Do you have multiple separate apps that can change a shared DB schema?
How do you keep that all in sync across your apps?
So usually one is the “main” but, personally speaking, I’d just say “don’t do it,” for the obvious issues I suspect you’re aware of ;-)
By “don’t do it,” I mean having multiple apps talk to one DB schema.
"don't do it" is the right answer. Others have pointed it out as well, many large SaaS companies I worked with, have had apps owning their databases. Anyone else needs anything - use APIs (and ETL if you need everything)
It's definitely not the right answer. It's actually the completely wrong answer.
Services are slow, restrictive, and don't enjoy the benefits of an actual DBMS, like transactions. You also add additional dependencies and failure points.
This isn’t a discussion of whether you should use services or not, only what to do in the case where you do have separate services interacting with a schema. And, in this case, trying to maintain consistency with multiple apps accessing the DB directly can quickly turn into a nightmare. Better to maintain separate services interacting over some contract to insulate multiple codebases from the internals of each other.
> Better to maintain separate services interacting over some contract to insulate multiple codebases from the internals of each other.
All you've done is shifted the dependency and made extra work when schemas change. Now we have schema updates, _and_ service updates.
As I said, you give up transactions, performance, and flexibility. You also increase the workload, and increase failures. There is almost no good reason for what you're proposing.
I am not sure I understand. The context was multiple applications managing/accessing the same database. No matter how you slice it, a schema update is accompanied by application updates. If anything, in a service oriented approach, you can potentially isolate the impact by versioning APIs and writing compatibility logic in a single place.
> a schema update is accompanied by application updates
1. DB -> App = Two updates, the schema and the app
2. DB -> Service -> App = Three updates, the schema, the service, and the app
In both cases, obviously the DB changes.
In both cases, obviously the App changes.
In the second case, you also get to update your services.
No, it’s not obvious the DB changes because the team depending on that schema doesn’t necessarily talk to your team much, and you might not even know that team exists.
The same is true for API contracts but the culture of changing an API is much more understood as something that must be communicated and processes are more rehearsed.
> No, it’s not obvious the DB changes because the team depending on that schema doesn’t necessarily talk to your team much, and you might not even know that team exists.
Literally everything you just said applies to services and API contracts as well. You haven't solved anything.
> the culture of changing an API is much more understood as something that must be communicated and processes are more rehearsed
You're just making this up. There are thousands of incidents every day because someone changed an API without communicating to every stakeholder.
"Services are better than databases because service owners are better at communicating changes than database owners" isn't a very compelling argument.
I'd work to fix the communication problem before I adopted a terrible tech solution.
>> the culture of changing an API is much more understood as something that must be communicated and processes are more rehearsed
> You're just making this up. There are thousands of incidents every day because someone changed an API without communicating to every stakeholder.
What you wrote isn't a contradiction of what I wrote. I wrote "more understood" not "perfectly solved problem."
> "Services are better than databases because service owners are better at communicating changes than database owners" isn't a very compelling argument.
Except they are because the tooling is better. I can check calls to an API endpoint using the same tooling that I would for checking end user API calls. I can push out comms to use the new API, I can confirm calls go to 0 before deleting the old endpoint.
For checking that a table or column is being accessed before changing / deleting it? Fuck, I don't actually know how I'd orchestrate that. And once I've made that change (vs when I make an API change), well, I hope I got it 100% correct because reverting that change is harder.
It is more that services allow more backward compatibility than direct DB changes, because you can also code additional logic to maintain that.
Of course in a narrow case where all apps that need the same DB need to be upgraded/fixed in lockstep, maybe you can get away by directly updating the database. But more than 2 teams, it will quickly be a nightmare of conflicting priorities across teams, amount deployment coordination required.
"Oh you cannot change this schema right now because 3 other apps depend on it and they dont have time to make the necessary changes in their app to accommodate this right now, for the next 6 months"
Except if your schema changes in a way that forces your API contract to change, you have a big red alarm to tell you to investigate what’s using that contract and update it too.
When your apps connect to a DB schema, the change is silent. You just run ALTER TABLE and… whoops, you just brought down a service you didn’t know existed because the contract was implicit.
In situations like these, the database admins are the ones responsible for the schema; the apps are mere users.