[jubrad]

Problem

mz_catalog.mz_cluster_replica_sizes leaves out disabled sizes, but replicas keep running on them. Anything that joins replicas to sizes loses those replicas: the Console's replica list says "This cluster has no replicas", and mz_cluster_replica_utilization and mz_cluster_replica_history drop rows or report NULL credits. #39557 will disable sizes that were removed from the flag but are still in use, making this routine, so this should land first.

Solution

• mz_cluster_replica_sizes lists every size and gains a disabled column.
• A builtin schema migration replaces the table's shard (it is rewritten at every startup anyway).
• The Console's size picker filters on NOT disabled, gated on environment version so it keeps working against environments without the column.

Testing

• sqllogictest: rewrote the column list, index accounting, and plan outputs that shift with the new column.
• test/cluster test-sql-cluster-disk now expects a disabled size to be listed with disabled = true.
• test/restart disabled-cluster-replica-size-survives-restart now checks that a replica on a disabled size still joins to its size row, with disabled = true.
• Console: snapshot test for the gated picker query.

Gotchas for reviewers

• This changes what a stable mz_catalog table returns: anything that lists sizes to pick from now needs WHERE NOT disabled. Worth a release note.
• DISABLED_REPLICA_SIZES_MIN_VERSION in the Console is 26.50.0-dev and has to match the release this lands in. Bump it before merging if it lands later.
• Disabled size names, including internal ones, are now visible to every user. The Console already hides mz_% sizes.
• Other tools that pick sizes from this table need the same filter, e.g. mz-clusterctl (environment.py), whose DISTINCT ON (processes, workers) ... ORDER BY credits_per_hour could now pick a disabled size.

🤖 Generated with Claude Code