|
I'm searching for good, general literature* on how to design and implement access control management. I'd like to build up the knowledge to design a system myself and/or make informed decisions about the architecture of one.
My focus is authorization, not authentication.
*literature = online,offline,free,paid; anything goes.
Some concrete questions could be
- "in which roughness should I store permissions?"
- "in which format should I store my permissions?"
- "should I use RBAC, UBAC, ACL,... and how would I concretely implement them, cleanly?"
- "How do I combine all that with a system that contains several individual organizations with their own set of permissions?"
Does anyone have recommendations? Even a "look in this corner of the internet would help".
(I really cannot find anything useful about this topic. All search results contain the same hipster stuff)
Just to make sure ... what I'm not looking for:
- A cook book.
- Answers that just say "use this or that, because it's the standard"
- Some "hip" book after which you nod, say "yes, aha, I see", but didn't really make you smarter.
|
|
|
|
|
|
Do you have any recommendations of the many?
Also: Be aware that google is very different depending on the country you are searching from and the locale on your computer. You might get better results than linkedIn and companies trying to sell their products - which is what I get.
|
|
|
|
|
D4rkTrick wrote: Do you have any recommendations of the many?
You need to review what you find and decide for yourself which best meets your requirements.
|
|
|
|
|
D4rkTrick wrote: "in which roughness should I store permissions?"
Having considered that many times what I came up with is that you cannot answer that question without business requirements. And probably some business domain knowledge yourself.
Consider these cases...
Who should have permission to edit a social security number at a company? Who should be able to view it?
Contrast that with who should have permission to change the delivery address? Who should be able to view it? (Even more prickly can someone change a delivery address on an order that has already been delivered?)
Should there be a admin (root admin) that can change everything?
Are there assets that will be shared between companies? If so then who sets the permissions? (That was a very real business case that I had to deal with.)
Answering questions like that is FAR more important than how you implement it in general. But it will certainly drive implementation details.
I would be very wary of suggestions that you should 'plan for the future'. For instance in one of many parallel futures you might be asked to add a field on one form, so obviously the solution is that every single field on every form must have a permission from the beginning. And then the users (not developers) who will be actually setting that up will need to understand hundreds of permissions and implications (like what if the social security number is on two different forms?)
I worked for a company with hundreds of permissions. No one knew what they all did. There was also disagreement about what they should do. So those tasked with changing them just had to experiment.
D4rkTrick wrote: "should I use RBAC, UBAC, ACL,... and how would I concretely implement them, cleanly?"
I doubt 'cleanly' exists in any medium size application. And really unlikely when it gets bigger.
So just pick one and look at it in detail.
And then really study how to log changes that users make. All users. Log in this case means to a persistent data store. Adds and updates are easy. Deletes are much harder.
|
|
|
|
|
I didn't mean to try to find the answers here on the board, sorry for the misunderstanding. The questions were merely examples. I'm aware that "clean" or "good" are adjectives that are strongly coupled to the use case.
My main goal is to find good literature.
|
|
|
|
|
Can't tell if you want to secure a door or an app.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
|
|
|
|
|
App 
|
|
|
|
|
|
Thank you. The information seems quite specific. Do you also have literature for a more abstract - close to mathematical - level?
|
|
|
|
|
It's a "Pattern"; like doors and windows. An abstraction of a "door" or a "window"? That would be a rectangle; with a hole. Math? Width and height.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
|
|
|
|
|
I use IODA architecture since the very beginning, nearly 10 years now. I develop SW more than 30 years as a profi.
It is the first approach/technique, that offers the option to produce reusable, small classes like "lego building blocks".
It is ortogonal to other best practices like CleanCode, DRY and other.
It has a strong impact on unit testing: Unit testing will be easy.
The reason is simple:
IODA was designed to reduce dependencies between classes to a minimum.
I - Integration - calls all other, but does not contain logic or data
O - Operation - contains logic, just work with data, do not know other operations or integrations
D - Data - holds the data and knows/deals only about itself and its childs. No operations here.
