Re: [Maria-developers] Features Wish-list in MariaDB 10.1.
So my opinion on these below two points was requested, let's keep it on the public mailing list:
* Crash-Safe replication with filename and position (not depending on GTIDs) Something similar as what is described in the following link would be good: http://blog.booking.com/better_crash_safe_replication_for_mysql.html * Have a way to avoid "SELECT binlog_gtid_pos" when a slave connect to the master in MASTER_USE_GTID = no
So my personal opinion on this is that I prefer to focus on improving the MariaDB GTID implementation. Considerable effort was invested into this, and the idea is that future development will focus on GTID mode, keeping non-GTID mode only as a backward compatibility. Significant effort was made to ensure that MariaDB GTID is usable for all existing use-cases of replication. For the first point, GTID replication is already crash-safe. For the second point, for GTID we in any case need a solution for locating GTID position quickly, otherwise GTID replication can have performance problems in case of frequent slave connects (and Jonas Oreland has promised a patch for this any day now ;-). And once this solution is in, the second point should be moot anyway. Of course, others may have a different opinion. For the second case, I suppose the easiest thing is just to remove the code that does SELECT binlog_gtid_pos() from the slave, if no-one cares about that functionality anyway... Hope this helps, - Kristian.
update on GTID to binlog position handling 1) there is already an implementation available for the gtid index at https://code.google.com/p/google-mysql/ 2) however, we (i) have re-implemented that as part of a different project. (made a standalone class that provides functionality that includes reasonably comprehensive unit tests) 3) i've been waiting for pavels republishing of our internal tree, and then "port" that patch to your branch (e.g MariaDB 10.x) But, it already quite possible to "port" the published version to MariaDB 10.x 4) I don't know exact state of pavels republishing effort. Hope this helps! /Jonas On Fri, May 22, 2015 at 12:42 PM, Kristian Nielsen <knielsen@knielsen-hq.org
wrote:
So my opinion on these below two points was requested, let's keep it on the public mailing list:
* Crash-Safe replication with filename and position (not depending on GTIDs) Something similar as what is described in the following link would be good: http://blog.booking.com/better_crash_safe_replication_for_mysql.html * Have a way to avoid "SELECT binlog_gtid_pos" when a slave connect to the master in MASTER_USE_GTID = no
So my personal opinion on this is that I prefer to focus on improving the MariaDB GTID implementation. Considerable effort was invested into this, and the idea is that future development will focus on GTID mode, keeping non-GTID mode only as a backward compatibility. Significant effort was made to ensure that MariaDB GTID is usable for all existing use-cases of replication.
For the first point, GTID replication is already crash-safe.
For the second point, for GTID we in any case need a solution for locating GTID position quickly, otherwise GTID replication can have performance problems in case of frequent slave connects (and Jonas Oreland has promised a patch for this any day now ;-). And once this solution is in, the second point should be moot anyway.
Of course, others may have a different opinion. For the second case, I suppose the easiest thing is just to remove the code that does SELECT binlog_gtid_pos() from the slave, if no-one cares about that functionality anyway...
Hope this helps,
- Kristian.
_______________________________________________ Mailing list: https://launchpad.net/~maria-developers Post to : maria-developers@lists.launchpad.net Unsubscribe : https://launchpad.net/~maria-developers More help : https://help.launchpad.net/ListHelp
Jonas Oreland <jonaso@google.com> writes:
1) there is already an implementation available for the gtid index at https://code.google.com/p/google-mysql/
Oh, thanks, I did not realise that the patch was already published.
2) however, we (i) have re-implemented that as part of a different project. (made a standalone class that provides functionality that includes reasonably comprehensive unit tests)
Right, that seems like a useful thing to have (I mean in general, having the ability to work with GTID sequencing in binlog, outside of the server code).
But, it already quite possible to "port" the published version to MariaDB 10.x
Ok, cool. I'll start taking a look, I found the patch as commit a266fea907d1ed37a602d009b291ad6d327fae4f in the tree. Let me know if there is anything you want me to be aware of with porting the patch - eg. to avoid conflicts with your newer version or whatever. - Kristian.
participants (2)
-
Jonas Oreland -
Kristian Nielsen