When the stone is in restoring transaction logs, under some rare
internal conditions the pageservers may not write all the required
pages to the extents that are needed as part of a checkpoint. If
the stone is shut down, this missing information may cause the stone
to fail to restart. This problem may occur during orderly shutdowns
as well as crashes. If the stone does not shutdown, and a later
checkpoint is correct, there are no consequences of this bug.
This bug has never been observed in customer systems, it was uncovered during testing.
If the problem occurs, the extents are corrupt, and you will need to restart from the backup and replay all the transaction logs again.
Reducing the number of times you shut down a warm standby reduces the risk.
Last updated: 11/12/09