Sun Java System Messaging Server 6.3 Administration Guide

20.12.5 Considerations for Partial Restore

A partial restore is when only a part of the message store is restored. A full restore is when the entire message store is restore. The message store uses a single-copy message system. That is, only a single copy of any message is saved in the store as a single file. Any other instances of that message (like when a message is sent to multiple mailboxes) are stored as links to that copy. Because of this, there are implications when restoring messages. For example:

The following examples demonstrate what happens to a message used by multiple users when a partial restore is performed. Assume there are three messages, all the same, belonging to three users A, B, and C, as follows:


Example 1. In the first example, the system performs a partial backup and full restore procedure as follows:

  1. Back up mailboxes for users B and C.

  2. Delete mailboxes of users B and C.

  3. Restore the backup data from step 1.

In this example, B/INBOX/1 and C/INBOX/1 are assigned a new inode number and the message data is written to a new place on the disk. Only one message is restored; the second message is a hard link to the first message.

Example 2. In this example, the system performs a full backup and a partial restore as follows:

  1. Perform full backup.

  2. Delete mailboxes for user A.

  3. Restore mailboxes for user A.

A/INBOX/1 is assigned a new inode number.

Example 3. In this example, partial restore might require more than one attempt:

  1. Perform full backup.

    B/INBOX/1 and C/INBOX/1 are backed up as links to A/INBOX/1.

  2. Delete mailboxes for users A and B.

  3. Restore mailboxes for user B.

    The restore utilities ask the administrator to restore A/INBOX first.

  4. Restore mailboxes for users A and B.

  5. Delete mailboxes for user A (optional).

    Note –

    If you want to ensure that all messages are restored for a partial restore, you can run the imsbackup command with the -i option. The -i option backs up every message multiple times if necessary.

    If the backup device is seekable (example: a drive or tape), imsrestore seeks to the position containing A/INBOX/1 and restore it as B/INBOX/1. If the backup device is non-seekable example: a UNIX pipe), imsrestore logs the object ID and the ID of the depending (linked) object to a file, and the administrator must invoke imsrestore again with the -r option to restore the missing message references. To Restore Messages from a Mailbox that has Been Incrementally Backed-up

If you are restoring messages from a mailbox that has been incrementally backed-up, and if that mailbox exists on the server on which you wish to restore the messages, then restoring the messages requires a straightforward imesrestore. However, if you wish to restore messages from a mailbox that has been incrementally backed-up, and if that mailbox no longer exists, you must follow different restore procedures.

Use one of the following procedures to restore messages to a mailbox that does not exist on the message store server:

The reason why these instructions must be followed for restoring an incremental backup is as follows: When a mailbox has been deleted or is being migrated, the imsrestore utility recreates the mailbox with the mailbox unique identification validity and message unique identifications (UIDs) stored in the backup archive.

In the past, when imsrestore would recreate a deleted or migrated mailbox, it would assign a new UID validity to the mailbox and new UIDs to the messages. In that situation, a client with cached messages would have to resynchronize the mailbox UID validity and message UIDs. The client would have to download the new data again, increasing the workload on the server.

With the new imsrestore behavior, the client cache remains synchronized, and the restore process operates transparently with no negative impact on performance.

If a mailbox exists, imsrestore assigns new UIDs to the restored messages so that the new UIDs remain consistent with the UIDs already assigned to existing messages. To ensure UID consistency, imsrestore locks the mailbox during the restore operation. However, because imsrestore now uses the mailbox UID validity and message UIDs from the backup archive instead of assigning new UID values, UIDs could become inconsistent if you perform incremental backups and restores.

If you perform incremental backups with the -d date option of the imsbackup utility, you might have to invoke imsrestore multiple times to complete the restore operation. If incremental backups were performed, you must restore the latest full backup and all subsequent incremental backups.

New messages can be delivered to the mailbox between the restore operations, but in this case, the message UIDs can become inconsistent. To prevent inconsistency in the UIDs, you need to take one of the actions describe above.