Add support for co-locating sequencers and partition processor leaders #2115
Conversation
| fn sequencer_node(&self) -> Option<PlainNodeId> { | ||
| match self { | ||
| LogletConfiguration::Replicated(configuration) => { | ||
| Some(configuration.sequencer.as_plain()) |
There was a problem hiding this comment.
I'm not entirely sure why is this a plain node id.
There was a problem hiding this comment.
I think this is wrong. Will change it to return GenerationalNodeId.
|
Thanks for the review @pcholakov and @AhmedSoliman. I've addressed your comments and pushed another commit. |
| let sequencer = preferred_sequencer | ||
| .and_then(|node_id| { | ||
| // map to a known alive node | ||
| observed_cluster_state.alive_nodes.get(&node_id.id()) |
There was a problem hiding this comment.
Don't we lose strictness of preference by potentially swapping the generational node id that was requested with whatever generation that we have in the list of alive nodes?
Is that intention here to fallback back in this order?
- Generational node if it's alive
- Other generation of the same node if alive
- Any other alive node
There was a problem hiding this comment.
The problem is what if a GenerationalNodeId is specified that is not known to be alive? In this case, the logs controller will try to reconfigure the log on the next tick. That's why the order, you've described, is what is being followed.
Right now, the scheduler won't give a GenerationalNodeId hint since the chosen partition processor leader is not bound to a generation.
tillrohrmann
force-pushed
the
issues/2081
branch
from
616f229 to
0539022
Compare
Let Scheduler respect replicated loglet placements Let LogsController respect the preferred sequencer node This fixes restatedev#2081.
tillrohrmann
force-pushed
the
issues/2081
branch
from
0539022 to
141bd1c
Compare
Let Scheduler respect replicated loglet placements
Let LogsController respect the preferred sequencer node
This fixes #2081.