Only recently, there was a post on GRAKN.AI which seemed heavily inspired by Prolog.

This is good news for Prolog: Modern Prolog systems provide many features that are important in the domain of databases, such as JIT indexing, transactions, and dedicated mechanisms for semantic data.

You can quite often apply Prolog in actual practice if you know it. For instance, see how often "logic" is mentioned just in the context of the present discussion. It's nice to know a logic programming language that can elegantly express business logic and business rules!

person(Firstname, Lastname, Gender, Age, State, Country).

Accessing the attributes by position makes the code hard to read because one has to remember the position and mistakes are not catched because there is no type system.

Would you recommend using the libray record or using dicts in swi-prolog? This would make the code non portable. Moreover, do you recommend dedicated strings or list of atoms to represent strings?

The Prolog syntax is very elegant but it could be clunky, I totally agree. Our syntax has resource and role names instead of relying on position:

e.g.

match $x has lastname "some last name";

If ever you decide you want to migrate away from Postgres, you'll need to rewrite all the business logic into some other out of process language. If ever you want to scale this, such that you want to calculate states in batches, you'd want to have the business logic somewhere else...

> If ever you decide you want to migrate away from Postgres, you'll need to rewrite all the business logic into some other out of process language.

Yeah, but I don't consider this a bad thing. IMO migrating between databases should always require a lot of rewriting. If it doesn't, you're most likely underutilizing the features provided by your database.

> If ever you want to scale this, such that you want to calculate states in batches, you'd want to have the business logic somewhere else...

I'm not sure I follow. The application I'm using this in has > 1 billion rows and we frequently re-compute the state of all our entities across the entire data set in batches after we make changes to our logic. Having the code that does this in the database avoids having to move large amounts of data between our db and the application.

Very likely true. But when you get locked into Oracle and it makes it nearly impossible to move because of the features your relying on, has a big effect on shaping your perspective on this.

> The application I'm using this in has > 1 billion rows and we frequently re-compute the state of all our entities across the entire data set in batches after we make changes to our logic.

That is very impressive. Mind sharing the rate of change across all of those rows?

Also, I'm curious. Is this all that DB instance does? Or is it responsible for other data as well?

I hear you. Picking a database is major decision and shouldn't be made lightly. That being said, I hear a lot of companies are successfully migrating from Oracle to Postgres these days. And in fact, this application was actually migrated from Couchbase to Postgres, but that's a story for another day perhaps :).

> That is very impressive. Mind sharing the rate of change across all of those rows?

Nowadays its about 5 million inserts / day with peak rates of 150 inserts / sec. Not too crazy, but it adds up over time :).

> Also, I'm curious. Is this all that DB instance does? Or is it responsible for other data as well?

It does a few other things, but this is the main workload.

That said, I much prefer to put all this logic into SQL, and to use PostgreSQL, than the alternatives. There's any number of reasons for this:

- direct access to the DB is not dangerous if the logic needed to keep it consistent... is in the DB

- you don't have to replicate this logic if you add new front-ends

- SQL is very expressive, so you'll end up having less code using SQL than anything else -- this also means that RDBMS migration costs need not be as high as one might think, since there will be less SQL than the equivalent for the front-end

- updating business logic is easier to do atomically

[1] http://www.techrepublic.com/article/is-sql-a-standard-anymor...

it's one of the things to keep in mind. Depending on your application, migrating away from postgres will be more or less painful. If you're already deeply invested in postgres features, this is just one more problem to solve.

> If ever you want to scale this, such that you want to calculate states in batches, you'd want to have the business logic somewhere else

yes. This has the all the usual issues of putting (some) business logic into the database. On the other hand, by using this, you're basically just creating a data integrity constraint similar to a foreign key, just one not as widely supported.

Still. If you ever plan to move to a database that doesn't support foreign key constraints, you will have to implement the business logic somewhere else.

