[Date Prev][Date Next]
(ITS#8475) Feature request: MDB low durability transactions
Full_Name: Ben Trask
Submission from: (NULL) (22.214.171.124)
Transaction commits are one of the few bottlenecks in MDB, because it has to
fsync twice, sequentially.
I think MDB could support mixed low and high durability transactions in the same
database by adding per-page checksums and a third root page. The idea is that
when committing a low-durability transaction, no fsyncs are performed. One of
the two other roots is always consistent on-disk to serve as a "backstop" in the
event of power loss/kernel panic. The two other pages are allowed to alternate,
with consistency only ensured by checksums.
During a restore after a crash, the non-durable roots (if any) get compared
against the third durable root. Any new pages get checksum'd to ensure
integrity. In the worst case, the durable root is used.