Incorrect "Excess TblRSz" and "Excess IdxRSz"
I'm not sure if this is a bug or a roadmap item. For now I'm calling it a bug.
With "proserve sports" I get the following in the Configuration Viewer:
# of Tables: 9 Excess TblRSz: 40 # of Indexes: 32 Excess IdxRSz: 17 # of LOBs: 0
With "proserve sports -basetable -100 -baseindex -100", I get this:
# of Tables: 9 Excess TblRSz: 141 # of Indexes: 32 Excess IdxRSz: 118
ProTop says it got better , when it actually got worse. BTW, I should note that PT does display correct values for the params (BT/TRS/BI/IRS).
I know I'm odd for using negative basetable and baseindex values, but it is a valid use case.
Here is another example; somewhat contrived, but it demonstrates a point:
"proserve sports -basetable 100 -baseindex 100"
# of Tables: 9 Excess TblRSz: -59 # of Indexes: 32 Excess IdxRSz: -82
I don't know how many people use BT or BI greater than one, so this may not matter to you, but again the numbers don't make sense to me.
One other small thing, and this is a more reasonable scenario. If a negative "excess" is meant to indicate the amount by which IRS or TRS is too low (given reasonable values for BT and BI), it's off by one. In sports, the highest object numbers are table 9 and index 32. If I proserve with "-tablerangesize 7 -indexrangesize 29", then TRS is actually too low by 2 and IRS by 3. I get this in PT:
# of Tables: 9 Excess TblRSz: -3 # of Indexes: 32 Excess IdxRSz: -4
What follows is more of a roadmap idea. It would be nice if there was a way to have PT give recommended values for the stats params, given the schema. It's hard to do that justice in the fairly confined space of the Configuration Viewer, and anyway it's not something that would really belong there as it isn't part of the current configuration. Maybe it could appear as an overlay frame, like "Sample Type" ("R") does today.
Here is some food for thought from the output of my health-check code:
__________________________________________________________________________________________________________________________________ Database objects and VST statistics __________________________________________________________________________________________________________________________________ # of Objects Statistics --------------- ------------------------------ Total Sys Appl ----- ---- ---- # of storage objects: 237 Low # High # Excess Tables : 136 127 9 # of VSTs : 52 ------ ------ ------ Indexes: 201 176 25 -basetable : 1 Tables : VST range : 1 7 BLOBs : 3 3 -baseindex : 1 Appl. range: 1 9 0 CLOBs : 1 1 -tablerangesize : 7 Indexes: VST range : 1 29 Total : 341 307 34 -indexrangesize : 29 Appl. range: 8 32 0 WARNING: missing CRUD stats for 2 application tables. Adjust -tablerangesize to at least 9. WARNING: missing CRUD stats for 3 application indexes. Adjust -indexrangesize to at least 32.
I prefer to look at it as "a difference of opinion". But your opinion is kind of persuasive.
(FWIW I suspect that the number of people using a non-default -basetable is pretty low.)
In the interest of simplicity and complete coverage I'm thinking I should stop trying to be fancy and take a "worst case" approach and just use the total number of _file and _index records.
What do you think of that?
Or maybe take the total that is >= the -base* parameter
I think the stat that will help most people and be simplest to report is: how many application tables and indexes do not have stats coverage. That could be because they are below base* or because they are beyond the rangesize.
Tables I can do that. Indexes are not clean. So from a consistency perspective I was thinking that it makes more sense to just count all of each.
For one thing it's a whole lot easier to explain. And there is a lot to be said for that.
Yes, that will result in about three hundred more objects than they really absolutely need -- but in the big scheme of things is that really a problem?
(A certain famous application has a configurator tool that applies the same -tablerangesize to every database in its scope, which results in thousands of excess *rangesize being allocated because other than the main db none of them need more than a handful...)
I do like the idea of an optional more detailed report too.
Rob Fitzpatrick last edited by Rob Fitzpatrick
Just curious, why are indexes not "clean"? I mean, apart from the unindexed query it might require to look for application indexes. But you're already caching them in a temp-table so it shouldn't be an issue.
Certainly it's the case that some people are interested in their object counts as their schemas evolve. Mine grow from release to release, so I like to know how many tables and indexes I have. But I think object count, while interesting in and of itself, is insufficient for judging whether you have all the stats you need.
When I was first learning about the metaschema and discovered the stats tables and params, my first stab at sizing them was by counting tables and indexes. But I quickly realized I was missing stats records. Then I discovered that indexes start at 8, not 1. I also discovered, primarily in internal (dev/qa/test) environments with lots of schema changes, that there could be gaps in the numbers. So a database with 100 tables and 200 indexes could have a highest table number above 100 and a highest index number above 200. Indexes especially can have gaps due to the temp indexes sometimes created in incremental schema files.
So I settled on my current process of finding the lowest and highest object numbers, calculating the start and end numbers of the stats "window", and determining if any of the application objects are outside that window, i.e. there are no stats kept for them.
Yes, that's a lot to explain and it may be over a newbie's head. It would have been for me, initially. But in my opinion, it's one of the things a newbie needs to learn to rise above being a newbie.
Of course PT reports a lot more than CRUD stats. But they are an important part of what it does, and an important reason why a person goes to the trouble of using it at all, as opposed to just running promon which is much easier and more straightforward. Using it to look at CRUD could be frustrating if the user learns, after the fact, that they weren't seeing the whole picture. You use highlighting to call out badly-set params in other parts of PT. I think it would be useful for users if it were somehow obvious that they aren't able to see CRUD stats for all of their application objects.
