As part of the initial cluster unification work (https://github.com/MaterializeInc/cloud/issues/4929), we unified storaged and computed into a single clusterd process, but left the storage and compute timely clusters separate. That is, a clusterd process hosts two timely clusters: one for storage and one for compute.
To support multipurpose clusters, we need to unify these two timely clusters. There are many design questions to shake out:
Do we unify the command streams?
If so, do we unify the command streams before we unify the timely clusters?
What do we do with introspection relations like mz_compute_frontiers that will now contain compute and storage frontiers? See #10089 for a past pondering on the name of this relation.
Part of the multipurpose clusters epic (#17413).
As part of the initial cluster unification work (https://github.com/MaterializeInc/cloud/issues/4929), we unified
storaged
andcomputed
into a singleclusterd
process, but left the storage and compute timely clusters separate. That is, aclusterd
process hosts two timely clusters: one for storage and one for compute.To support multipurpose clusters, we need to unify these two timely clusters. There are many design questions to shake out:
mz_compute_frontiers
that will now contain compute and storage frontiers? See #10089 for a past pondering on the name of this relation.cc @guswynn @aljoscha @antiguru @teskje