Re: [BackupPC-users] Backups getting behind

[Craig B. <cr...@at...>]

> I have one very long-running backup (a 100 gig partition) and
> it appears that when this is doing a full backup the other
> machines get behind instead of starting $Conf{MaxBackups}
> at the hourly intervals: 
> 
>  PCs will be next queued at 12/13 11:00. 
>    Other info: 
>    528 pending backup requests from last scheduled wakeup, 
>    0 pending user backup requests, 
>    1 pending command requests, 
> 
> What makes them stay 'pending'?  I saw the same thing when I
> had a problem with ssh to one machine that made that backup
> run hang until I killed it.  Shouldn't it be counted as one
> job while the rest of the system continues? 

Normally, yes.  But once a day BackupPC needs to wait for all
jobs to finish so it can run BackupPC_nightly.  I suspect this
is what is happening.

The BackupPC_nightly script (that cleans the pool and does other
administrative work) needs to run when there are no other backups
running (to avoid race conditions).  So on the first wakeup after
midnight all the PCs are queued, and then BackupPC_nightly is
queued.  BackupPC_nightly acts like a "block" on the queue: it
won't run until all prior jobs are complete, and no more newer
requests will be run until BackupPC_nightly finishes.  So a single
long-running job will stall everything else if it collides with
BackupPC_nightly.

How long does the backup take?  If it is more than 24 hours then
there is not a good solution.  Perhaps BackupPC_nightly needs
an option to only run every 24hrs * n, or perhaps it can be
setup to run at the same time as backups.

Perhaps things are worse if it has settled on a schedule where
the long backup starts a little before BackupPC_nightly.  Eg,
if you wakeup at 11pm and 1am, and the long backups happens
to start at 11pm, then BackupPC_nightly gets scheduled at 1am,
and everything after that will then be stalled until the long
backup and BackupPC_nightly are done.  If so, shift the backup
start time to after 1am (eg: by manually running an incremental
early in the morning, or changing IncrPeriod to, say, 1.2 for a
day, and then changing it back).

Craig