[Orca-users] Wierd "bug" in HTML and PNG name generation

[Cameron Paine]

[This is a repost. I waited 48 hours. Apologies if you've
already received it. --cbp]

Maybe someone has seen this before. It's got me stumped.

As you probably know, plot files (and the HTML files that
host the time series of plots for a given group/subgroup)
have names of the form:

g_sg_dt_s1,__s2, ... __sN-pt.*

where,
g == group,
sg == subgroup (hostname in my case),
dt == data type (guage, counter, etc),
s1, s2 ... sN == data source names (column headings),
pt == plot type (daily, weekly, etc),
* == html or png.

Sorry about the crappy notation but I didn't want your
MUA to munge my description. I'd post the real names but
they're over 160 characters in length.

I have 30 hosts, all of which generate the same format
source file. There are some differences in the exact
composition of the "columns" but AFAICS they're
essentially the same. They're all in the same group.

One of my plot definitions has four data sources. In fact,
it's almost a copy of the CPU Usage plot in the distribution
orcallator.cfg. For most of the hosts (subgroups) this
definition results in file names of the form:

g_sg_dt_s1,__s2,__s3,__s4-pt.*

However two of the hosts (subgroups) result in file names
of the form:

s4-pt.*

i.e. the leading g_sg_dt_s1,__s2,__s3,__ is missing.

One side-effect of this is that the HTML file name is no
longer unique (the PNG files get written to a subgroup
directory whereas the HTML files go in the top-level
directory) and so the second one overwrites the first.
This is what initially drew my attention to this problem.

Let's call that symptom 1.

Symptom 2 is that the plot for these two hosts is AFU. All
four data sources are represented but they're plotted in the
wrong order. As near as I can tell, the incorrect plots
reorder the data sources as: 4, 1, 2, 3. It could also be
4, 2, 1, 3 but that seems unlikely. The legends and colours
are the same on *all* plots.

What else can I tell you? The RRD files are named correctly
(and identically) for both the correct and incorrect hosts.

All 30 hosts have these four data sources and they all appear
in the same order in the source files.

The generated names are all shorter than the maximum size so
the md5 hash work-around is not invoked.

I can't think of anything else that might be relevant.

I've looked at the code where (I think) the names are generated
and I can't see anything that even hints at why these two hosts
are processed differently. So I'm throwing this at the list to
see if anyone else has observed similar behaviour.

Thanks for reading and for any clues you might be able to offer.

Cameron Paine