design: multi-replica scheduling for singleton sources (aka hot standby) #31205
Conversation
aljoscha
force-pushed
the
design-multi-replica-singleton-sources
branch
from
06a025f to
6a126ef
Compare
aljoscha
changed the title
|
|
||
| ## Non-Goals | ||
|
|
||
| - Add a failure detection mechanism for replicas and update source scheduling |
There was a problem hiding this comment.
👍🏽👍🏽👍🏽
I'm very in favor of explicitly excluding "fault tolerance" to keep the scope as limited as possible.
There was a problem hiding this comment.
This seems reasonable to me. Stepping away from the assumption that all replicas receive the same commands will add a bunch of complexity in the controller, but it doesn't look like we have a choice. At least I don't think the alternative will turn out any simpler.
| We want to support zero-downtime ALTER on clusters. The plan for this is to | ||
| turn on a new replica with the changed parameters and turn off the old replica | ||
| when the new one is "sufficiently ready". This in turn requires that we are |
There was a problem hiding this comment.
Nit: This might be me being dumb but I at first thought this was describing some form of ALTER SOURCE. It's describing what we usually call "graceful cluster reconfiguration", so maybe it would be helpful to mention that term here?
There was a problem hiding this comment.
adding a clarification!
| expected to shut down. | ||
|
|
||
| A: This is caught by the mechanisms we already have today for making sure there | ||
| is only one active ingestion dataflow. We need this for correctness in the fact |
There was a problem hiding this comment.
| is only one active ingestion dataflow. We need this for correctness in the fact | |
| is only one active ingestion dataflow. We need this for correctness in the face |
aljoscha
force-pushed
the
design-multi-replica-singleton-sources
branch
from
20f48f8 to
7a666be
Compare
Rendered: https://github.com/aljoscha/materialize/blob/design-multi-replica-singleton-sources/doc/developer/design/20250127_multi_replica_scheduling_singleton_sources.md
Motivation
Tips for reviewer
Checklist
$T ⇔ Proto$Tmapping (possibly in a backwards-incompatible way), then it is tagged with aT-protolabel.