The whole approach to monitoring drains has ordering issues and lacks
state to output useful error messages.
AFAICT to get the tests passing reliably I needed to change the behavior
of monitoring.
Parts of these tests are skipped in CI, and they should be rewritten as
e2e tests.
This change updates the console log message when performing a node
drain and particulary when a node has marked all allocs for
migration. Previously it logged 'drain complete' which was a little
confusing to operators as the node is not drained at this point.
Closes#4183
Before the drain CLI would block until the node was marked as completing
drain operations. While technically correct, it could lead operators (or
more likely: scripts) to shutdown drained nodes before all of its
allocations had *actually* terminated.
This change makes the CLI block until all allocations have terminated
(unless ignoring system jobs).
Also delay "node complete" after the node has been marked complete to
capture a few more alloc events. There are other ways to implement this
that could trade off correctness for responsiveness as technically a
node is considered drained when all of its allocs have been marked to
stop and not when they've actually stopped (which may not happen for a
long time).
* Truncate all UUID identifiers to eight characters by default
* Refactor the node identifier to an auto-generated UUID
* Created and updated tests and mocks
Tim Gross