Hi,
This is an intended design rather than a bug.
A PreFillEntryLong is essentially a batch of updates occurred over
time. Such a batch is flushed to disk upon the entry/batch is full. If
a system crash occurred before the current entry/batch is flushed,
there is no guarantee that updates from the current batch is available
for read upon restart. However, the indexes.dat maintains a SCN
(monotonically increasing system change number) from the last
successfully flushed batch. Such a SCN represents a checkpoint, from
where a client should re-apply lost updates.
Thanks.
Jingwei