Hi,
This afternoon, I migrated one of the main production English Wikipedia slaves, db59, to MariaDB 5.5.28. We've previously been testing 5.5.27 on the primary research slave, and I've been testing the current build for the last few days on a slave in eqiad. All has looked good, and I spent the last few days adapting our monitoring and metrics collection tools to the new version, and building binary packages that meet our needs.
A main gotcha in major version upgrades is performance regressions due to changes in query plans. I've seen no sign of this, and my initial assessment is that performance for our workload is on par with or slightly improved over the 5.1 facebook patchset.
Taking the times of 100% of all queries over regular sample windows, the average query time across all enwiki slave queries is about 8% faster with MariaDB vs. our production build of 5.1-fb. Some queries types are 10-15% faster, some are 3% slower, and nothing looks aberrant beyond those bounds. Overall throughput as measured by qps has generally been improved by 2-10%. I wouldn't draw any conclusions from this data yet, more is needed to filter out noise, but it's positive.
MariaDB has some nice performance improvements that our workload doesn't really hit (better query optimization and index usage during joins, much better sub query support) but there are also some things, such as full utilization of the primary key embedded on the right of every secondary index that we can take advantage of (and improve our schema around) once prod is fully upgraded, hopefully over the next 1-2 months.
The main goal of migrating to MariaDB is not performance driven. More so, I think it's in WMF's and the open source communities interest to coalesce around the MariaDB Foundation as the best route to ensuring a truly open and well supported future for mysql derived database technology. Performance gains along the way are icing on the cake.
-Asher
Asher,
This is awesome! Thank you for your hard, careful work and dedication in taking this huge first step in moving to MariaDB!
--peter
On Tue, Dec 11, 2012 at 4:10 PM, Asher Feldman afeldman@wikimedia.orgwrote:
Hi,
This afternoon, I migrated one of the main production English Wikipedia slaves, db59, to MariaDB 5.5.28. We've previously been testing 5.5.27 on the primary research slave, and I've been testing the current build for the last few days on a slave in eqiad. All has looked good, and I spent the last few days adapting our monitoring and metrics collection tools to the new version, and building binary packages that meet our needs.
A main gotcha in major version upgrades is performance regressions due to changes in query plans. I've seen no sign of this, and my initial assessment is that performance for our workload is on par with or slightly improved over the 5.1 facebook patchset.
Taking the times of 100% of all queries over regular sample windows, the average query time across all enwiki slave queries is about 8% faster with MariaDB vs. our production build of 5.1-fb. Some queries types are 10-15% faster, some are 3% slower, and nothing looks aberrant beyond those bounds. Overall throughput as measured by qps has generally been improved by 2-10%. I wouldn't draw any conclusions from this data yet, more is needed to filter out noise, but it's positive.
MariaDB has some nice performance improvements that our workload doesn't really hit (better query optimization and index usage during joins, much better sub query support) but there are also some things, such as full utilization of the primary key embedded on the right of every secondary index that we can take advantage of (and improve our schema around) once prod is fully upgraded, hopefully over the next 1-2 months.
The main goal of migrating to MariaDB is not performance driven. More so, I think it's in WMF's and the open source communities interest to coalesce around the MariaDB Foundation as the best route to ensuring a truly open and well supported future for mysql derived database technology. Performance gains along the way are icing on the cake.
-Asher
Ops mailing list Ops@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/ops
Hello Asher,
Thanks so much for your hard work on supporting our MariaDB migration. I think that it's a big step for the MariaDB Foundation.
— Patrick
On Tue, Dec 11, 2012 at 4:17 PM, Peter Youngmeister py@wikimedia.org wrote:
Asher,
This is awesome! Thank you for your hard, careful work and dedication in taking this huge first step in moving to MariaDB!
--peter
On Tue, Dec 11, 2012 at 4:10 PM, Asher Feldman afeldman@wikimedia.org wrote:
Hi,
This afternoon, I migrated one of the main production English Wikipedia slaves, db59, to MariaDB 5.5.28. We've previously been testing 5.5.27 on the primary research slave, and I've been testing the current build for the last few days on a slave in eqiad. All has looked good, and I spent the last few days adapting our monitoring and metrics collection tools to the new version, and building binary packages that meet our needs.
