The book “High Performance MySQL” states the following about using LVM snapshots with innodb tables: “All innodb files (InnoDB tablespace files and InnoDB transaction logs) must be on a single logical volume (partition).” Here is portion of a df command performed on one of our hosts:
/dev/mapper/vg01-db 2.5T 2.0T 567G 78% /db /dev/mapper/vg00-innodb 8.0G 2.0G 6.1G 25% /db/innodb /dev/mapper/vg02-binlog 503G 140G 363G 28% /db/binlog /dev/mapper/vg06-data4 755G 652G 103G 87% /db/data /dev/mapper/vgc2-data8 6.2T 644G 5.6T 11% /db/data8 /dev/mapper/vgc3-data9 6.2T 1.8T 4.5T 29% /db/data9 Where /db/innodb contains the innodb logs and the one ibdata file. However, we use innodb_file_per_table so all the /db/datax filesystems have .ibd files (many of the tables in the datadir, /db/data, are sym-linked to /db/data8 and /db/data9 where the data actually resides. We use LVM snapshots to move the data around, since our databases are several terabytes. Does this mean our snapshots are inconsistent? There’s no way we can get all innodb data on a single partition. Thanks, Jim Lyons -- Jim Lyons Web developer / Database administrator http://www.weblyons.com
