Peter, you write:
> Chris pointed out to me that there are a few cases where InterMezzo has some
> difficulties with nested transactions: look at truncate. Nesting here would
> disable the possibility in Ext3 to break the transaction up.
>
> Instead Chris suggested that we involve the underlying file system not
> through a refcount, but through an indication: please include X bytes from
> me. In order to deal with those transactions that you break up into
> multiple one, it makes sense for InterMezzo to have the option of:
>
> - adding data with the first of the sequence of transactions you do
> - adding data with the last.
I believe that when you request the size of the original transaction (up
to the max transaction size) ext3 guarantees that that many blocks will
be atomic. If you run the sibling transactions first, they are sure to
be in the same commit as the start of the transaction (given that presto
does not add very many blocks to any ext3 transaction).
The fact that the rest of the truncate transaction may span multiple
commits to disk should be transparent to presto, since the cleanup is
handled internally at recovery time. What IS important is that the
KML and expect changes are bundled with the start of the truncate/unlink,
where the inode information is changed.
Cheers, Andreas
--
Andreas Dilger \ "If a man ate a pound of pasta and a pound of antipasto,
\ would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/ -- Dogbert
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]