Re: Backup methods

[Alvin Starr <alvin@xxxxxxxxxx>]

On 05/10/2018 06:29 PM, Anatoli wrote:

Actually, mysqldump performs a lock on the records it's dumping. If its for a MyISAM db, the entire table is locked. If it's for InnoDB and similar, an internal snapshot is created and only the records the dump is reading are unavailable for writing.

Mysql provides table consistency by locking but that is only table consistency.
Multiple updates across multiple tables could easily result in an inconsistent database even though the tables are individually consistent.
With some tables that take hours to dump locking the tables is problematic.
This is why some people use LVM snapshots combined with "FLUSH TABLES WITH READ LOCK".

Cyrus could also implement a per-user lock, but in reality it doesn't need that complex syncro mechanisms, a simple global write lock would be enough (reading would not be affected, son only I, not I/O, and not to stop it but just to suspend). After all, the *write* lock would last only a second or so, the fs snapshots are almost instantaneous. If you can't tolerate a 1 second delay for writing in Cyrus, you are probably not a SME.

The trouble is that read operations can alter a files state so it may not be just a simple matter of a write lock.
If the mailboxes are on something like EXT4 you can do an LVM snapshot bacause of the built in auto checkpointing and for something like xfs there is freeze.

Cyrus has multiple databases that would also need to be frozen and flushed before the snapshot is taken.
If you spread your mailstorage and metadata storage over multiple file systems trying to co-ordinate snapshots becomes more complex.

And you don't need to hold the data to transfer it. You can dump it directly to a nfs share or pass it as stdout to ssh: mysqldump | xz -9 | ssh remote_server "cat > /bkp/`date +%y%m%d_%H%M`.sql.xz". With a couple of pipes more you can encrypt the data on the fly so it's secure to store it in a cheap VPS overseas... or you could upload it to dropbox.

There may be an argument that could be made for 2 backup stratagies.
1) where the mailstoreage and metadata can exist on a single volume and flushing the various databases is a short duration event. Then an LVM snapshot could be used
2) for distribted large scale mail systems where only an online live backup system can be used.

Backup for 100 users has different requirements than backup for 100000 users so why not support a few different backup strategies.

From: Jason L Tibbitts Iii
Sent: Thursday, May 10, 2018 18:41
To: Anatoli
Cc: Info-cyrus
Subject: Re: Backup methods

"A" == Anatoli  <me@xxxxxxxxxx> writes:

A> What about mysqldump > dump.sql, then mysql < dump.sql? Also a wrong
A> way and didn't have to be implemented?

No, that's exactly my point.  Thanks for making it for me!  The analog
to the way you indicated that you would like it to work would be having
the mysql server stop IO so that you can take a filesystem snapshot
while the database is in a consistent state.  But instead, the database
(like cyrus) implements a backup method which you can use to extract the
data.  And it also requires disk space to hold the backup until you can
transfer it to your backup medium.

 - J<

----
Cyrus Home Page: http://www.cyrusimap.org/
List Archives/Info: http://lists.andrew.cmu.edu/pipermail/info-cyrus/
To Unsubscribe:
https://lists.andrew.cmu.edu/mailman/listinfo/info-cyrus

-- 
Alvin Starr                   ||   land:  (905)513-7688
Netvel Inc.                   ||   Cell:  (416)806-0133
alvin@xxxxxxxxxx              ||

----
Cyrus Home Page: http://www.cyrusimap.org/
List Archives/Info: http://lists.andrew.cmu.edu/pipermail/info-cyrus/
To Unsubscribe:
https://lists.andrew.cmu.edu/mailman/listinfo/info-cyrus