Manual/Auto Re-Indexing cause data sources to loose their indexes

[HorridFlyer]

I'm running 1.2.11, after upgrading from 0.8.8b. The old version is still running on the production server, while I'm testing the new one on a VM.
They both use Spine as the poller.

While the old version is running as smooth as silk, 1.2.11 is having a lot of problems, and here is the last one:
if I run a manual re-indexing of the "SNMP Interface statistic" data-query associated to my given device, its data-sources got messed-up and I start getting "you have 1 Device with bad SNMP Indexes" in the logs. To solve this, for each affected data-source I have to manually switch from IfDescr to IfName (or IfIndex) in the "Index Type" dropdown menu.
This also happens after an auto re-indexing, e.g. after my network element reboots (as my "Re-Index Method" is set to uptime, which changes after a reboot).
Once I switched the data-sources from IfDescr to ifName, they no longer have troubles after a manual/auto reindex.

Is this a bug? Or does is it only happen with imported devices/data-sources from older versions?
Anyway, is there a way to quickly fix those broken data-dources? I wasn't able to find a way to do it directly into the DB

[TheWitness]

ifName was adopted over ifDescr as it has been reported that some ifDescr's change after firmware updates for some device types, so that can not be used a primary keys. The ifName is much more stable.

When Cacti finds an interface that has disappeared is will remove it from the poller cache and remove the index. However it's not forgotten.

This is a new behavior, but we are considering a way that you can continue to attempt to track the data source even though the sort field value has gone away for some types of indexes that come and go.

I have to say that the feature is Handy when you have thousands of graphs and want to prune the dozens that are orphaned.

[HorridFlyer]

Ok I understand.
So do I have to manually switch each DS to ifName untill a patch will be released? No way to fast-fix them operating directly into the DB?

[TheWitness]

So, you are saying that they are stuck at ifDescr now? Some of them anyway?

[TheWitness]

I've seen this happen, but it only happened when there was not an ifName match.

[HorridFlyer]

Yes they're stuck with ifDescr untill I manually switch to ifName. And in the meantime the graphs don't populate.

If i click on the affected DS i see Index Type is still on ifDescr, but the interface description shows up regularly and it is the same as ever: a machine reboot doesn't cause interface descriptions to change! Nor they're affected by the manual re-indexing of-course (I click on the green arrows to reload the SNMP - Interface Statistics data query). Please don't forget the same DS/graphs are still running smoothly on 0.8.8b, they never stop working in both scenarios, and this is the proof nothing has changed in the interface description.
So if 1.2.11 has a reason to behave this way, I really cannot understand it. Anyway I was trying to find out a way to quickly switch all ifDescr to ifName before the graphs get screwed up, but still didn't find it.

[TheWitness]

Just modify the interfaces.xml to have ifDescr first, and everything will switch back on the next reindex. Not the best option for everyone, but maybe for you.

[HorridFlyer]

Changing the order between ifName and in ifDescr in the interface.xml file didn't solve this issue: after the inversion the order between 0.8.8b and 1.2.22 is the same, but only in 1.2.11 data-sources looses their indexes after a data-query reload.
Here I have to admit not every DS get screwed up, only those where the description on the network device effectively differs from the one Cacti picked up, e.g.:

DS ifDescr = "test port - not yet activated"
network device port description = "test port"

In such a case I still have to manually switch the DS to ifName.

[TheWitness]

Yes, we can do a setting to prevent this changes in the future. For now, just hang in there