Replica Set Data Synchronization

Logical Initial Sync Process

When you perform a logical initial sync, MongoDB:

1. Clones all databases except the local database. To clone, the mongod scans every collection in each source member database and inserts all data into its own copies of these collections.

2. Builds all collection indexes as the documents are copied for each collection.

3. Pulls newly added oplog records during the data copy. Ensure that the destination member has enough disk space in the local database to temporarily store these oplog records for the duration of this data copy stage.

4. Applies all changes to the data set. Using the oplog from the source member, the mongod updates its data set to reflect the current state of the replica set.

When the initial sync finishes, the member transitions from STARTUP2 to SECONDARY.

To perform an initial sync, see Resync a Member of a Self-Managed Replica Set.

File Copy Based Initial Sync

Available in MongoDB Enterprise only.

File copy based initial sync runs the initial sync process by copying and moving files on the file system. This sync method can be faster than logical initial sync.

Initial Sync on NVMe Clusters

You must perform an initial sync on clusters that use the local Non-Volatile Memory Express (NVMe) SSD storage option, including if you're using Atlas auto-scaling. Atlas NVMe clusters auto-scale to the next higher tier when 90% of the storage space is full. An initial sync takes longer to complete compared to subsequent syncs, and reduces the performance of the primary from which the data is read.

Fault Tolerance

If a destination member performing initial sync encounters a persistent network error during the sync process, the destination member restarts the initial sync process from the beginning.

A destination member performing initial sync can attempt to resume the sync process if interrupted by a temporary network error, collection drop, or collection rename.

By default, the destination member tries to resume initial sync for 24 hours. You can use the initialSyncTransientErrorRetryPeriodSeconds server parameter to control the amount of time the destination member attempts to resume initial sync. If the destination member cannot successfully resume the initial sync process during the configured time period, it selects a new healthy source member from the replica set and restarts the initial synchronization process from the beginning.

The secondary attempts to restart the initial sync up to 10 times before returning a fatal error.

Initial Sync Source Selection

Initial sync source selection depends on the value of the mongod startup parameter initialSyncSourceReadPreference:

• For initialSyncSourceReadPreference set to primary (default if chainingAllowed is disabled), select the primary as the source member. If the primary is unavailable or unreachable, log an error and periodically check for primary availability.

• For initialSyncSourceReadPreference set to primaryPreferred (default for voting replica set members), attempt to select the primary as the source member. If the primary is unavailable or unreachable, perform sync source member selection from the remaining replica set members.

• For all other supported read modes, perform sync source member selection from the destination members.

Members performing initial source member selection make two passes through the list of all replica set members:

If the destination member cannot select a source member after two passes, it logs an error and waits 1 second before restarting the selection process. The secondary mongod can restart the initial sync source selection process up to 10 times before exiting with an error.

Oplog Window

The oplog window must be long enough so that a destination member can fetch any new oplog entries that occur between the start and end of the Logical Initial Sync Process. If the window isn't long enough, there is a risk that some entries may fall off the oplog before the destination member can apply them.

It is recommended that you size the oplog for additional time to fetch any new oplog entries. This allows for changes that may occur during initial syncs.

For more information, see Oplog Size.

Streaming Replication

Source members send a continuous stream of oplog entries to their destination members. Streaming replication mitigates replication lag in high-load and high-latency networks. It also:

• Reduces staleness for reads from secondaries.

• Reduces risk of losing write operations with w: 1 due to primary failover.

• Reduces latency on write operations with w: "majority" and w: >1 (that is, any write concern that requires waiting for replication).

Use the oplogFetcherUsesExhaust startup parameter to disable streaming replication and using the older replication behavior. Set the oplogFetcherUsesExhaust parameter to false only if there are any resource constraints on the source member or if you wish to limit MongoDB's usage of network bandwidth for replication.

Multithreaded Replication

MongoDB applies write operations in batches using multiple threads to improve concurrency. MongoDB groups batches by document ID (WiredTiger) and simultaneously applies each group of operations using a different thread. MongoDB always applies write operations to a given document in their original write order.

Read operations that target secondaries and are configured with a read concern level of "local" or "majority" read from a WiredTiger snapshot of the data if the read takes place on a secondary where replication batches are being applied.

Reading from a snapshot guarantees a consistent view of the data, and allows the read to occur simultaneously with the ongoing replication without the need for a lock. As a result, secondary reads requiring these read concern levels no longer need to wait for replication batches to be applied, and can be handled as they are received.

Flow Control

Administrators can limit the rate at which the primary applies its writes with the goal of keeping the majority committed lag under a configurable maximum value flowControlTargetLagSeconds.

By default, flow control is enabled.

For more information, see Flow Control.

Replication Sync Source Selection

Replication source member selection depends on the replica set chaining setting:

• With chaining enabled (default), perform source member selection from the destination members.

• With chaining disabled, select the primary as the source member. If the primary is unavailable or unreachable, log an error and periodically check for primary availability.

Members performing replication source member selection make two passes through the list of all replica set members:

If the member cannot select a sync source after two passes, it logs an error and waits 1 second before restarting the selection process.

The number of times a source member can be changed per hour is configurable by setting the maxNumSyncSourceChangesPerHour parameter.