For me, data integrity is paramount. If ever I can put a data integrity check directly into the database, I will do it because bugs in the application logic can exist and when the database itself enforces integrity constraint, I'm protected from those.

I don't want to have to deal with, to stay in the framework of the article, a shipped, but unpaid order. Was this a bug in the application? Did it actually ship? Did the payment fail?

If the database blows up on any attempts to store invalid data, I'm protected from having to ask these questions.

I'm always surprised when people fight using constraints. During dev, I see them as a godsend for spotting problems - I don't know how many bugs having self-enforcing data structures has caught.

I will say that I have, under protest, turned them off in production once for performance. A particular heavily-used flow was annoying due to a large number of FK checks on an intermediate step. But never during development, and given the number of FK-violation errors I've seen in production code, preferably not even then.

Fixing code is almost always so much easier than fixing data.

Loose coupling has a lot of other advantages but this seems to have become the biggest selling point for a loosely-coupled data layer.

I wish I could say that. I've done more than I care to count. The issue is usually one of the following:

1) converting between types of storage layers (SQL/NoSQL/flat file). I've never seen an abstraction layer that could handle MongoDB and later be converted to Postgres for example (a real migration I had to do once) and still do justice to either backend.

2) Hopefully you pick a storage layer for a good reason. For example, say you picked Postgres because you have Geospatial needs. Business later dictates you have to use MySQL "for reasons", what abstraction layer is going to help there to not require code refactor?

3) If you are doing things right, there's more than App code that interfaces with the DB. You've got the entire devops chain (backups, automation, etc) that will likely need to be rewritten. This can take an immense amount of work as well.

Anyway, not saying it's not possible, but I've found it a good litmus test of another engineer if they think moving to another storage layer should be simple. It's not always, but still more often than not shows inexperience or over optimism.

It is an interesting question about how successful people are in migrating to different storage engines.

Not that I doubt you, because I know with the right trade offs it's possible, but is the operative word "could" or did you actually ever migrate one of those architectures?

> Obviously to go to no-sql or other less conventional storage is a different story.

Ruling those out feels a bit disengenuous. There are plenty of good reasons to need to migrate to or from a relational DB.

And even where there are no feature or syntax issues there may well be optimisation differences. For instance going from MSSQL to postgres you might hit a significant difference with CTEs because MSSQL can perform predicate optimisations through them but postgres doesn't - this might mean core queries need to be refactored significantly for performance reasons (to avoid extra index or table scans) if not functional ones. (not intending to pick on postgres here, I'm sure there are similar examples in the other direction and between other engines, but this is the most significant example that immediately springs to mind).

For constraints, validity checking, things like adding/updating timestamps, and other things that are about data integrity, about the only time I don't do that in the DB is when outside information is involved such that it can't be. Otherwise, to the extent possible, I want the datastore to only accept valid data. This goes to the notion that fixing code is easier than fixing data, so the store can not only defend itself, but also help catch bugs.

There are also times when dealing with huge amounts of data that doing whatever you're doing in an SP is the only way to get decent performance. Dragging enormous tables over the network to process is sometimes really wasteful. If you're tuning indexes and whatnot against that model, you're already changing the DB, and doing so at a level that is implicit rather than explicit, and in ways that can change out from under you (if the statistics change).

How is this any different than anything else where DB data & application logic have to be kept in sync?

version control + loader/change scripts should handle it, no?

Of you follow this rule you'll never be able to use the most powerful features of postgres.

> If ever you want to scale this, such that you want to calculate states in batches, you'd want to have the business logic somewhere else...

Not due I'm totally on board with the premise here, but the need to calculate state outside the database is a generally valid one and so it might be a good idea to write the core of the transition function in JavaScript and implement the postgres functions in PLv8.

If there is significant functionality we want to take advantage of in this new platform, then that means that we have to refactor anyway to use this, because by definition we either don't do it today, or we do it with an insufficient workaround in Postgres.

The migration necessitates large refactoring by its very justification.

