Backup and restore strategy for AEM 6.4 Author

[deepaksuresh94]

The architecture of my AEM 6.4 author instance has the following 3 components:

Nodestore: MongoDB

Datastore: Filesystem on NAS drive mounted on the server

crx-quickstart: Situated in one of the mounts on the server

As per adobe documentation (https://helpx.adobe.com/in/experience-manager/6-4/sites/administering/using/backup-and-restore.html), MongoDB, NAS mount and crx-quickstart mount each has daily snapshots backup scheduled.

Coming to restoration activity (which is not very clear in the adobe doc), let us say that MongoDB is corrupted and needs restoration from a previous known good snapshot. Should File datastore and crx-quickstart also be restored to the same timestamp of the MongoDB snapshot? If not, will there be consistency/sync issues ? Same question arises for other 2 scenarios when File datastore and crx-quickstart fail and have to be restored.

[berliant · Accepted solution]

Ideally, a file datastore and crx-quickstart should also be restored to the same timestamp of the MongoDB snapshot and the same for your second scenario.

[Jörg_Hoh]

I agree to berliant​ regarding the ideal and always best case.

A shortcut, which you can take and which saves you to restore the datastore: If between the state before the breakdown and the time of the mongo snapshot not datastoreGC has run (and only then!) you can omit the restore of the datastore. In that case you have some unreferenced blobs, but that's not a problem (will be fixed with the next datastoreGC).

[deepaksuresh94]

But, wouldn't the backup times of file datastore and mongoDB be different even if they were to be initiated at the same time ? I feel this might still cause some mismatch between the references in MongoDB and the blobs in the file data store if data is being written into both. Please advise.

[Jörg_Hoh]

Probably yes, especially if you don't do a filesystem backup atomically. If you can (for example using a filesystem snapshot), it's atomic and has a single timestamp. But on the other hand, you need to coordinate these 2 processes (MongoDB backup and FS snapshot).

For restore I would always restore the MongoDB to the timestamp you require. And then restore the datastore from the next youngest timestamp. It might contain unreferenced blobs, but these don't matter.