Chris LuandGitHub 8c1be63c92 ecbalancer: let a non-overflow parity shard leave a data-bearing rack (#11438) ...

The parity pass only queued shards past the per-type cap, so a single
parity shard sharing a rack with data was never a move candidate even
when an empty data-free rack existed (2+1 over 3 DCs settled 2/1/0).
Non-overflow candidates now move too, but only to a rack without data;
overflow shards keep the existing data-rack fallback.

2026-09-25 02:14:23 +08:00

..

balancer_test.go

refactor(ec_balance): make the balance planner per-volume ratio-capable (#9960)

2026-06-14 11:33:31 -07:00

balancer.go

ecbalancer: let a non-overflow parity shard leave a data-bearing rack (#11438)

2026-09-25 02:14:23 +08:00

place_test.go

ec placement: spread EC shards evenly across machines, not onto the lowest-id one (#9855)

2026-06-07 20:45:17 -07:00

place.go

ec placement: spread EC shards evenly across machines, not onto the lowest-id one (#9855)

2026-06-07 20:45:17 -07:00

shard_ratio.go

refactor(ec_balance): make the balance planner per-volume ratio-capable (#9960)

2026-06-14 11:33:31 -07:00

small_ratio_regression_test.go

ecbalancer: let a non-overflow parity shard leave a data-bearing rack (#11438)

2026-09-25 02:14:23 +08:00

snapshot_test.go

ec: skip physically near-full disks when placing EC shards (#10167)

2026-06-30 20:01:55 -07:00

snapshot.go

ec: skip physically near-full disks when placing EC shards (#10167)

2026-06-30 20:01:55 -07:00