While debugging a multi node validation failure I found a few gaps making debugging hard:

1. disagg.preserve not enabled together with disagg.multi=1 disagg.multi_validation=1.
2. Even with preserve enabled, the first process that finishes preservation and starts core dumping seems to get the other process killed
3. When the process preserves the data it reads the latest data, but it should use the same timestamp that was used to calculate the checksum.