I like to think of each layer in a tiered system as having differing deployment requirements and time tables. Front-end systems will be very frequent, backend systems possibly less so, though not necessarily, and then DBs ideally infrequent and they generally take longer.

At a minimum, keeping them decoupled is freeing for patching bugs and releasing features independently. It does raise the bar for keeping all changes compatible with existing systems.

Of course updating a database is much harder than overwriting executable or library files, but a lot of the objects in a database should be safely updatable by simply dropping and recreating them.

The tricky changes are of course things like, e.g. splitting one column into two or merging two columns in two tables into one. But those changes are even harder to do the 'dumber' your database is, i.e. the less logic there is in it that enforces a certain level of quality in its data.

My favorite, and the only one I've used extensively, is [DB Ghost](http://www.dbghost.com/). What I like about it compared to all others I've run across is that it, by default, will automatically sync your target DB (e.g. a production DB) with a model source DB that it also builds automatically.

So instead of scripting out every schema change as explicit SQL statements and queries you just maintain the scripts to build the model source DB, e.g. to add a column to a table, instead of creating a SQL script file to `ALTER TABLE Foo ADD COLUMN Bar ...` you just update the existing SQL script file with the `CREATE TABLE Foo ...` statement. When you deploy changes – 'sync' a target DB in the DB Ghost terminology – it automatically detects differences and modifies the target to match the source.

The benefit being that neither you nor the DB Ghost program needs to explicitly perform every single migration since the beginning of time. Only changes that need to be explicitly handled as migrations, but, with a little customization, doing that is pretty easy too.

The bar now for me working with databases is whether I can create a new 'empty' database (with test data) in a minute or two and whether I can automatically deploy changes (or, as I'm doing now, generate a deployment script automatically). Given that, I can actually do something like TDD for databases, which is really nice, especially if there's significant business logic in the database (which there almost always is in my experience to-date).

You'll rewrite your front end 20 times in 20 different "frameworks", and your app 5 times in 5 different languages, for everytime you actually change databases

On one hand, using a bunch of PL/pgSQL makes sense for transaction isolation and faster execution. I'm not sure how much of the arguments about business logic matter in this case, but a strong argument for using PL/pgSQL is that these queries written in Python (for example) are going to be significantly slower, and that really matters when there are 100 concurrent users hitting the database every 10 seconds. No one likes waiting for their system to update... they'll just switch over to Excel. I think that PL/pgSQL is a great use-case for situations where preoptimizing for speed isn't a mistake, though the trade off is that you need to find the rare expert on processing languages (what order to triggers fire and how do you prevent this from being an issue?), and who can work through the complex logic of the system.

I wonder why the author didn't use windowing instead the lateral example. You can window and sum over composite rows, and it goes pretty fast. I never did a direct comparison between lateral and windowing, but windowing would be much cleaner and not require you to generate a series.

It's almost as crazy as the Excel spreadsheet I found to simulate a neural network.

I've always loved seeing how far you could push SQL and databases to do things that (on the surface) would seem extremely difficult or impossible.

What it usually turns out to be is possible - but in the long-term unmaintainable. This FSM is not that - not yet (if it gained even more states, it could get there). But I have seen (and I have unfortunately written) queries in SQL that could make your hair stand up. Insane extreme monstrosities that I am both proud and ashamed of (fortunately, I don't work for that company any longer).

On a different but related note, I do recall one query that a friend of mine wrote to allow the querying of a zip code database (which had lat/lon columns for each zip code), to be able to calculate distances from a given address - using SQL. It was a simple form of geo lookup he had to do for a particular project. I later used the same code as a part of a lookup process for placing markers on a google map (markers would indicate whatever was needed for "show me locations near my address within N miles"). It was a pretty interesting piece of SQL for the reason of doing the specialized distance calculation (can't remember which calc, but one of the simplest ones that didn't take into account certain things about the earth's "roundness" to make things super-accurate - that wasn't needed for short distances and purpose).