A - API - Application Interfaces - To deal with the environment or to provide general functions like data validation and other stuff
All of these classes are devolped by the project. Standard-Libraries can be found below this code.
If you then divide the Operations in "IO operations" and "other operations" you get the following:
IO_operation "reader" reads "data_ger",
operation "translator" translates "data_ger" to english data "data_eng"
and then the IO operation "View_data" presents it to the user.
And the integrator "TextManager" has a method like
void viewAsEnglish(int textIdGer, string readLang = "ger")
{
var data_ger = reader.Read(textIdGer, readLang);
var data_eng = translator.Translate(data_ger, "eng");
dataViewer.View(data_eng);
}
After longer practice you will automatically think in these four types, when creating a new class.
And yes, it will work fine in real projects.
Now lets start discussion about it, if you want.
**Links**
Detailed description:
https://www.infoq.com/news/2015/05/ioda-architecture/[^]
Bigger sample project (documentation only in german, sorry)
GitHub - jpr65/VersandEtiketten: This is a reference projekt for IODA and also a developer test ticket. Only in german, possible later also in english.[^]
** Images **
IODA - Principle Diagram
IODA - Libraries
modified 1-Dec-23 2:53am.
|
|
|
|
|
I haven't seen this architecture before. Seems interesting but there is one thing I don't get:
Ralf Peine 2023 wrote: I - Integration - calls all other, but does not contain logic or data
O - Operation - contains logic, just work with data, do not know other operations or integrations
If I need to to implement something like "if operation1 fails do operation2", where do I place the logic? I cannot place it in the integration units (cannot contain logic or data) and cannot place it in one of the operations (do not know other operations).
What is the solution?
Mircea
|
|
|
|
|
Yes, that's some important point.
You may add minimal logic into the integrator to set switches for the workflow.
However, the decision why a switch is set should again be made in an operation, if it is not a simple true/false decision.
It is ***allowed*** to put logik in integrators, sometimes it is easy for the moment, but this leads to technical debt and is sometimes difficult to test. Especially while doing hotfixes, you will break the IODA principles, and this is ok, for the moment.
The more you stick to the IODA principles, the more testable and reusable the code will be, that you create.
|
|
|
|
|
Like perhaps everything along these lines it has some utility as long as it is not taken as an absolute.
I certainly do not want to see every single operation in a million line application decomposed like this.
For example I have written reports at various times in various languages and attempting to decompose them often lead to nothing but making it harder to understand the whole.
Then after that one must consider in standard business programming of maintaining an application over years with multiple changes in employees is this likely to be rigorously maintained? If it isn't then doesn't it just devolve into what the original code might have looked like the first place?
|
|
|
|
|
I see ETL: extract, translate, load.
Your sample "integrator" looks more like a "translator".
While my "data" may not have methods, it resides in a repository that does: add, delete, update, query. Data as another object. The methods represent an "API".
And I have "operations" that depend on other operations: "distance" calculators for anything that needs to calculate a distance (in pixels, paces, yards, or feet); or an angle; or "a collision detector" while moving.
I don't disagree with the architecture, but things aren't that clear cut.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
|
|
|
|
|
This example and the dokumentation are very small. Please read the detailed description of Ralf Westphal.
"Translation" is what the integrator manages, and "integration" is the type of the class in the meaning of IODA.
Reason for IODA approach is the following
"Functional dependencies are evil!"
Only the integrator classes are allowed to call operation classes.
So, in an ideal world, you have no functional dependencies.
|
|
|
|
|
If you want people to appreciate the benefits of this approach, write an article, hosted here on Code Project showing a non-trivial application created using these techniques. I find myself deeply distrustful of prescriptive ways of working when they are presented as absolutes (e.g. "Functional dependencies are evil!"). No approach is 100% perfect and, generally, there are always be exceptions to the absolutes that are the driving reason for a new way of working.
|
|
|
|
|
I'm working on a WPF app for a client that gets invoice remittance data from their customers as CSV files. My app extracts the Invoice Date, Reference Number, and Amount using metadata defined by my client for the input files. This metadata defines the columns in the file, and the index position of the Invoice Date, Reference Number, and Amount.
