I would like to get your input on your desired behavior for Manual Recalculation for Analytics.
You can only comment after a vote. I voted "No" as i don't see any good reason yet. Manual recalculation is posisble at the moment, just remove the data and do a backfill. Any more advanced logic might need custom tooling anyway. So this seems as a lot of work (on any larger critical production environment) for little gain. Any insight in why this would be requried would be handy to evaluate the reasoning behind.
As for automatic recalculation, now that is a more enticing prospect where i would more likely vote a yes.
(of course trying to vote for the companies i work for.. don't have issues upgrading my Dev environment)
An upgrade on the PI Data Archive side here is to better enable the PI Analysis Service recalculation. In pre-AF Client 2.8 and pre-3.4.400 servers, to delete or replace values, the client first has to get all the data. AF Client 2.8, PI Bufss, and a new PI Data Archive will support a server RPC called by ReplaceValues that can perform replace and/or delete in one logical server call regardless of the existing events. Good thing is there are some good use cases for ReplaceValues outside of analysis too!
Sorry it appears I neglected to answer Roger. Sorry Roger :-).
There are complications with replacing values in the PI Data Archive, some of which Barry has already spoken to. In addition, if you have a PI Collective, you need to be concerned that the outputs from PI Analysis Service is synchronized between members of the collective during recalculation. We needed a way to ensure that we can be absolutely sure that recalculated values are exactly the same between the collective members and that all previous outputs are removed/replaced. The worst thing that can happen is if you have a mix of previous output values and new recalculated output values and no ability to discern old from new. The new features we put into PI Server 2016 (soon to be released, I promise :-)) ensure that we can do that (at least in theory).
Finally back on Pi Square. Ah, understand, like distributed transactions.
Now i have another good reason why i try to avoid collectives when you can do without them. ;-)
I voted yes, an upgrade is an upgrade, hardly takes any time these days.
Hopefully the upgrade fixes the issue with removing lots of data at once.
How much is "lots"?
I asume that analytics means AF Analytics. Is there a difference between automatical and manual recalc?
Willing to upgrade and being able to do it are two different sites of the same medal. I´m working in pharmaceutical Industries and due to regulatory requirements I need to validate every update - it is an huge effort especially because there come more functions then "only" the recalc. So my testing efford is huge. In fact I can do an upgrade every five to six years - in between just in case of mayor bugs.
So new features are great but until I´m able to use them a lot of time passes by. Nevertheless I appreach new features!
Thanks for your comment. I understand that in certain industries it's not so easy to upgrade. However, sometimes it's necessary as it's hard to add new features without having a new release :-).
We understand that with newer version of PI Data Archive there can be many new features which can be useful. In collective PI scenario, we want analyses to be in collective mode also as mentioned by Stephen in reply to Roger's comments and we can upgrade our dev environment, but unfortunately not production ones.
I have to agree with flost that it's not quite a yearly upgrade that client approves unless there is major bug detected and resolved, and the decision making time has also been long for companies in Oil & Gas sector I worked with.
You people have already thought about it but just suggesting that can't this feature be integrated in patch, so clients' with one older version can also utilize this.
Is a patch considered differently than an upgrade? Either way you're installing software resulting in a new version of the software. I'm curious how you differentiate between these two.
In both cases clients see the impact of the new installable on their current system.
Patch upgrade generally considered as small software update, in which after deciding impact and consequences with stakeholders, can be installed (in this case impact regarding on analyses services).
In case of whole new PI Data Archive upgrade, the concerns are bigger as the stability of whole system put in to consideration and decision making procedure itself for this likely to take quite long time.
After reading all of the preceding comments, I voted yes.
We have a significant number of PI servers, many of which are in HA environments.
For the AF analyses we have enabled, there was often a period of time, upwards of a month or two, when the calculations were running, but need to be 'tweaked'.
The data was deleted from the archive, and the calculations reprocessed to backfill... it was painful.
While we may not upgrade every PI Archive System as fast as I'd like, having a need may put the upgrade higher on the back log.
Retrieving data ...