That being said, I think things can be somewhat tamed by applying the same best practices to your SQL as you apply to your regular code: I.e. lots of tests, good comments and documentation, extracting logic into either set returning functions or functions suitable for lateral joins (both can be inlined if done right [1]), keeping things in version control, etc.

But yes, applying those practices can be harder in SQL than it is in your application layer language for various reasons. So I'll always recommend avoiding going too crazy. You can often get the best of both worlds by making pragmatic choices about what should be done at which layer. No need to enslave yourself to a false dichotomy.

[1]: https://wiki.postgresql.org/wiki/Inlining_of_SQL_functions

I like the type safety of ENUMs, and I'm actually using them in my application, but they can cause a lot of problems. E.g. it's currently almost impossible to remove an ENUM value [1], and adding a new value can't be done inside of a transaction.

The state transitions could definitely go into a table, but I feel it's overkill in many cases, and would have probably obfuscated the main ideas presented in the article.

Anyway, thanks for your feedback. There is definitely a lot more to explore for anybody who is interested in applying these ideas in practice :)

[1] https://www.postgresql.org/message-id/21012.1459434338%40sss...

In the end it is just quibble about style anyway and one usually end up with what is measured faster in testing..

I agree with your comments about ENUMs, though, especially having built many "static" products that suddenly need their ENUMs changed, and having a group of developers visiting me and asking why it's not as possible as they had assumed :)

YMMV but I don't think a blanket "just use enum" is the correct approach.

    BEGIN;
        ALTER TYPE my_enum ADD VALUE 'bar' AFTER 'foo';
    COMMIT;

Also, removing a value is a pain (though, that may have changed now).

I know you can do it — but I've found after trying a bunch of approaches that text with constraints is a better starting point. Just wanted to point it out because it's something I researched a fair amount. Many guides say that enum can be a pain but I decided that I'd use them to be pure, then discovered they were a pain, and now don't use them as a starting point.

ERROR: ALTER TYPE ... ADD cannot run inside a transaction block

"Reduce locking required for adding values to enum types (Andrew Dunstan, Tom Lane)

Previously it was impossible to run ALTER TYPE ... ADD VALUE in a transaction block unless the enum type was created in the same block.

Now, only references to uncommitted enum values from other transactions are prohibited."

https://www.postgresql.org/docs/devel/static/release-10.html...

Secondly, what are your thoughts on using enum generally? Would you recommend them in cases where you don't need to really optimise for narrower columns?

It's not that index structure specific atm, even though it could possibly be avoided for some types (e.g. hash, although there's considerable visibility issues to make that possible). Consider e.g. a btree index, if the deleted enum value ends up in an inner page, you really need to know how it compares to know where to descend to.

> Secondly, what are your thoughts on using enum generally? Would you recommend them in cases where you don't need to really optimise for narrower columns?

I like using them. There's cases where the transactional restrictions make it too problematic, but other than that I think the documentational advantage is substantial, besides just the width.

That would likely add an unnecessary read and slow things down just a little bit.

https://github.com/agentm/project-m36

> Unlike most database management systems (DBMS), Project:M36 is opinionated software which adheres strictly to the mathematics of the relational algebra. The purpose of this adherence is to prove that software which implements mathematically-sound design principles reaps benefits in the form of code clarity, consistency, performance, and future-proofing.

My e-mail is in my profile.

That being said, I'm in Shanghai frequently to sync with my colleagues and it's an amazing city. I've found pretty much most conveniences available to me in Berlin and to some extend even better. There is an active expat community, and a fascinating vibe.

My manager (from SF, USA) has been there for over a year now and has enjoyed the experience a lot and extended his stay. YMMV, but I'm fairly well traveled and would say that it's a pretty great spot if you're willing to emerge in a foreign language and culture.

Also your colleagues may curse you down the line when they need to change the state machine's behaviour. Would of course be possible to keep note of a "state machine version number" on each row but you would end up with a transition function that had to know about every historic version of the state machine...

