some improvements/wishes

[Sancho]

Like it

Up

0

Down

Drop it

::

Hi all,
after several weeks of tests and one presentation, some improvements came out:

Would be possible to take a look to this presentation to improve ours ?

+ present alerts as a tree: when you have an alert on a module of an agent in a category, then this category appears red (different from others),
then one can click on it to deploy the tree and make Agents from this category appear, then the Agent containing the module which triggered the alert appears in red (or anything different from others) ,
then click on this Agent to deploy its alerts list.
That could be great to have this kind of alerts view for large networks, and maybe this could be implemented for agents view as well?

Nice idea, I like it very much, could you submit to our bugtracker/whishlist ?

+ It would be great and _really_ usefull to implement pre-defined AND user-defined templates for modules/agents/alerts that enables, when creating a new agent, to choose from a list of pre-defined modules and alerts (with some check boxes or drop-down lists). Then this could be used to generate the Agents conf files easily and graphically.

Yes!, and it’s a feature planned for 1.3 ! 🙂

+ The implementation of an RRD mechanism in the MySQL Pandora database to avoid keeping _every_ single value and processing them all each time we open a graph view.
It would be great to keep every received values for no more than 3 or 4 days, the data older than 3 or 4 days is calculated from 10 or more original values, then for the data older than 2 weeks, they are generated from 10 or more values calculated precendently.

This the “default” way to make things, it’s the method used by the majority of FLOSS solucions, like Zabbix or ZenOSS, in the first days of Pandora (when it’s name was other) this was also our approximation, but I can give you the more important reasons for not to use this:

More load, more problems to generate graphs, Pandora actually requires also a lot of CPU to generate graphs, but think about what it’s the percent of total available graph you are usually using, really do you want to generate periodically all possible graphs ?

Also, RRD requires a lot of space to store temporal graphs. Other reason for not to use RRD it’s that you never have “real” data, only data generated in a batch process. Batch process also introduces heavy load every X minutes, including for data that actually is never requested, so so need a really fast machine only to generate graphs.

I have experience generating more than 500+ graphs for individual pieces of information with RRD and it’s a f****g hell. It’s impossible to achieve high-performance results using this approach (I’m talking about have monitoring more than 10,000 elements), you would need to generate, supposing a 5min batch time, about 2000 graphs per minute (about 33 graphs per second). RRDGraph can’t work too fast.

Of course, if you want to generate “custom” graphs and reports, RRDtool it’s not valid since you have not a SQL backend to ask, your reporting needs always fallback to SQL 🙂

I hope this answer “WHY” Pandora is not using RRD.

+ Maybe the Pandora Server can be implemented in Java, as Java XML parsing is efficient and can avoid charset issues because Java can treat and convert many charsets.

Actually XML parsing it’s a problem in client side, Perl have also very powerful XML management libs based on libxml, that it’s the library we will use in Pandora 2.x branch. Clients have to create “manually” and this always have problems.

I cannot make Java Clients because I have many many many problems using java in the past for huge projects deployments because, for example, you have not Java in a IPSO, or probably Java VM in a AIX it’s too old for you, and of course, I hate to make systems calls from java/C code, for this I simply prefer to use shellscript or the tools system provide me.

+ maybe developp some Java console to have the choice between html/PHP and a heavy console app. Just a suggestion though.

This is a nice idea, our plan for make that is to improve, little by little our Pandora Applet to make in the future a graphical full-console in GTK (Cross plattform), but this far away ..

Thanks for your comments Dagget, it’s very important for the project to discuss about this….

[daggett]

Like it

Up

0

Down

Drop it

::

Ok, thanks, I did an add to the sourceforge “wishes” for the tree thing.

I also think this is important to discuss things like these, and I will continue to do it ! 😉

Thank you again for your answer.

Concerning the presentation I made, it was finally without any document, I just made the presentation with an open console, and some .conf examples.

I will do a prensentation with OO Impress, and send it to you once it’s done.

bye for now,
Denis

[Sancho]

Like it

Up

0

Down

Drop it

::

Ok, thanks, I did an add to the sourceforge “wishes” for the tree thing.

I also think this is important to discuss things like these, and I will continue to do it ! 😉

Thank you again for your answer.

Concerning the presentation I made, it was finally without any document, I just made the presentation with an open console, and some .conf examples.

I will do a prensentation with OO Impress, and send it to you once it’s done.

bye for now,
Denis

Will be nice to share this stuff too, we want to create a product’s whitepaper, similar to this (another FLOSS solution we use in our company):

http://www.artica.es/docs/yixue_whitepaper_en.pdf

So ideas will be welcomed !