Analysis Recalculation produces Out of Order Events, which means they are not compressed.
The reason for this behavior is, that there is a more recent snapshot value existing. This also happens, when Recalculation is performed till "Now" (*), because the Analysis Service performs the Recalculation in chunks (see "Will PI Analysis Recalculation follow compression and exception?")

Would it be possible to make the Analysis Service first delete the time range fully (when recalculating up to "Now") and let it perform the recalculation in chunks afterward?
This would not produce OOO Events and compression can be applied.

The only workaround at the moment is to delete values manually (with Powershell for example).

The PI data archive (and potentially other systems) can generate many automated error messages to assist with troubleshooting. One such key example is "[-10767] Client exceeded maximum concurrent queries in RPC thread pool", which occurs when a service (such as AF Analyses) doesn't have adequate threads to retrieve input values.

If this error is given because of a system issue, the Asset Analysis service will still process the logic with that error message being used a generic string. If the logic is something similar to Input <>1, then this would resolve as TRUE. In this scenario, this would be a false positive result for the customer.

Because these error messages come from the Data Archive, this scenario could affect a customer's entire set of analyses.

Can we include functionality to flag, filter, or reject input variables that are PI system generated error messages? This doesn't resolve the issue of data quality, but it cuts down on inaccuracy coming from the central system itself.

If I want to create an analysis that generates an event frame in PI System Explorer, often times it's useful to check the values of variables to make sure that a calculation is being done as expected to mitigate error on my end, and make sure I am doing the calculation properly and have correct syntax for the order of operations (as an example). This can be done with expression types, which makes it easy to see the values of all the attributes and variables used in the analysis for a given time. There may also be a threshold that I may have to find by using the "guess and check" method, which is impossible with an event frame type. Ideally, the "Preview Results" window has a toggle switch that would still allow me to see event frames as they stand right now (start time, end time, name of event, etc), and I could just switch over to a preview results window that is similar to an expression type analysis.

As a consumer of event frame data, I would like to be able to get a summary of the children event frame duration based on the trigger or severity over the duration of parent event frame when a parent event frame closes. The output should be stored where it can be trended and analyzed.

For example, over the duration of the parent event frame there may be several sets of the children event frames triggered by different complex start triggers (and having different severity) - i.e. starttrigger1, starttrigger2, strarttrigger3, starttrigger2, strarttrigger3, starttrigger1, etc - as a consumer of data I would like to see on the parent event frame the duration of all child event frame with severity “Warning” or the duration of all child event frame triggered by start trigger 2.

I understand, that partially it can be done using output at close option with functions like TimeEQ or TimeGT, but this is applicable only if the triggers are simple (using single tag) and not complex (several attributes with different conditions, for example ‘attr1’=0 AND ‘attr2’>100).

Another consideration is to have this information specifically on the parent level of the event frames only, which is not possible with current functionality where the same event frame template is used for both parent and children.

The Analysis server uses the PI Buffer subsystem in order to apply Exception and Compression testing on the results of Analysis. However, when deleting data (either through Recalculation or deleting data via Powershell or PI SMT, or whatever) the PI Buffer subsystem on the Analysis server never resets the stored snapshot value/timestamp.

The consequence of this is that if the same Analysis is recalculated multiple times (for example after corrections, optimization, etc), compression is only ever applied the very first time an Analysis is recalculated. After that, compression is never applied up to the timestamp that the recalculation was run up to, as the PI Buffer subsystem never resets the actual current snapshot value/time.

In the referenced thread, the only workaround is to completely stop the PI Buffering Subsystem on the Analysis Server, delete the cache, release ownership of the PI Points on the Data Archive, restart the Buffering Subsystem, and then hope that everything works.

It would be way easier to be able to tell the Buffering Subsystem to get the current Snapshot from the Data Archive.

Reference in this thread on PI Square:
https://pisquare.osisoft.com/thread/43696-analysis-recalculation-ignores-pi-point-compression-settings#comment-146126

KB00868 recommends Asset Analytics as the right job, but there are some major limitations that inhibit ease of use and deployment, specifically around operations on >1 attribute or tag simultaneously.