This will get much easier to do now with the new "pointer passing interface" in SQLite3 3.20.0 (http://sqlite.org/bindptr.html).

I'm not recommend this, nor against this. Just observing that it's a legitimate design pattern.

Woot? Apple uses Go?

Anyway, YMMV and I'm not recommending to apply the ideas in this article in every situation. But after having had all of this logic in the application layer before migrating to Postgres, I find this approach more maintainable.

Furthermore, I'm challenging the utility of putting the transaction logic in the database. It doesn't make sense other than to seem elegant. His application needs to actually respond to the events so it's just silly to separate the transitions. This is a complete redo.

So let me explain in more detail how you could solve this without all of the overhead you have created. Assuming a true FSM, there is no need to store the state as a value in your database (e.g. "awaiting_payment" in a [state] field). Instead, you should be storing your states as BIT fields:

[awaiting_payment], [awaiting_shipment], [shipped], etc

where "1" indicates the current state and every other state field is "0". In addition, you will need a nullable timestamp for each state:

[awaiting_payment_timestamp], [awaiting_shipment_timestamp], etc

to store when each state was entered, if at all. In this way, every order is in a single row and your analytics become trivial queries that don't require any complex joins or correlated sub-queries. Additionally, using simple constraints on each state field (e.g. the [awaiting_shipment] field <> 1 if the [awaiting_payment_timestamp] is NULL) would remove the need for your trigger and, again, reduce overhead.

This is called an "accumulating snapshot" table, and it is used in data warehousing to solve the exact problem you are dealing with here - tracking an entity through a series of predetermined events to monitor it's status and mark milestones. Using a "transaction" table like in your example, while certainly possible, is storing your data in a way that you cannot easily answer the questions you may want to ask of it. You are spreading your entity across multiple rows instead of putting it all in one row.

Think about how the data translates to an object in your domain. Let's call this table "order_state" ("order_events" is a bad name for two reasons: it's plural - which is just wrong - and it doesn't faithfully describe your entity). I'm going to assume "order" and "order_item" tables as well. These three tables create an aggregate root that can be used to derive an Order object in your domain. This Order will have, at the very least, 3 properties corresponding to your entity's attributes in your database:

id, items, state

where "items" is a collection of OrderItem and state is OrderState (see where I'm going here?). The question becomes about how the OrderState is populated. If your Order object represents the "context" in your FSM, then the OrderState is the base class for all states:

AwaitingPaymentState, ShippingState, CancelledState, etc

This means polymorphism is inevitable. There are 3 common ways to deal with polymorphism in a database, but using a transaction table (like in your example) is not one of them. For your problem, utilizing "single table inheritance" (where all objects in a hierarchy map to a single table row) makes the most sense because you are not storing different data in each child class, rather, just a marker to indicate which state is current (this is also generally the most performant solution). It also means you don't need to add any additional tables, because your "order_state" table can serve both purposes.

Using a transaction table, you are adding a 3rd orthogonal dimension to your aggregate (another set of primary keys), so your object graph becomes needlessly large and it becomes a chore to choose and hydrate your current state.

My last critique is in regard to the choice of putting your transition logic in the database itself. I just don't see how that can be useful other than to "seem" clean. What I mean is, somehow an order actually has to be shipped or an item has to be paid for, and adding a row to a database table doesn't do that. So somewhere else in your stack you must have duplicated your transaction logic in some fashion in order to facilitate the correct database calls when the events take place in your domain.

For example, using my above objects, when a the "pay" event is raised on your Order object "context", it will invoke the "pay" event on it's current state, AwaitingPaymentState, to undergo transition. In this scenario, there must be some method (an "action") that actually handles the payment. Either there is no state machine at all in your domain and you need to validate the state before processing the payment, OR you have a state machine like my example and you need to transition to the AwaitingShippingState. In either case, the logic is duplicated.

What you have created is a cool academic solution that may work (I don't doubt that it works), but this problem was solved 30 years ago and if this was done on my team, I would insist it be redone to adhere to the basic principals of the entity relationship model and database design.