Indexes aren't "clean" because "system" indexes and application indexes are all mixed together. You don't have segregation by negative idx-num like you do for meta-schema tables. (There are some negatively numbered system indexes.)
It is relatively easy to say that if you want all application tables you need to set the base to X and the range to Y. And there won't be anything else in there.
With indexes you will be including things that are not "application".
That's why I don't see them as "clean".
Ok, that makes sense. But that's just life in OE; there's nothing you can do about it.
Well, having written that, there is something you could do about it if it really mattered a lot. Your results go into temp tables before you display them, so you could filter out index records where the index name begins "_". For the record, I'm not suggesting you do that. Seems like a lot of work for not much gain.
On a side note, I'd really like to see Progress renumber their indexes. It shouldn't break anything; no one should be taking dependencies on indexes having certain numbers (though I wouldn't be shocked if they did). Release 12.0 would be a good place to do that type of thing. They would have to rebuild their sample and empty DBs but it would be a win for customers.
Rob Fitzpatrick last edited by Rob Fitzpatrick
How about something like:
excessTbl = (( tblRange - maxTblNum ) - abs( _TableBase )) excessIdx = (( idxRange - maxIdxNum ) - abs( _IndexBase ))
I think that works for positive and negative -base*. At least it seemed to make sense when I tried a few test cases.
- maxTblNum is the maximum table number less than 32768
- maxIdxNum is the maximum index number under 993 if there are no application indexes > 993
That should avoid over-allocating for people who do not deliberately want to monitor system stuff.
Offhand I'd say no.
lowest table #: 100
highest table #: 490
actual excess: 9
calculated excess: (TRS - maxTbl#) - abs( BT ) = (400 - 490) - 100 = -190
I think the values to be compared are:
- highest table number ("HT")
- highest stats table ("HST")
The latter is my arbitrary name for the highest table # for which stats will be collected, given BT and TRS. It is equal to TRS + BT - 1.
E.g., it gives the correct answers in these cases:
BT 1, TRS 50: HST = 50 + 1 - 1 = 50
BT -500, TRS 505: HST = 505 + (-500) - 1 = 4
BT 100, TRS 400: HST = 400 + 100 - 1 = 499
excess = HST - HT
excess = (TRS + BT - 1) - HT
Does that make sense?
Separately, it's also worth comparing BT (basetable) to LT (lowest table #). If BT > LT then BT is too high. (It also means the currently-calculated "excess" may be inaccurate.)
BT is not likely to be set wrong, if set at all, but it's an easy check to do and it will keep monkeys like me out of trouble.
We may have conflicting goals. I am primarily trying to ensure that ProTop users can always monitor their application tables. So a base value > 1 isn't something that I want to encourage or support.
The over-arching goal of the "excess" metrics is to make sure that Joe Ordinary User has an indication that application data is not being adequately monitored.
Optionally I do sort of want to ensure that when certain persons who won't be named but whose initials might be "Rob Fitzpatrick" set the base less than zero or otherwise muck around doing fun stuff that it isn't needlessly wrong. But some of these use cases don't seem like something that I can support without losing the usefulness of the metric for the types of people that are the primary target.
As you suggested, I think the ultimate solution will have to be to go down the path of showing a bunch of different scenarios such as "monitor application data", "monitor meta schema", "monitor everything" etc and show the combination of values needed for those scenarios.
I agree I'm an "outlier" and I'm really not trying to tailor PT to my specific needs. You've been very accommodating and I'm not trying to push that.
That said, I like the calculation I provided (the excess one, not the basetable one) as it works, as far as I am aware, in all cases whether BT is 1 or not. So I think it is a reasonable way to calculate excess stats slots for Joe Ordinary.
BT 100, TRS 400: HST = 400 + 100 - 1 = 499
This, I think, (hope) is the last bit giving me trouble.
If you were monitoring sports2000 with these settings, using these calculations ProTop would indicate that you have an "excess" of 499 table "slots". But in reality you are monitoring no table stats at all.
So perhaps this leads to a different thought -- I am really trying to illustrate "unmonitored (application) tables" (or indexes) and "excess" is probably the wrong way to communicate that.
It has been a long and winding road to get there -- but I think that is really what I need to recognize.
So, keeping our friend Joe Ordinary in mind, maybe two things are relevant:
- do I have unmonitored objects?
- if I don't today, how close am I to being there if my schema grows?
┌────────────────────────── Excess Range Calculation ───────────────────────────┐ │ │ │ _tableBase: -500 _indexBase: 1 │ │ -tablerangesize: 600 -indexrangesize: 50 │ │ Highest Stats Table#: 99 Highest Stats Index#: 50 │ │ │ │ Actual Number of App Tables: 25 Actual Number of App Indexes: 55 │ │ Minimum App Table#: 1 Minimum App Index#: 8 │ │ Maximum App Table#: 25 Maximum App Index#: 62 │ │ Unmonitored App Tables: 0 Unmonitored App Indexes: 12 │ │ │ │ Lowest Monitored Table#: -361 Lowest Monitored Index#: 1 │ │ Highest Monitored Table#: 25 Highest Monitored Index#: 50 │ │ │ │ Excess Table Range: 74 Excess Index Range: -13 │ │ │ └───────────────────────────────────────────────────────────────────────────────┘
My "scratch" numbers... soon to be a hot key or something along those lines
Cool. Very comprehensive.
One bit of feedback: Joe is probably more familiar with the param names, -basetable and -baseindex, than with their VST field names.
One more: looks like "excess index range" is off by one.