Serious slowness with rrdgraph from rrdtool / 2018-10-12

One of the things still needing migrating is the NTP server stats which obviously uses rrdtool. Because I want to keep the history I migrated the datasets with:

/usr/local/rrdtool/bin/rrdtool dump ntpvals-stardate.cs.uu.nl.rrd \
| ssh newhost /usr/bin/rrdtool restore -f - ntpvals-stardate.cs.uu.nl.rrd

And then create a graph of the plloffset for example using:

/usr/bin/rrdtool graph /tmp/plloffset-stardate.cs.uu.nl-24hours.png \
--title "stardate.cs.uu.nl pll offset (last 24 hours)" --imginfo \
'<img src="tmpgraphs/%s" WIDTH="%lu" HEIGHT="%lu" alt="Graph">' \
--start -24hours --end now --vertical-label="Seconds" --color BACK#0000FF \
--color CANVAS#c0e5ff --color FONT#ffffff --color GRID#ffffff \
--color MGRID#ffffff --alt-autoscale --imgformat PNG --lazy \
DEF:offset=ntpvals-stardate.cs.uu.nl.rrd:plloffset:AVERAGE \
CDEF:wipeout=offset,UN,INF,UNKN,IF CDEF:wipeoutn=wipeout,-1,* \
LINE1:offset#000000:"Offset\:" \
GPRINT:offset:LAST:"Current\:%.3lf%s" \
GPRINT:offset:MIN:"Min\:%.3lf%S" \
GPRINT:offset:MAX:"Max\:%.3lf%S" \
GPRINT:offset:AVERAGE:"Average\:%.3lf%S" \
AREA:wipeout#e0e0e0 AREA:wipeoutn#e0e0e0

But on the old server this takes 0.026 seconds, on the new server 3 minutes and 47.46 seconds. No idea what is happening, strace shows nothing strange and rrdtool uses 1 cpu at 100% all that time.

Update 2018-10-13: Searching through the options in use found that the difference between 0.039 seconds and 3 minutes 56.473 seconds is the --alt-autoscale parameter. Enabling it makes rrdgraph very slow.