Does this make more sense?

There are certainly many ways to model this problem in the database. But to me the main idea of the article is implementing a FSM as a user defined aggregate. This is technique is actually not mutually exclusive with an "accumulating snapshot" table, and in fact the trigger could be modified to maintain such a table instead or in addition to the order_events table.

I don't quite understand why you insist on not having a transactions (order_events) table. IMO it's very helpful for auditing purposes, and can be used to capture additional data for each state transition. E.g. you could easily put an IP column on it. From what I can tell you'd have to add one column per event for doing this with your accumulated snapshot table. Our application has ~10 events, so I'm not sure I'd like to have an explosion of columns on it.

Also, I think your solution falls apart completely when it's possible to re-enter a previous state. (or would you overwrite the first timestamp?)

Last but not least, there are many advantages to putting logic into the db. I know you object to the advanced analytics because they can be done differently when you control the DB schema from day one. But imagine you inherit a database with an events table that can't easily be modified to use an accumulating snapshot table. How would you write analytical queries against it? Another advantage of keeping logic in the database is reducing the number of queries (and network overhead) that are needed.

Anyway, you've made many good suggestions, but I think you've taken the toy example a bit too serious. It's very hard to come up with good real world examples without disclosing details of your actual application (which I can't in this case).

Anyway, using FSMs backed by user defined aggregates is definitely not an academic solution. This powers a mission-critical application of one of the worlds big companies for a database with over a billion records. It's also not the first iteration, and previous implementations at the application layer had caused problems that FSMs inside the DB have solved. Last but not least, this modeling the problem domain as FSMs has allowed the entire team (including non-technical people) to deeply and correctly understand how our analytics work.

Along those same lines I'm not insisting you can't have a transaction table. If you need to record more information, like an IP address or user_id with each transition, then it would require a transaction table (although the article mentions none of this).

Honestly, the storage concerns I bring up (that is the schema) are really far less of a concern than the basic idea of storing your transitions in the database. Put simply, you are storing the wrong information. Instead of storing the actual state, you are storing a series of inputs that allows you to derive the state. This is a HUGE mistake. What if you add a new state? Say, [in_transit]. Every one of those 1 billion rows that are already in there will no longer be able to be processed correctly because they won't have the correct order of events.

This entire thing could be one table that stores the [order_id] [state] [timestamp] [is_current], with a stored procedure to update it that takes an "order_id" and "event_name". Instead, you have created a house of cards.

Anyway, coming back to the example from the post, if I change the FSM (e.g. add a state/event) I need to make sure all existing data conforms to it one way or another (by making the change backwards compatible or by migrating the data). But the same problem exists for any other invariant you're trying to enforce/add to your DB (FKs, Checks, etc.). It's just the cost of enforcing an invariant.

I'm also confused at this point how you jump between saying it's okay to have a transactions table and then later backtrace and say it's a huge mistake because it's not storing the actual state. I already pointed out that the two are not mutually exclusive, and that the trigger could maintain an accumulated snapshot in addition to the order_events table. In fact, the trigger could event derive the new state based on the latest value from the accumulating snapshot table which may also address some of your concerns about adding new states.

Also, I haven't created a house of cards. I had the novel idea of for leveraging user defined aggregates as finite state machines and decided to share it with the PostgreSQL community. I couldn't use my companies business domain as an example, so I had to come up with a bogus one that people would understand easily. Now don't get me wrong, I'm totally enjoying the discussion around this bogus example, but I'm not enjoying your unwavering faith in the absolute correctness of your ideas and the HUGE mistakes you claim I'm making. So is is probably my last reply unless you're cool with toning it down a bit.

