[epic] ValkeyCluster scheduling & placement (shard spread, primary balance, zones, failover)

[jdheyburn]

Purpose

Umbrella for where ValkeyCluster pods run and how that ties to HA and failover: shard-level placement, even distribution of primaries across workers, zone-aware behaviour, defaults, surfacing scheduling failures, and integration with proactive failover replica selection.

Use this issue as the parent / tracking hub; link implementation issues here as they are opened.

Existing GitHub issues (anchor this epic to these)

Track	Issue	Role
Shard-aware placement design	#146	Open design: API (e.g. topologySpreadConstraints vs operator policy), defaults, failure surfacing, interaction with user affinity / future zone work.
Primaries balanced across workers	#247	Cluster-level: avoid too many primaries on one worker; rebalance after node events; complementary to #146 per that issue.
Proactive failover replica selection	#249	Replica choice for rolls / write-pause; natural tie-in for zone-aware primary preference later.

Workstreams (file or link sub-issues when ready)

1. Close / summarize [Design] Shard-aware placement policy for primaries and replicas #146 — record a written decision (Option + defaults + how strict failures surface) before treating any implementation spec as final.
2. Shard-aware placement implementation — labels, optional default soft spread, spec.placement surface, conditions, docs, E2E — blocked on [Design] Shard-aware placement policy for primaries and replicas #146 outcome.
3. Primary balance across workers (Primaries balanced across workers #247) — scheduling and/or controlled failovers to rebalance; coordinate with (2) so policies do not fight.
4. Zone-affinity for primaries (future issue) — preferred AZ, integration with feat: fail over to the highest-offset replica #249 for CLUSTER FAILOVER target selection, optional fail-back — depends on [Design] Shard-aware placement policy for primaries and replicas #146 not painting the API into a corner; often ships after (2).

Cross-cutting

• PDBs, single-shard / tiny-cluster behaviour, CI-friendly defaults.
• Discussion #228 for zone / failback context where relevant.