Question About PITR Continuity

MongoDB Percona Backup for MongoDB
, ,
1

Hello,

We are currently using MongoDB 8.x and Percona Backup for MongoDB (PBM).

In our production environment, we perform one full logical backup per day with PITR enabled continuously.

We are planning to use these backups to restore the production database to a separate MongoDB server. Our intended restore strategy is as follows:

  1. Restore a full backup to the target server only once as the initial baseline.

  2. After the initial full backup restore, continuously apply/replay the PITR backups to the target server to keep it synchronized to a desired point in time.

  3. We do not intend to restore the subsequent daily full backups to the target server.

We would like to confirm whether this restore strategy is supported and technically feasible with PBM.

The reason for this question is that we have observed in the PBM logs that PITR backup/slicing appears to be interrupted when a full backup is running.

For example:

  • 00:00 - Full logical backup starts

  • 00:00–00:30 - PITR slicing appears to be interrupted

  • 00:30 - Full backup completes

  • After 00:30 - PITR slicing resumes

In this scenario, we would like to understand how PBM handles the oplog entries generated during the period when the full backup is running.

Specifically:

  • When PITR resumes after the full backup completes, does PBM capture the oplog entries for the entire period during which PITR slicing was interrupted?

  • In other words, is there a continuous PITR/oplog chain across the full backup window, with no gap in the recoverable timeline?

  • If we restore only the first full backup and then continuously replay PITR data, can we also recover all changes that occurred during subsequent daily full backup windows?

  • Is pbm oplog-replay the appropriate/supported method for maintaining this type of restore workflow on the target server?

  • Are there any limitations or risks associated with continuously replaying PITR data on top of an initially restored full backup over multiple days?

Additionally, the PBM documentation appears to recommend using the most recent full backup as the base backup when performing a PITR restore.

Could you please clarify why the most recent full backup is recommended or required?

For example, assuming we have:

Day 1: Full Backup A + PITR
Day 2: Full Backup B + PITR
Day 3: Full Backup C + PITR

Would it be possible to restore Full Backup A only once and then continuously replay all PITR/oplog data through Day 2 and Day 3, including the periods when Full Backups B and C were running?

Or does PBM require/recommend restoring Full Backup C when restoring to a point in time on Day 3?

We would appreciate clarification on how PBM maintains PITR continuity during logical full backups and whether our proposed restore strategy is supported.

Thank you.

2

Hi, the scenario is technically feasible. PBM will resume saving oplog from the proper position after full backup is done (there should be no gaps). One thing to keep in mind is oplog apply is single threaded, so applying a large batch of oplog could take significant amount of time. This is why the most recent full backup is always recommended as the base backup when doing PITR.

For your particular use case perhaps it is better to use Percona ClusterSync for MongoDB to continuously replicate data (it handles initial data copy as well)