On 19.01.23 11:04, Albert Shih wrote:
Jan 19 10:20:37 FQDN sync_client[56922]: Error in do_sync(): bailing out! The remote Server(s) denied the operation Jan 19 10:21:27 FQDN sync_client[56922]: MAILBOX received NO response: Unknown error Jan 19 10:21:27 FQDN sync_client[56922]: Error in do_sync(): bailing out! The remote Server(s) denied the operation Jan 19 10:21:39 FQDN sync_client[56922]: IOERROR: zero length response to MAILBOX (end of file reached) Jan 19 10:21:39 FQDN sync_client[56922]: Error in do_sync(): bailing out! Bad protocol
What's in the slaves' logs when this happens?
Beside this problem, what are the actual advantage between sync_client and some classique zfs snapshot MAILBOX zfs send SNAPSHOT --> other server ?
Snapshots are consistent from a file system perspective, but not necessarily from a logical perspective of cyrus itself. Since the mailboxes and databases are split across multiple files, you might end up snapshotting an "inbetween" state with some files already updated and others not yet. Same as if the server were to crash the moment you're running the snapshot.
So you might end up having to repair the database manually to bring it back into a consistent state, losing some mails that were in flight during the time the snapshot happened.
But that doesn't seem to be too much of a constraint for your setup, since you're using your slaves more as a backup than a replication solution anyway…? ZFS replication should be fine then, as long as you're allowed to drop some mails.
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
<<< multipart/mixed; boundary=------------Fm3kxC05DCgYkGnO0moBEaen; protected-headers=v1; boundary="----------=_1675067475-224681-1"; charset="UTF-8": Unrecognized >>>
