OK, to close this off (in the absence of any way to override the behaviour of svcstatus.cgi), I made sure that I had exactly one RRD file (everything up to the first comma in the filename, or the .rrd at the end) to match the entry in GRAPHS, and all of the broken graphs disappeared. I'm happy now.
On 27 March 2017 at 12:35, Jeremy Laidman <jlaidman at rebel-it.com.au> wrote:
On 27 March 2017 at 12:16, Jeremy Laidman <jlaidman at rebel-it.com.au> wrote:
It's weird how the exact same graph definition attempts to display one too many graphs for one server, and two too many graphs for another. It's a mystery how svcstatus.cgi determines how many graphs to show.
To answer my own questions, actually, it's fairly simple, and is based on the number of RRD files matching the testname. The test name that's broken for me is "bindstats". I have a file called "bindstats.rrd", but I also have a bunch of other files with names of the form "bindstats,<string>.rrd" (starting with "bindstats" and followed by a comma and some other text). In all, there are 14 files matching in this day, for one server, and 11 for another. It seems that svcstatus.cgi has a maximum of 6 lines per graph, so for 11 files, it thinks there are two graphs to create (6+5), and for 14 files, it thinks there are three graphs (5+5+4).
I think that my mistake was making different files, for different graphs, but giving them essentially the same names (with different suffixes separated by comma). If I rename the graphs to be completely different, then svcstatus.cgi will only see one file, and won't try to create graphs for non-existent lines.
J