[1] http://disq.us/p/1l10by0

So ruling out a snapshot table (I think you have made clear this is not an option), your best choice would then be a transaction table where your [state] is a Type 2 Slowly Changing Dimension (SCD2) with an indicator field for the current state. I want to be clear here: you should be storing your state, NOT the events (arguments) that can be used to derive your state. I cannot express this enough. Your problem is not unique. You simply cannot be naive enough to think you are the first person to want store the state history of an FSM in a database. This is literally data warehousing 101. The novelty of your solution stems from the fact that you are (dangerously) misunderstanding the problem. If, for some reason, you need to store what [event_name] caused the transition to the [state], just add a field. Simple.

As for invariants. Let's use an example: say you add a "package" event that transitions to an [in_transit] state between [awaiting_shipment] and [shipped] (note here we may also then want to change "ship" to "deliver"). This simple and totally reasonable change would make every order with an event queue: "create", "pay", "ship" result in [error]. If you just stored the [state], then a [shipped] order will remain [shipped] regardless of what series of events brought it there - which would be correct. There would be no need to go back and change historical data - nor would you want to. How would that even be done? Manufacture shipping information and timestamps? There are no invariants in this process (again, I want to reiterate that this is a solved problem). True FSMs (if this is actually modeled as one) only care about the current state anyway and maybe the previous state (see SCD3). What I mean is, future actions are determined using the current state. If the current state becomes [error] for 1 billion orders, that's a problem. A huge problem.

"Absolute correctness" is an interesting term... and I agree that I am probably coming off a bit aggressive in my comments. Look, I'm not trying to be a bully here, and I DO think that the FSM you have created is interesting, unique, and a cool example of how user-defined aggregates can be used. The problem with it is, unfortunately, an extremely common one. If I had a nickle for every time a programmer jumped into the world of databases and came up with a "new" solution to a "never-before-seen" problem... I'd be a wealthy man. Relational databases have been around for a long time and are one of THE MOST studied and written about pieces of infrastructure/software in existence (they're interesting for so many reasons!). What this means is that it's EXTREMELY unlikely you have found a problem that hasn't already been "solved" (of course there can be wiggle room) in a theoretically correct manner. It's almost always just an individual not recognizing (or breaking down) the problem efficiently.

And that's what I am seeing here: a problem that is present and handled in LOTS of databases (I've personally done this dozens of times), but is being approached from a less-than-perfect (albeit unique) angle. I can't say that I'm "absolutely correct"... I don't have the full grasp of your domain and business requirements. I don't know precisely the degree of constraint your schema is imposing. I don't know how well communication between teams and management of your stack is handled (yes, these can be important - you did mention you had problems in these areas). Maybe yours really is the best solution - though I remain unconvinced. But I can say, that given what I know about the problem, the entity relational model, database design, and data warehousing that your solution is simply not sound. It's difficult to put that nicely. I don't mean that your wrong (apparently this is working for you). It's not black and white. Just that if one of my clients came to me with similar requirements, I would follow "the book", because this has already been written.

Here's what I'd do (SCD6):

    Order [OrderId] ... [additional data]

    OrderStateHistory [OrderStateId] [OrderId] [TransitionTimestamp] [PreviousStateName] [EventName] [StateName] [IsCurrent] ... [additional data] 

    OrderTransition [CurrentStateName] [EventName] [NextStateName]
    

order_state_history_tr @current_state_name, @event_name -- note this the PK of OrderTransition

to facilitate transitions. This stores BOTH the transitions and the state.

https://docs.microsoft.com/en-us/sql/relational-databases/ta...

A second option would be to have a transaction table to record each transition (and a reference table containing all valid states), and then have foreign keys in the accumulating snapshot table to the transaction table. Although this would be kind of silly for the current problem due to its simplicity.

A third option would be to use an [is_valid] calculated field to record whether or not the state is valid.

I'm sure I could come up with more, but it's all a silly exercise because I would do (and recommend) none of these and keep business logic where business logic belongs. This would allow for more flexibility for different "kinds" of orders, more complex validation, etc.

When you can query the SQL schema using SQL queries (though I admit that the pg_* tables kinda suck) and when you have things like record types, you really do have a very powerful SQL.