On 6/29/11 2:47 PM, Rajesh wrote:
Tell me if I am wrong. There can be multiple indexes for a
partition(depending on which attributes user wants to index)
Right.
and each index
is a different file. So it is possible that some indexes are updated and
others are not updated when power goes/ of program is killed.
yes. Bt this is not an issue. In this case, the index can be rebuilt
from the master table, using the configuration which describes the index
we need.
I did not find any transaction log implementation also. Do it is correct to
assume that it is possible for some indexes or even the Master Table get
corrupted(Partially updated) when some system failure happens?
yes, but this has nothing to do with transaction. The only thing is that
the system *must* be able to guarantee that when started, the indexes
will be correctly updated. We have a journal mechanism for that, where
each modification is tagged as ACK'ed when the backend is done.
Note that it is currently being refactored heavily in 2.0. Also note
that's really really a bit more complex than just having a journal...
--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com