A main gotcha in major version upgrades is performance regressions due to changes in query plans. I've seen no sign of this, and my initial assessment is that performance for our workload is on par with or slightly improved over the 5.1 facebook patchset.
Taking the times of 100% of all queries over regular sample windows, the average query time across all enwiki slave queries is about 8% faster with MariaDB vs. our production build of 5.1-fb. Some queries types are 10-15% faster, some are 3% slower, and nothing looks aberrant beyond those bounds. Overall throughput as measured by qps has generally been improved by 2-10%. I wouldn't draw any conclusions from this data yet, more is needed to filter out noise, but it's positive.
MariaDB has some nice performance improvements that our workload doesn't really hit (better query optimization and index usage during joins, much better sub query support) but there are also some things, such as full utilization of the primary key embedded on the right of every secondary index that we can take advantage of (and improve our schema around) once prod is fully upgraded, hopefully over the next 1-2 months.
The main goal of migrating to MariaDB is not performance driven. More so, I think it's in WMF's and the open source communities interest to coalesce around the MariaDB Foundation as the best route to ensuring a truly open and well supported future for mysql derived database technology. Performance gains along the way are icing on the cake.
-Asher
Ops mailing list Ops@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/ops
Ops mailing list Ops@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/ops
Nice!
The main goal of migrating to MariaDB is not performance driven. More so, I think it's in WMF's and the open source communities interest to coalesce around the MariaDB Foundation as the best route to ensuring a truly open and well supported future for mysql derived database technology. Performance gains along the way are icing on the cake.
If it works out, then at some point we should probably tell the MariaDB peeos that they can mention that the WMF uses it. :-)
On Tue, Dec 11, 2012 at 5:49 PM, Terry Chay tchay@wikimedia.org wrote:
Nice!
The main goal of migrating to MariaDB is not performance driven. More
so, I think it's in WMF's and the open source communities interest to coalesce around the MariaDB Foundation as the best route to ensuring a truly open and well supported future for mysql derived database technology. Performance gains along the way are icing on the cake.
If it works out, then at some point we should probably tell the MariaDB peeos that they can mention that the WMF uses it. :-)
We've been talking to Monty Widenius who visited the WMF office prior to the Foundation announcement, and are fostering mutual support between the Wikimedia and MariaDB Foundations. Win-win for the open source community at large!
-A
Asher,
This is great news! Thanks for your perseverance and setting up MariaDB 5.5 for en.wp :-) Monty will be thrilled!
Best, Alolita
On Wed, Dec 12, 2012 at 7:53 AM, Asher Feldman afeldman@wikimedia.org wrote:
On Tue, Dec 11, 2012 at 5:49 PM, Terry Chay tchay@wikimedia.org wrote:
Nice!
The main goal of migrating to MariaDB is not performance driven. More
so, I think it's in WMF's and the open source communities interest to coalesce around the MariaDB Foundation as the best route to ensuring a truly open and well supported future for mysql derived database technology. Performance gains along the way are icing on the cake.
If it works out, then at some point we should probably tell the MariaDB peeos that they can mention that the WMF uses it. :-)
We've been talking to Monty Widenius who visited the WMF office prior to the Foundation announcement, and are fostering mutual support between the Wikimedia and MariaDB Foundations. Win-win for the open source community at large!
-A _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Awesome!
On Tue, Dec 11, 2012 at 8:49 PM, Terry Chay tchay@wikimedia.org wrote:
If it works out, then at some point we should probably tell the MariaDB peeos that they can mention that the WMF uses it. :-)
The enwiki article on MariaDB has claimed MediaWiki "officially" supports it since October 2012.[1] Perhaps that's a {{citation needed}}.
[1]: https://en.wikipedia.org/w/index.php?diff=prev&oldid=519286745
On 12 December 2012 14:28, Brad Jorsch bjorsch@wikimedia.org wrote:
The enwiki article on MariaDB has claimed MediaWiki "officially" supports it since October 2012.[1] Perhaps that's a {{citation needed}}. [1]: https://en.wikipedia.org/w/index.php?diff=prev&oldid=519286745
I would have thought it would have Just Worked in MediaWiki (the software) for a long time.
- d.
On Wed, Dec 12, 2012 at 9:34 AM, David Gerard dgerard@gmail.com wrote:
On 12 December 2012 14:28, Brad Jorsch bjorsch@wikimedia.org wrote:
The enwiki article on MariaDB has claimed MediaWiki "officially" supports it since October 2012.[1] Perhaps that's a {{citation needed}}. [1]: https://en.wikipedia.org/w/index.php?diff=prev&oldid=519286745
I would have thought it would have Just Worked in MediaWiki (the software) for a long time.
