That's an impossible problem to solve, which is one of the many reasons
that retrievals consider the index to be potentially noisy.
That said, in the original implementation, at least, a transaction
couldn't commit until everything touched had been written. But what
constituted a write was always problematical.
NuoDB, incidentally, allows an administrator to specify exactly what is
meant by "commit." Among the options:
* Commit message sent to an storage manager
* Confirmation of the commit message from the storage manager
* Confirmation of the commit message and everything before it has been
written to stable storage by a storage manager
* Any of the above from a controllable subset of storage managers.
Write guarantees are implemented by logging replication messages to a
journal written with non-buffered writes. You can't really rely on disk
microcode to honor non-buffered writes -- they cheat to win benchmarks.
But, hey, what else can you do?
On 5/8/2014 10:13 AM, Dimitry Sibiryakov wrote:
Hello, All.
If a record is inserted or modified, what will be order of page written:
data page
first then index page or vice versa?
------------------------------------------------------------------------------
Is your legacy SCM system holding you back? Join Perforce May 7 to find out:
• 3 signs your SCM is hindering your productivity
• Requirements for releasing software faster
• Expert tips and advice for migrating your SCM now
http://p.sf.net/sfu/perforce
Firebird-Devel mailing list, web interface at
https://lists.sourceforge.net/lists/listinfo/firebird-devel
