Garbage Collection

How garbage is created

Dolt creates on disk garbage. Dolt transactions that do not have a corresponding Dolt commit create on disk garbage. This garbage is most noticeable after large data imports.

Specifically, writes to Dolt can result in multiple chunks of the prolly tree being rewritten, which writes a large portion of the tree. When you perform write operations without committing or delete a branch containing novel chunks, garbage is created.

How to run garbage collection

Garbage collection can be run offline using dolt gc or online using call dolt_gc().


If you have access to the server where your Dolt database is located and a Dolt sql-server is not running, navigate to the directory your database is stored in and run dolt gc. This will cycle through all the needed chunks in your database and delete those that are unnecessary. This process is CPU and memory intensive.


You can run garbage collection on your running SQL server using call dolt_gc through any connected client. To prevent concurrent writes potentially referencing garbage collected chunks, running call dolt_gc will break all open connections to the running server. In flight queries on those connections may fail and must be retried. Re-establishing connections after they are broken is safe.

At the end of the run, the connection which ran call dolt_gc() will be left open in order to deliver the results of the operation itself. The connection will be left in a terminally broken state where any attempt to run a query on it will result in the following error:

ERROR 1105 (HY000): this connection was established when this server performed an online garbage collection. this connection can no longer be used. please reconnect.

The connection should be closed. In some connection pools it can be awkward to cause a single connection to actually close. If you need to run call dolt_gc() programmatically, one work around is to use a separate connection pool with a size of 1 which can be closed after the run is successful.

NOTE: Performing GC on a cluster replica which is in standby mode is not yet supported, and running call dolt_gc() on the replica will fail.

Automated GC

We eventually want to support automated garbage collection during events that generate a lot of garbage, such as file imports. It will need to be interruptable.

Last updated