It probably did. But [[mw:Manual:Installation guide]] still says it's not "officially" supported.
On 12 December 2012 14:38, Brad Jorsch bjorsch@wikimedia.org wrote:
On Wed, Dec 12, 2012 at 9:34 AM, David Gerard dgerard@gmail.com wrote:
On 12 December 2012 14:28, Brad Jorsch bjorsch@wikimedia.org wrote:
The enwiki article on MariaDB has claimed MediaWiki "officially" supports it since October 2012.[1] Perhaps that's a {{citation needed}}. [1]: https://en.wikipedia.org/w/index.php?diff=prev&oldid=519286745
I would have thought it would have Just Worked in MediaWiki (the software) for a long time.
It probably did. But [[mw:Manual:Installation guide]] still says it's not "officially" supported.
I expect when it's on more of the cluster someone can announce its enhanced status here :-)
- d.
This is awesome! Is there any write-up of the migration process floating around?
-Tom
On Wed, Dec 12, 2012 at 10:06 AM, David Gerard dgerard@gmail.com wrote:
On 12 December 2012 14:38, Brad Jorsch bjorsch@wikimedia.org wrote:
On Wed, Dec 12, 2012 at 9:34 AM, David Gerard dgerard@gmail.com wrote:
On 12 December 2012 14:28, Brad Jorsch bjorsch@wikimedia.org wrote:
The enwiki article on MariaDB has claimed MediaWiki "officially" supports it since October 2012.[1] Perhaps that's a {{citation needed}}. [1]: https://en.wikipedia.org/w/index.php?diff=prev&oldid=519286745
I would have thought it would have Just Worked in MediaWiki (the software) for a long time.
It probably did. But [[mw:Manual:Installation guide]] still says it's not "officially" supported.
I expect when it's on more of the cluster someone can announce its enhanced status here :-)
- d.
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
On 12 December 2012 15:32, Thomas Fellows thomas.fellows@gmail.com wrote:
This is awesome! Is there any write-up of the migration process floating around?
+1
In fact, this would be a nice thing to put on the WMF blog. It'll certainly get a lot of linkage and reporting around the geekosphere.
- d.
On Wednesday, December 12, 2012, David Gerard wrote:
On 12 December 2012 15:32, Thomas Fellows <thomas.fellows@gmail.comjavascript:;> wrote:
This is awesome! Is there any write-up of the migration process floating around?
+1
In fact, this would be a nice thing to put on the WMF blog. It'll certainly get a lot of linkage and reporting around the geekosphere.
A detailed blog post is definitely my intent, I'm just waiting until at least one major project is 100% on mariadb and I have more data and hence confidence in drawn conclusions. I don't think that's far off at all, potentially later this month. If that occurs and goes well, the eqiad data center migration in late January may also be a migration to all mariadb.
On Wed, 12 Dec 2012 06:45:24 -0800, Antoine Musso hashar+wmf@free.fr wrote:
Le 12/12/12 01:10, Asher Feldman a écrit :
This afternoon, I migrated one of the main production English Wikipedia slaves, db59, to MariaDB 5.5.28.
Congratulations :-)
Out of curiosity, have you looked at Drizzle too?
IIRC Drizzle uses a completely different client and we'd need to write a new database class for it.
...something I've wished I could do for a long time.
On Wed, Dec 12, 2012 at 6:45 AM, Antoine Musso hashar+wmf@free.fr wrote:
Le 12/12/12 01:10, Asher Feldman a écrit :
This afternoon, I migrated one of the main production English Wikipedia slaves, db59, to MariaDB 5.5.28.
Congratulations :-)
Out of curiosity, have you looked at Drizzle too?
I've spoken with Drizzle developers at OSCON in the past. I haven't seen anyone advocate it as a production quality database though, and it doesn't currently seem to have a lot of development momentum behind it, with Brian Aker no longer putting in a lot of time. Lots of interesting ideas and features, especially around replication, but they make it incompatible with MySQL in enough ways where a gradual migration wouldn't be practical even if it was otherwise desirable.
-A
wikitech-l@lists.wikimedia.org