Compact Command

v2.0+

The compact command compacts and defragments a collection. Indexes are rebuilt at the same time. It is conceptually similar to repairDatabase, but works on a single collection rather than an entire database.

To run (example from the mongo shell):

> db.runCommand( { compact : 'mycollectionname' } )
OR
> db.mycollection.runCommand( "compact" )

The command will not return until compaction completes. You can view intermediate progress in either the mongod log file, or by running db.currentOp() in another shell instance.

Because this blocks all other activity, the compact command returns an error when invoked on a replica set primary. To force it to run on a replica set primary, include force:true in the command as a parameter.

Running a compaction

The compact command generally uses less disk space while performing its work than repairDatabase does (though in worst case can need just as much). Additionally, compact is faster.

Although faster than repairDatabase, this command blocks all other operations while running, and is still slow. Run during scheduled maintenance. If you run the command on a replica set secondary, the secondary will automatically demote itself to a "recovery" state until the compaction is complete.

Killing a compaction

At the beginning of compaction, indexes are dropped for the collection. At the end, indexes are rebuilt. Thus, if you kill the compaction in the middle of its operation, either with killOp or a server failure, indexes may be missing from your collection. If you are running with --journal, no data should ever be lost on a crash during compaction, although on startup the indexes will not be present. (Regardless of this, always do a backup before system maintenance such as this!) When the indexes are rebuilt, they will be in the 2.0 index format.

If there's a crash while the command is running, then as long as journaling is enabled, your data will be safe.

Additionally, if a compaction operation is interrupted, much of the existing free space in a collection may become un-reusable. In this scenario, it is recommended that compaction be run again, to completion, to restore use of this free space.

Effects of a compaction

This command fully compacts the collection resulting in no Padding Factor for existing documents, but the paddingFactor number is kept for the collection (unlike when a repair is done). Thus, updates which grow documents will be slower after compaction as documents will move much more often on updates (at least at first).

You may wish to run the collstats command (db.collectionname.stats() in the shell) before and after compaction to see how the storage space changes for the collection.

Compaction may increase the total size and number of your data files, especially when running compact for the first time on a collection. Even in this case, total collection storage space consumed will not increase.

File System

This operation will not reduce the amount of disk space used on the filesystem. Storage size is the amount of data allocated within the database files, not the size/number of the files on file system.

Details

• Compact may be run on replica secondaries and slaves. Compact commands do not replicate, so each host needs to be compacted separately.
• Currently, compact is a command issued to a mongod. In a sharded environment one would do each shard separately as a maintenance operation. (This is likely to change in future versions, along with other enhancements.)
• Capped collections cannot be compacted. (The documents of a capped collection are not subject to fragmentation, however.)

Replica Sets

• Compact each member separately.
• Ideally compaction runs on a secondary (see comment regarding force:true above).
• If compact is run on a secondary, the secondary will go into "recovering" state automatically (to prevent reads from being routed to it while compacting). Once the compaction is finished, it will automatically return to secondary state.
• A partial script demonstration how to automate step down and compaction can be found here.

See Also

• repairDatabase command