some questions:
1. I see quite a lot fo mediawiki's tables use auto-increment as primary key but we know that this is no good for innodb, do you guy aware of this problem?
2. Why the datasebase dump (/download.wikimedia.org) do not use a unqiue format? eg. some files are XML and some files are SQL
3. Why not output as CSV format so we can use mysqlimport, which is a LOT faster for all of us.
thanks.
On 1/11/07, howard chen howachen@gmail.com wrote:
- I see quite a lot fo mediawiki's tables use auto-increment as
primary key but we know that this is no good for innodb, do you guy aware of this problem?
The only reason I can think that someone would say that autoincrement is bad w/ Innodb is because innodb is always clustered on the primary key. In some applications the use of autoincrement can result in poor locality. In our usage, I don't see why autoincrement would result in any worse locality than any other choice. Also, the use of sparse primary keys w/ Innodb results in poor packing and wasted space.
howard chen escribió:
- Why the datasebase dump (/download.wikimedia.org) do not use a
unqiue format? eg. some files are XML and some files are SQL
Time ago it used SQL. Mediawiki tables were exposed, like the cur table. With MediaWiki 1.5, there was a change on the db format, so some tables went out, and others appeared. This broke some usages. Instead of giving new sql files, which may get outdated, export changed to the current XML format, so download format will remain stable, no matter how is it really saved (e.g. article contents are not on the db, but on 'external storage').
http://meta.wikimedia.org/wiki/Data_dumps#What_happened_to_the_SQL_dumps.3F
Some files have kept SQL format, and there's probably no need to change it for now.
- Why not output as CSV format so we can use mysqlimport, which is a
LOT faster for all of us.
thanks.
I hope you're comparing against mwdumper, and not against importDump.php ;) You can always translate them into any format (SQL, CSV...) you want. The reasoning above applies here as well.
Platonides
wikitech-l@lists.wikimedia.org