On Mon, Nov 12, 2018 at 06:40:56 -0500, Gene Heskett wrote:
> amanda@coyote:/root$ /usr/local/sbin/amadmin Daily balance
>
> due-date #fs orig MB out MB balance
> ----------------------------------------------
> 11/12 Mon 1 32963 7875 -53.6%
> 11/13 Tue 1 7688 7688 -54.7%
> 11/14 Wed 2 22109 22109 +30.3%
> 11/15 Thu 4 75027 46623 +174.9%
> 11/16 Fri 4 3506 3506 -79.3%
> 11/17 Sat 14 12127 7581 -55.3%
> 11/18 Sun 4 21281 16842 -0.7%
> 11/19 Mon 14 27513 15343 -9.6%
> 11/20 Tue 1 49240 25295 +49.1%
> 11/21 Wed 22 24718 16774 -1.1%
> ----------------------------------------------
> TOTAL 67 276172 169636 16963
> (estimated 10 runs per dumpcycle)
>
> Which is only microscopicly better. Snd the churn returns
Hmmm, yes, now things are getting messy....
However, I think much of the churn you saw in this run can be explained
based on the balance figures:
In your balance table from a day ago, the entry for today's run was:
11/11 Sun 1 10886 10886 -35.8%
... but note in the table after today's run that the 11/21 entry (which
normally would have included the same DLE(s) as the 11/11 did in
yesterday's table) is pretty close to a zero balance.
So I think basically the idea is that the planner noticed that the total
of full dumps scheduled for that last run was much less than the target
(average) daily size, and so it moved forward a whole bunch (i.e. 21)
smallish DLEs from 11/16, 11/17, and 11/19, thus pulling the runs total
up to the target size.
However, the catch is that it's really just borrowing DLEs from other
days that already were pretty undersized and now after the promotions
are left with negative balance figures, so the shuffling is not actually
solving the balance problem over the course of the full cycle....
(I don't know the details of how the planner makes these choices, but I
would assume that that the reason it doesn't promote DLEs from one of
the positive-balance days is that those DLEs are all so big that you
can't promote any of them without making today's run too large.)
We might still notice something interesting as we watch the balance
reports over the course of the dump cycle, but off hand I don't think of
any advice to give you, other than the "break up your huge DLEs" advice
you've already heard....
> NOTES:
> planner: Incremental of coyote:/var bumped to level 6.
> planner: Incremental of coyote:/home/gene/Mail bumped to level 2.
> planner: Incremental of coyote:/home/amanda bumped to level 2.
> planner: Full dump of coyote:/usr/local promoted from 6 days ahead.
> planner: Full dump of coyote:/root promoted from 6 days ahead.
> planner: Full dump of lathe:/etc promoted from 6 days ahead.
> planner: Full dump of shop:/etc promoted from 6 days ahead.
> planner: Full dump of GO704:/usr/local promoted from 6 days ahead.
> planner: Full dump of coyote:/usr/bin promoted from 6 days ahead.
> planner: Full dump of lathe:/home promoted from 6 days ahead.
> planner: Full dump of GO704:/home promoted from 8 days ahead.
> planner: Full dump of shop:/home promoted from 8 days ahead.
> planner: Full dump of picnc:/boot promoted from 8 days ahead.
> planner: Full dump of lathe:/usr/src promoted from 8 days ahead.
> planner: Full dump of GO704:/root promoted from 6 days ahead.
> planner: Full dump of shop:/lib/firmware promoted from 6 days ahead.
> planner: Full dump of lathe:/lib/firmware promoted from 6 days ahead.
> planner: Full dump of GO704:/lib/firmware promoted from 6 days ahead.
> planner: Full dump of shop:/root promoted from 6 days ahead.
> planner: Full dump of coyote:/usr/lib promoted from 5 days ahead.
> planner: Full dump of lathe:/root promoted from 6 days ahead.
> planner: Full dump of shop:/usr/lib/amanda promoted from 6 days ahead.
> planner: Full dump of GO704:/usr/lib/amanda promoted from 6 days ahead.
> planner: Full dump of coyote:/home/gene/Documents promoted from 5 days
> ahead.
> taper: tape Dailys-42 kb 18022402 fm 67 [OK
>
> But the rest of the report shows that coyote /var was actually a level 0.
> coyote /var 0 10971 10971 -- 9:07 20543.3 2:55 64197.3
Okay, that does seem odd.
Normally I would say something about how the "bumped to level N" lines
don't tell you that Amanda is actually going to do an incremental on
that run, but only tells you what level it would plann to do if it ends
up deciding to do an incremental for that DLE.
However, off hand I would have expected colyote:/var to be mentioned in
the "promoted from N days ahead" lines, but it's not there (and all 21
promoted DLEs are listed, so it seems something else is up).
So, if you want to do the deep dive to try to debug that quirk... the
next step will be to look through the
/var/log/amanda/Daily/amdump.20181112*
log file (assuming the run in question started after midnight this
morning; if not the datetime-stamp would be 20181111* of course).
Find the "ANALYZING ESTIMATES..." section, and cut-and-paste the log lines
that discuss coyote:/var from that section on down through INITIAL
SCHEDULE, PROMOTING DUMPS IF NEEDED, and on through the GENERATING
SCHEDULE section.... and we'll see if those lines tell us anything
useful...
(Also, "amadmin Dailys info coyote /var" might possibly tell us
something interesting as well.)
Nathan
----------------------------------------------------------------------------
Nathan Stratton Treadway - [email protected] - Mid-Atlantic region
Ray Ontko & Co. - Software consulting services - http://www.ontko.com/
GPG Key: http://www.ontko.com/~nathanst/gpg_key.txt ID: 1023D/ECFB6239
Key fingerprint = 6AD8 485E 20B9 5C71 231C 0C32 15F3 ADCD ECFB 6239
