[HyperDB] separating reads and writes

Andy Skelton skeltoac at gmail.com
Mon Apr 6 17:41:50 GMT 2009


On Mon, Apr 6, 2009 at 12:27 PM, Emiliano Compean
<e.bardomiano at gmail.com> wrote:
> So basically, I want all reads to go to one server and writes to another.
> Is this correct?

This is generally correct but your write-only server will still be
sent read queries sometimes. Due to inevitable lags in replication, we
designed HyperDB to read from the write database after any writes.
This prevents the application reading a previous version of row
written during the same page load.

The current implementation keeps track of which datasets have received
writes. It then bypasses the slaves for all subsequent reads on those
datasets.

Cheers,
Andy


More information about the HyperDB mailing list