In February I expressed concern that when scheduling and promoting,
amanda's planner does not properly consider DLEs with longer than
default dumpcycles.
In my case I had a default dumpcycle of 1 week. But more than
half my DLEs had custom dumpcycles of 2, 3, or 4 weeks. When
looked at as "what does the full dump schedule look like for
the next 28 days" using "amadmin DS1 balance --days 28", there
were days with NO level 0s scheduled and others with 5 and 7
large DLEs schedules totalling up to 4 times the "balance".
Sure, the "no scheduled level 0" days will get some promoted
dumps, but the monster dump days never seemed to go away.
This is not a recent configuration/installation. There have
been years to get the balance adjusted to a reasonable state.
As described in the Feb 25 message below, I lengthed my default
dumpcycle and runspercycle to 28 days. How I did it is described
in the message but is not recommended except for an experiment
like this. After the included February message I describe the
results of the experiment.
On Tue, Feb 25, 2020 at 05:32:28PM -0500, Jon LaBadie wrote:
> In a recent discussion on widely out of balance dumps and
> frequent promotions to level 0 dumps I noted this about
> my configuration.
>
> On Thu, Feb 13, 2020 at 11:50:34PM -0500, Jon LaBadie wrote:
> ...
> >
> > My default {dump/runsper}cycle is 7 days, set in amanda.conf.
> > But more than 1/2 of my DLEs have custom dumpcycles (14,21,and
> > 28 days) set in their dumptypes. You can not set custom values
> > for runspercycle in dumptypes (bummer!!).
> >
>
> I decided to give amanda's planner a different scenario, long
> dumpcycles and daily runspercycle to see if that would have an
> effect on the balances and promotions. It seems to be working.
>
> Using "amadmin <config> balance", the report shows the
> expected high balance over the next 28 days only exceeds 57%
> one time. On that one date, 5 DLEs are scheduled to dump 350%
> of the balanced amount. But the 3 days before that are well
> under balance, so perhaps the planner will promote some of
> the 5 DLEs.
>
> Here are the changes I made.
>
> In "amanda.conf", dumpcycle and runspercycle were both changed
> from 7 to 28 days.
>
> In "disklist" I had many entries that defaulted to 7 day cycles
> like:
>
> cyber.jgcomp.com Root / comp-root-tar
>
> and many othere entries with longer dumpcycles like:
>
> cyber.jgcomp.com Vault-Monthly /opt/vault-monthly {
> comp-user-tar
> dumpcycle 28
> }
>
> I made NO changes to disklist. Thus it seems like Root above
> would get the default 28 day dumpcycle rather than the desired
> 7 day. To fix this I edited the "dumptypes" file. The dumptype
> "global" is incorporated into nearly every other dumptype so
> I added a dumpcycle line.
>
> define dumptype global {
> comment "Global definitions"
> index yes
> record yes
> auth "bsdtcp"
> # added jhl 2/13/2020
> dumpcycle 7
> }
>
> So now, DLE Root gets its dumpcyle from the global dumptype
> (7 days), Vault-Monthly overrides the global value to 28 days,
> and amanda's planner uses the default value from amanda.conf,
> also 28 days.
>
> The primary downside is an administrative one. If someone
> adds a new DLE they might expect it to get the default from
> amanda.conf when that is actually overridden by the global
> dumptype.
>
> If I see any surprise effects from these changes I'll report
> back. But so far (12 days) it has been fine.
>
> Jon
Well, not a hick-cup. Ran smoothly the entire 2 months since
I made the changes. As you will see in the 28 day schedule
that follows, there are no days with zero full dumps scheduled
and the largest days are only 2x the balance size.
$ amadmin DS1 balance
due-date #fs orig MB out MB balance
----------------------------------------------
4/14 Tue 1 107003 106124 +59.5%
4/15 Wed 3 76638 65826 -1.0%
4/16 Thu 3 80792 51735 -22.2%
4/17 Fri 4 192853 133781 +101.1%
4/18 Sat 7 110353 88447 +33.0%
4/19 Sun 2 79663 58196 -12.5%
4/20 Mon 1 123755 97200 +46.1%
4/21 Tue 1 63332 51495 -22.6%
4/22 Wed 3 76638 65826 -1.0%
4/23 Thu 3 80792 51735 -22.2%
4/24 Fri 3 120982 61484 -7.6%
4/25 Sat 6 38743 17041 -74.4%
4/26 Sun 1 79663 58196 -12.5%
4/27 Mon 1 123755 97200 +46.1%
4/28 Tue 1 40558 39703 -40.3%
4/29 Wed 3 76638 65826 -1.0%
4/30 Thu 4 108679 79443 +19.4%
5/01 Fri 2 85869 27801 -58.2%
5/02 Sat 7 90187 61737 -7.2%
5/03 Sun 2 79663 58196 -12.5%
5/04 Mon 1 123755 97200 +46.1%
5/05 Tue 1 19439 19326 -70.9%
5/06 Wed 4 76638 65826 -1.0%
5/07 Thu 3 80792 51735 -22.2%
5/08 Fri 4 132022 73292 +10.2%
5/09 Sat 6 38743 17041 -74.4%
5/10 Sun 2 159184 137520 +106.7%
5/11 Mon 1 123755 97200 +46.1%
----------------------------------------------
TOTAL 80 2590884 1896132 67719
BALANCED 2556708 1862553 66519
DISTINCT 30 1100328 909156
(estimated 28 runs per dumpcycle)
I'm completely happy with the results. If you are
having balance problems and do have some longer than
default dumpcycles, consider making a similar change.
But my situation may be unusual in that I have a high
proportion of long dumpcycle DLEs.
Time to eliminate the administrative problem I allude
to in the February message and convert the 7 day
dumpcycle DLEs to "custom" dumpcycles.
Jon
--
Jon H. LaBadie j...@jgcomp.com
11226 South Shore Rd. (703) 787-0688 (H)
Reston, VA 20190 (703) 935-6720 (C)
