Critical Description
On cluster activation we restore partition states from checkpoint and WAL. But before that action we pre-create partitions by ideal assignment on "beforeExchange" phase and own it in case of first or next activation. This partition state change is logged to WAL and override actual last state of partition during restore.
Possible solutions:
1) Pre-create partitions after actual restore.
2) Do not log to WAL partition own on pre-create phase.
Attachments
Issue Links
- links to
