Docs Menu

Docs HomeDevelop ApplicationsMongoDB Manual

The AutoMerger

On this page

  • Behavior
  • Details
  • Example

Starting in MongoDB 7.0, the balancer can automatically merge chunks that meet the mergeability requirements.

The AutoMerger runs in the background as part of balancing operations. For most use cases, the default settings perform well. For details on which settings to customize for your deployment, see AutoMerger Policy.

When the AutoMerger runs, it squashes together all sequences of mergeable chunks for each shard of each collection.

Unless explicitly disabled, the AutoMerger starts the first time the balancer is enabled and pauses for the next autoMergerIntervalSecs after the routine drains.

When AutoMerger is enabled, automerging happens every autoMergerIntervalSecs seconds.

For a given collection, AutoMerger guarantees that subsequent merges are delayed at least the amount specified by autoMergerThrottlingMS.

If a balancing window is set, AutoMerger only runs during the window.

Automerging happens as part of balancing operations. In order to decide if and when to execute automerger, the settings are taken into account in this order:

  1. Global balancing settings

  2. Per-collection balancing settings (configured by configureCollectionBalancing)

  3. Global AutoMerger settings

  4. Per-collection AutoMerger settings (configured by configureCollectionBalancing)

mergeAllChunksOnShard finds and merges all mergeable chunks for a collection on the same shard. Two or more contiguous chunks in the same collection are mergeable when they meet all of these conditions:

  • They are owned by the same shard.

  • They are not jumbo chunks. jumbo chunks are not mergeable because they cannot participate in migrations.

  • Their history can be purged safely, without breaking transactions and snapshot reads:

This example assumes that history is empty for all chunks and all chunks are non-jumbo. Since both conditions are true, all contiguous intervals on the same shard are mergeable.

These chunks belong to a collection named coll with shard key x. There are nine chunks in total.

Chunk ID
Min
Max
Shard
A
x: 0
x: 10
Shard0
B
x: 10
x: 20
Shard0
C
x: 20
x: 30
Shard0
D
x: 30
x: 40
Shard0
E
x: 40
x: 50
Shard1
F
x: 50
x: 60
Shard1
G
x: 60
x: 70
Shard0
H
x: 70
x: 80
Shard0
I
x: 80
x: 90
Shard1
1
db.adminCommand( { mergeAllChunksOnShard: "db.coll", shard: "Shard0" } )

This command merges the contiguous sequences of chunks:

  • A-B-C-D

  • G-H

2
db.adminCommand( { mergeAllChunksOnShard: "db.coll", shard: "Shard1" } )

This command merges the contiguous sequences of chunks E-F.

After these commands have completed, the contiguous chunks have been merged. There are four total chunks instead of the original nine.

Chunk ID
Min
Max
Shard
A-B-C-D
x: 0
x: 40
Shard0
E-F
x: 40
x: 60
Shard1
G-H
x: 60
x: 80
Shard0
I
x: 80
x: 90
Shard1
← Migrate Ranges in a Sharded Cluster