weird issue after virtualbox to vmware migration

[robotfish1911]

I've been running Cacti for a couple of weeks now, and obtained some decent baselines. After determining that I was going to stick to Cacti, I moved the installation from Virtualbox running on a laptop onto a server blade running vmware.

I don't think the migration from one virtual platform to another is really the issue, but what I did find was that once I fired up my server on the vmware box, the poller would run fine and not report any errors, but my graphs were not continuing to populate.

At first, I thought it was an issue I could solve by doing a "Rebuild poller cache". It didn't seem to help, so I figured I was doing something wrong. I then found the script that The Witness wrote, to do a poller cache rebuild from a command line. I ran it from the location documented (put the php file in my htdocs/cacti folder) from a command prompt and didn't get any output.

After derping around further, I decided I was going to go into my htdocs/cacti/rra directory and leave the numbered folders corresponding to my data sources there, but back-up and remove the *.rrd files and let rrdtool rebuild them.

That worked out perfectly; after letting the polling service run for 15 minutes or so graphs started to reappear and start populating properly.

I then decided I would find a data source I knew was working, take the working .rrd file out, and put one of my backups back in. The old data from the past couple of weeks was there, but again, graphing did not start working.. I just ended up with a huge gap that grew, as the time in the timeline progressed.

At this point, I fired up Procmon to check for file permission issues (possibly caused by the migration) and found absolutely none. When I filtered it down by process and by directory, I did see that rrdtool.exe was opening the backup file, but not saving to it. I also noted that in windows explorer, the timestamp was not incrementing, which is indicative that rrdtool is not updating the .rrd files at all, once they have been removed and re-added to the directory they came out of.

Does anyone have any ideas as to why simply removing the .rrd file from it's location in the rra directory of Apache, and readding it might make it so that the graph is no longer updated?

Throughout the process, I bounced the server serveral times.

If anyone has any ideas, I'd be grateful. I can live with having to start baselining again, but I'm sure there's something I'm missing.

It's worth mentioning that it's running on Server 2008R2, and I used the unattended windows installer that BSOD put together (apache, mysql, dependencies, etc).

[phalek]

Just to clarify: You did migrate the whole virtual host and did NOT just move the cacti directories/files around ?

[robotfish1911]

I apologize if I was being a bit ambiguous there. I migrated the entire machine bit for bit using vCenter converter. I didn't copy and paste the directory structure.

This behavior is very confusing. When I get back in the office on Monday I'm going to go through all the troubleshooting steps again from the top. I was pretty burnt out on Friday and I may have missed something.

[BSOD2600]

1. Did you migrate between 32 and 64bit os per chance?
2. rrdtool info <filename.rrd> show any obvious structural differences between the old and new rrd files (take one from a simple data template like cpu or interface).
3. Could crank up the cacti logging level to high, catch what the rrdtool update command thats being used (for the good rrd file), and then attempt to run that same command against the old rrd file at the command prompt. see if rrdtool spits out any useful error messages.