Compress old revisions
From Wikitech
There is a script to compress individual old revisions:
- cd /home/wikipedia/common/php-new/maintenance
- nice php compressOld.php en wikipedia -t 1 -c 100 5467442
- -t 1 : the time to sleep between batches, in seconds
- -c 100: the number of old records per batch
- 5467442: the old_id to start at, usually 1 to start. Displayed as it runs, if you stop the job, note the last value reached and use it to resume the job later. You get a warning for every record which has already been converted, so don't start much below the point you need.
Progress for en wikipedia:
- As far as 5467442 of about 6.7 million
- resume off peak with nice php compressOld.php en wikipedia -t 1 -c 100 5467442
- batch size of 5000 is OK off peak
Completed. Left about 40GB lost to fragmentation. Will take a table rebuilt to free it but that can't be done on Ariel using an InnoDB table because it will add 40GB of space to the tablespace for the copy.