On 01/27/2016 01:35 PM, Gene Heskett wrote:

> Sounds to me like you are doing these backups by hand, and IMO

No siree. Cron is a wonderful thing.

> If you don't have a user "amanda" make one, then put the backup into the 
> amanda crontab to be done in the middle of the night when the machine is 
> mostly idle.

Her name is actually amandabackup (less likely to hire someone
with that exact name). The cron job has been running for years,
overseen by someone else. As usual when I inherit something, I
take a look at what it is doing to be sure it convinces me.

I'm sort of glad I haven't needed to use these tapes to restore.

> But amanda will catch up the next day if you have autoflush set true in 
> the amanda.conf file, so thats not a biggie in my observations.

There are complicating factors. There's a flush-threshold-dumped
setting, so she sometimes leaves stuff in the holdingdisk on
purpose. Tape drive compression is turned on, so she doesn't
really know how the holdingdisk size compares to what will
actually fit on the tape, and her guesses never improve with
experience, because she doesn't ask the tape drive after writing
(I started a different thread on -hackers about that one. There
are SCSI requests for "how much compressed space did that
really use?", though maybe not all HW or drivers support them).
I might turn on software compression and turn it off at the
drive, just so she'll know the real sizes and her fit algo
decisions will be better. Plus there's a runtapes setting.
Sometimes she stops because she hit that limit.

Last night she happened to write no tapes at all, meanwhile there
are dumps sitting in the holdingdisk that would be required
to make use of tapes she did write three days ago.

(A manually-started amflush is running as I type. :)

I am not averse to revising any of the config settings my
predecessor made. But the available settings that can be
configured seem to be things like thresholds that could
be fiddled with and probably increase the /likelihood/ of
getting dumps written in an order that makes sure the tapes
are useful. Sure, a probabilistic improvement is better
than a poke in the eye.

It just seems like a more direct and deterministic approach
would be a taperalgo that /chooses/ an order on purpose that
reflects the dump dependencies, and if there isn't one yet,
how hard would it be?

Are there amanda devs who follow this list and might know?

-Chap

Reply via email to