Comments on: MySQL Proxy for sharding http://fschiettecatte.wordpress.com/2008/04/18/mysql-proxy-for-sharding/ Thoughts from the edge of the 'net Mon, 13 Oct 2008 12:01:44 +0000 http://wordpress.org/?v=MU hourly 1 By: HScale and MySQL Proxy « François Schiettecatte’s Blog http://fschiettecatte.wordpress.com/2008/04/18/mysql-proxy-for-sharding/#comment-4277 HScale and MySQL Proxy « François Schiettecatte’s Blog Tue, 26 Aug 2008 22:58:28 +0000 http://fschiettecatte.wordpress.com/?p=445#comment-4277 [...] Scaling, Software Development — François Schiettecatte @ 5:58 pm I have written about HScale before and MySQL Proxy, raising a few questions about fault tolerance, but I did not anticipate the issues they ran [...] [...] Scaling, Software Development — François Schiettecatte @ 5:58 pm I have written about HScale before and MySQL Proxy, raising a few questions about fault tolerance, but I did not anticipate the issues they ran [...]

]]> By: François Schiettecatte http://fschiettecatte.wordpress.com/2008/04/18/mysql-proxy-for-sharding/#comment-4251 François Schiettecatte Tue, 22 Jul 2008 21:43:27 +0000 http://fschiettecatte.wordpress.com/?p=445#comment-4251 You are very right about running multiple MySQLProxies to mitigate the single-point-of-failure issue, my mistake for not seeing that especially since I developed exactly the same solution for the search engine at Feedster. I (with others) tested a product for MySQL which would create a multi-master setup (by which I mean that two independent servers maintain the same data). It would read from the least loaded servers and write to both maintaing a log in case of the servers went down, which it would replay against that server when it came back up before bringing it online. We did not use this product because it did not work as advertised and it was complicated to install. But I think multi-master with recovery could be very powerful for installations who wanted large MySQL servers. Cross-table sharding (splitting up a table across MySQL instances) is already very good and I think would cover 95% of the user cases. But I am not sure I understand cross-database sharding. You are very right about running multiple MySQLProxies to mitigate the single-point-of-failure issue, my mistake for not seeing that especially since I developed exactly the same solution for the search engine at Feedster.

I (with others) tested a product for MySQL which would create a multi-master setup (by which I mean that two independent servers maintain the same data). It would read from the least loaded servers and write to both maintaing a log in case of the servers went down, which it would replay against that server when it came back up before bringing it online. We did not use this product because it did not work as advertised and it was complicated to install. But I think multi-master with recovery could be very powerful for installations who wanted large MySQL servers.

Cross-table sharding (splitting up a table across MySQL instances) is already very good and I think would cover 95% of the user cases. But I am not sure I understand cross-database sharding.

]]> By: karel1980 http://fschiettecatte.wordpress.com/2008/04/18/mysql-proxy-for-sharding/#comment-4248 karel1980 Tue, 22 Jul 2008 07:11:57 +0000 http://fschiettecatte.wordpress.com/?p=445#comment-4248 MySQLProxy does not necessarily introduce a single point of failure. Simply run multiple mysqlproxy instances and configure your application to fall back to the other instance when one is going down. About fault tolerancy, MySQLProxy can do R/W splitting (script included) which goes a long way OR be used to create a multimaster setup. Using it for sharding seems a bit disappointing at the moment: at the moment, HScale (a sharding-with-MySQLProxy implementation) does not support cross-database sharding, only cross-table sharding (I assume one could use FEDERATED tables, but it's unclear to me what the pros/cons would be in that case). MySQLProxy does not necessarily introduce a single point of failure.
Simply run multiple mysqlproxy instances and configure your application to fall back to the other instance when one is going down.

About fault tolerancy, MySQLProxy can do R/W splitting (script included) which goes a long way OR be used to create a multimaster setup.

Using it for sharding seems a bit disappointing at the moment: at the moment, HScale (a sharding-with-MySQLProxy implementation) does not support cross-database sharding, only cross-table sharding (I assume one could use FEDERATED tables, but it’s unclear to me what the pros/cons would be in that case).

]]>