The key is the reference number. It is used to look up the invoice row in their accounting system. Once in a while their customer formats the reference number wrong. For example, one of the reference number formats is eight numbers, as in '20158011'. But from time to time, they send invalid data, as in 'RECOUP19718470_002ZT'.
Another customer's format is 'RDHC608965'. This column's reference number always starts with 'RDHC'. The remainder is the six-digit reference number.
What I would like to do is provide a place in the UI to define the required format of the reference number. If it was internal I would specify a regex, but this is something that my client has to define.
What's the right way to go about this?
In theory, theory and practice are the same. But in practice, they never are.”
If it's not broken, fix it until it is.
Everything makes sense in someone's mind.
|
|
|
|
|
"Reference data" is just that; there's nothing to validate. It's like a "comment"; subject to the whims of the client and not something to build a "system" around.
"Invoice payments" are generally applied to outstanding balances, and not any particular invoice; that shows how "important" the reference number is (not).
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
modified 15-Nov-23 20:27pm.
|
|
|
|
|
Gerry Schmitz wrote: "Reference data" is just that; there's nothing to validate. It's like a "comment"; subject to the whims of the client and not something to build a "system" around.
It's not 'Reference data'. It's a Reference Number. The Reference number in the customer's data file is a 'reference' to an Id in the client's accounting system. This matches an invoice from the client to a customer's payment. It's a reconciliation.
But that's not the question. See this[^]
Since my client can define new customers in the app, I need a way for them to tell me the FORMAT of the Reference Number. Then, my app is reading the file, I can verify the number so it can be looked up. The format of the Reference Number CAN be validated. I COULD use a RegEx, but defining the validation for it has to be user friendly.
The question, again, is how to allow the user some way of manually inputting the format of the reference number.
In theory, theory and practice are the same. But in practice, they never are.”
If it's not broken, fix it until it is.
Everything makes sense in someone's mind.
|
|
|
|
|
It's reference "junk" when it's in some one else's system that you have no control over. And anything you put in the way of "the payment process" makes you the problem.
A company issues their own "client numbers", Purchase Order numbers, Invoice Numbers, Product Numbers, etc. which becomes "reference junk" in someone else's system that they then send back to you as more "reference junk". You reconcile your own paperwork; not someone else's.
I'd send "checklists" of the outstanding invoices if they wanted to "reference" something.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
|
|
|
|
|
This whole process is outside of this discussion. I have no control over how my client processes their invoices.
Thank anyhow
In theory, theory and practice are the same. But in practice, they never are.”
If it's not broken, fix it until it is.
Everything makes sense in someone's mind.
|
|
|
|
|
You have (some) control when inputing from documents; then you can use "fuzzy" searching then and there.
You said CSV ... which means, you have to get it into the system "before" you can do anything with it. Throughput.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
|
|
|
|
|
Kevin Marois wrote: What's the right way to go about this?
Your requirements description is incomplete for a start.
You must also answer the question - what do they want to happen when it is wrong? For example discard the entire csv, ignore than one row, collect the failures and present immediately to a human, collect somewhere and allow for a report. Might be some others.
You should also decide who actually manages it? Is it set up on install? Can it change day to day? Can different users change it (so not the customer but individual customer users.)
Those drive how you configure it. For example if install then you would need to ask that during install. (This probably is not viable since their needs might change over time.)
But other than that your application should have a section specifically for customer configurations. Yes plural. Presume one now and more for the future. And possible one for the application (admin users only) and one for normal users. You application might not support multiple users so the different levels might not apply.
With both application level and user level then you need to decide if the user one overrides the application one completely or if both work.
You can save the configuration in a configuration file or a database. Or other persistent store. You would normally load the configuration on start up. Naturally updates while running must impact the loaded configuration also.
|
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
