Issue with full backup

[riccadus]

I'm currently running the network edition of 5.0.1

I've recently noticed that sometimes the scheduled full backup reports success but doesn't seem to back up properly:

Label: full-20080409.150135.256
Type: full
Status: completed
Started: Thu, 2008/04/10 01:01:35.256 EST
Ended: Thu, 2008/04/10 01:01:51.859 EST
Redo log sequence range: 133 .. 133
Number of accounts: 1

The output of a successful full backup looks like this:

Label: full-20080404.150014.043
Type: full
Status: completed
Started: Sat, 2008/04/05 01:00:14.089 EST
Ended: Sat, 2008/04/05 01:33:57.708 EST
Redo log sequence range: 126 .. 126
Number of accounts: 52

Note the "Number of accounts: 1" instead of the expected 52. The size of the backup as reported by a "du" command is also around 500KB for the broken ones instead of the usual ~30GB of the other full backups. The backup process also takes around 15 seconds instead of an expected 30 minutes.

Incremental backups on the other hand always seem to work fine.

The disk is nowhere near full so that can't be an issue.

The backup system is not something you want to be flakey! Are error conditions reported correctly by the backup system?

Any ideas?

Thanks.

[mmorse]

Might be Bug 20785 - Full backup starts everyday
(Not 100% sure as the key to identify that was 0 account fulls and you're 1 account fulls.)

Probably time to upgrade from 5.0.1 to 5.0.5
Of course after a good zmbackup -f -s server.domain.com -a all
A copy of your /opt/zimbra and any linked directories would 'play it safe'.
5.0.5 is Released!

[riccadus]

Thanks for the response.

I'd rather be sure that this will fix my issue as frankly zimbra upgrades make me nervous. Everytime I upgrade to fix an issue some other major function appears to break (I thought I was going well this time until I noticed this problem).

Sorry for the gripe, but we pay money for this for a reason - and that's because we need the enterprise features, and they need to work! I can get flakiness for free with the open source version.

[mmorse]

Totally understand, let's get some more info: zmbackupquery -v
CLI zmbackupquery Network Edition only - Zimbra :: Wiki

[riccadus]

Output attached. I've removed the list of email addresses though.

I've just noticed that the issue seems to occur after a new account is added (i.e. the total number of accounts increases).

[mmorse]

If no full backup for an account is found, the backup process performs a full backup on this account, even if only an incremental backup was specified.

For new users there will be no existing full backup; so you're fine if the accounts listed in the 'random' fulls are the recently added accounts.

By default:
Full backup is scheduled for 1:00 a.m., every Saturday.
Incremental backups are scheduled for 1:00 a.m., Sunday through Friday.
(check your zmschedulebackup -q and crontabs)

Look at the below 2 examples:
-You've simply added a user on the day of the two bold thrusdays below going from 52 (drop 1 & add 1 on thurs?) so back up 52 then to 53 users:
-Italics is a normal full on a saturday.

Type: incremental
Started: Fri, 2008/05/09 01:00:23.409 EST
Accounts: 53

Type: full
Started: Thu, 2008/05/08 01:01:35.338 EST
Accounts: 1

Type: incremental
Started: Thu, 2008/05/08 01:00:19.234 EST
Accounts: 52

----snip---

Type: incremental
Started: Sun, 2008/04/13 01:00:21.593 EST
Accounts 52

Type: full
Started: Sat, 2008/04/12 01:00:13.311 EST
Accounts: 52

Type: incremental
Started: Fri, 2008/04/11 01:00:33.542 EST
Accounts: 52

Type: full
Started: Thu, 2008/04/10 01:01:35.256 EST
Accounts: 1

Type: incremental
Started: Thu, 2008/04/10 01:00:27.421 EST
Accounts: 51

Type: incremental
Status: completed
Started: Wed, 2008/04/09 01:00:20.844 EST
Number: 52

Notice that wed at the end - Did you remove one account then add another on thurs?
Wed: 52
Thurs: drop 1 & add 1? (52)
Fri: 52

Either your getting new workers/doing account maintenance on Thursdays or there's something scheduled on one account somewhere? (Do the 1 account's omitted match?)

[mmorse]

All your action happens to be on Thursday.

So either your getting new workers and doing account maintenance on Thursdays or there's something else scheduled somewhere? (check your zmschedulebackup -q and crontabs)

I show you going from 52, drop 1 & add 1 on first thurs so back up to 52, then to 53 users.

If those 2 Thursday fulls are recently added accounts then you're fine, assuming their small because they don't have much mail yet.

[riccadus]

Hmmm... ok thanks for clarifying that this is expected behaviour.

Maybe I only add accounts on Thursday... strange coincidence but the sample size is too small to draw any grand conclusions.

Where this was biting me in the backside was that I was performing offsite backups by copying the most recent full backup onto an external drive. Alas now my life is not so simple...

[mmorse]

Where those 1 account fulls indeed new users?