>>>>> "Mike" == Mike Grice <Mike@???> writes:
Mike> On 27 Aug 2006, at 11:25, Radim Kolar wrote:
>> Are there any plans to implement write support for .db
>> (Berkeley db) files? Exim can read them, but not write.
>>
>> It would be very useful because SQL access is much slower than
>> simple hash-file and SQL connections have not fast start up.
I'd like to correct (hopefully accurately) a few errors about
(Sleepycat's) Berkeley DB :-
Mike> 1) To write to a berkeley db, you need to get an exclusive
Mike> write lock. This can cause other processes to deadlock
Mike> while they are waiting. See any comparable system where
Mike> contention is high, it can really be a sticking point that
Mike> brings the whole thing crashing down round you (example:
Mike> shared SpamAssassin AWL or Bayes DB). With SQL you can
Mike> generally implement a table schema that can use row-level
Mike> locking (such as InnoDB), which would eliminate deadlocks in
Mike> this way at least.
Well, Berkeley DB has a transactional mode which uses page locking
although I don't know how it works for heavy updates and I'd advise
anyone writing a BDB application with heavy writing to look at
OpenLDAP's experience with it.
Mike> 3) Berkeley DB doesn't work over NFS. Period.
Incorrect, see
http://dev.sleepycat.com/resources/faq_show.html?id=125
By the way, I'm not advocating using BDB.
Sincerely,
Adrian Phillips
--
Who really wrote the works of William Shakespeare ?
http://www.pbs.org/wgbh/pages/frontline/shakespeare/