To improve this, AF analyses functions that take in data of type "attribute attname" should be able to process an expression that returns type attribute. Perhaps limited to periodic analyses other toggleable somehow, as there is potential to use a lot of CPU cycles.

== Example 1: Totalized Mass Flow ==
Given FIT-001 and DIT-002, return the totalized mass flow (say, kg) over an interval. The current process:
1. Create short-interval periodic, or event-based analyses to write the product of flow & density to a tag-linked attribute WIT-003 (I don't think you can use an attribute as we care about the history).
2. TagTot(WIT-003, ''. '-1w')
Better solution:
AF should be able to see that I'm asking for a 1-week total of an expression, so run the expression and totalize the result.
TagTot((FIT-001 * DIT-002), ''. '-1w')
This is basically what the SUMPRODUCT function in Excel does, and we haven't found a better way to do it in AF yet.

== 2. Conditional Expressions ==
We monitor processes and create KPIs for utilization and value delivered. PIAF makes it cumbersome to compute utilization or value based on multiple conditions. Utilization of an application APP-001 might depend on x=1, y=2, z=3.
Current process:
1. Create short-interval periodic, or event-based analyses to write the result of ('x' = 1) AND ('y' = 2) AND ('z' = 3) to a tag-based attribute, MYTAG-001.
2. Create an analysis to run TimeEq('MYTAG-001', '', '-1w', 'true')
Better solution:
AF should be able to process something like this:
TimeEq((('x' = 1) AND ('y' = 2) AND ('z' = 3)), '', '-1w', 'true')

The current process is inefficient and a bad use of tags (both a waste of a tag license and adding a low-value intermediary calculated result to the data archive). There are also complexities around scheduling the preliminary analytics to ensure the results are ready for the insight-generating analytics.

Thanks for your feedback. I'd also be overjoyed to hear there's an easy solution we're overlooking.

If you have an analysis input that is the "string" data type but can be represented as a number (for example, the string "2.35"), then to convert this string into a numeric data type such as double it is necessary to take the string and output it to another attribute that has the correct data type. There are no usable data type conversion functions in AF Analytics, so it is not possible to take the string, convert it into a double, and then store it in an analysis variable to be used again in the same analysis. It is possible to take a string input and output it to an attribute of the correct data type to make the built-in automatic conversion happen, but this means users will end up with an unnecessary attribute every time they need to do a conversion.

It would be better if there was a way to convert data types (at least strings to numbers) directly inside of AF Analysis Expressions without having to rely on the automatic conversion that happens when you output the result to an attribute of the desired data type.

By now the NumOfChanges Function in AF Analytics does not take into account the value at the start of the timerange. NumOfChanges only counts the amount of changed values of the recorded values in the given timeframe.
If I query the NumOfChanges for today 12 am till now and my Archived Values are:
Val1, 11:59 pm (day before)
Val 2, 4 am
Val 2, 12 pm
Val 2, 8 pm

the function will result in 0 although in the time from 12 am to 4 am there was Val1.
Regarding TechSupport this is not a bug in the functionality, only a bad description of the functionality of NumOfChanges.

For now I built a workaround:
If TagVal(Tag,Starttime) <> NextVal(Tag,Starttime) THEN NumOfChanges + 1 ELSE NumOfChanges

But I think such a missleading function should be changed.

It would be nice if we could have a "real-time" and "recalculate" part in an analysis. As source data could have changed you probably want sometimes the analysis to act in another way. Example: 2 attributes are used to trigger an event-triggered analysis and assume they write both a value for the same time stamp. in real-time this would mean that the analysis is triggered only once (1 trigger possible per time stamp) and is missing the value of one of the 2 attributes (depends who writes first). but with a recalculation both values are present and because of that the result can be different. so it would be nice if we can anticipate on this by allowing a different analysis setup taking the differences into account