permissions to GET/LIST PV and PVCs on target cluster while draining machines (#275, @ggaurav10)
• [OPERATOR] Changes the drain logic to honour the individual pods' terminationGracePeriodSeconds instead of a fixed(though configurable) timeout. For the pods without volumes, the timeout would be a bit longer than the maximum of terminationGracePeriodSeconds of the pods without volume. For the pods with volumes, the timeout is its own terminationGracePeriodSeconds and a fixed (but configurable) volume detach timeout for each pod. (#288, @amshuman-kr)
  • The command-line flags machine-max-evict-retries and machine-pv-detach-timeout can be used to customize the default behaviour of the number of retries for eviction and wait period for the detaching of volumes after pods are evicted/deleted.
  • The default value for the existing command-line flag machine-drain-timeout has been increased to 12h which should be enough for most work-loads.

Improvements

• [OPERATOR] Increased OpenStack server status wait for a timeout during server creation from 5 to 10 mins. (#286, @kayrus)
• [OPERATOR] Improved the server status handling - It no longer waits for the timeout if the server status is different from BUILD during server creation. (#286, @kayrus)
• [OPERATOR] Force deletion of machine succeeds even on drain failures (#283, @prashanth26)
• [OPERATOR] dep is replaced by go mod. (#280, @ialidzhikov)
• [OPERATOR] When draining a node, pods with PVCs are evicted serially. Next pod eviction waits for PV of the previously evicted pod to detach from the node. Operators can expect faster machine drains for their machines. (#275, @ggaurav10)