[Box Backup] Large accounts

[Ben Summers]

On 21 Oct 2004, at 08:54, Martin Ebourne wrote:

> Oct 18 21:23:31 gliadin bbstored[26671]: Incoming connection from 
> 192.168.1.4 port 37172 (handling in child 26692)
> Oct 18 21:23:31 gliadin bbstored[26692]: Certificate CN: 
> BACKUP-11111111
> Oct 18 21:23:32 gliadin bbstored/hk[26672]: Housekeeping giving way to 
> connection for account 0x11111111
> Oct 18 21:23:32 gliadin bbstored/hk[26672]: On housekeeping, sizes in 
> store do not match calculated sizes, correcting
> Oct 18 21:23:32 gliadin bbstored/hk[26672]: different (store,calc): 
> acc 0x11111111, used (758,1710887), old (77,57272), deleted (0,77137), 
> dirs (73,9139)
>

And this is the simple clue I need to fix this bug!

If housekeeping is in progress when the client logs in, the 
housekeeping process will stop immediately to let the client log in and 
modify the account. This is what's happening in the above logs.

Unfortunately, the housekeeping process writes the sizes it's 
calculated SO FAR back to the store info. This underestimates the size 
considerably. Next time round, it's fixed properly, and you get another 
message. The chances of an interruption are far higher with larger 
accounts, which is why Imran was seeing so many of the messages. It 
also explains the wildly fluctuating figures in his logs, which were 
really puzzling me.

So there you go. It's relieving to know that no data would be lost. In 
the worst case, you might get an account going temporarily over quota.

Lesson learned: Always ask for full extracts from logs! :-)

Thanks for all the reports, all very helpful. I'll get a patch to Imran 
for him to test today.

Ben