Clojurians Log v2

Hello, datomickers! Newbie question: is it good practice to use Datomic entity ids as identifiers shared with the application in a similar way that we use surrogates primary keys on relational databases?

You probably should create application level ids if you are sharing them with any external place

Like other systems or even having them in URL routes or similar

I have a question about tuples; Say that I have a category with tags, and these tags are component/many of this category. Now, I’d like to add a composite (tuple) key to this tag entity that says `[tagName, category]` is unique, but there is no explicit relation from tag -> category. Do I have to reverse this relation / lose the component-ness to add this composite key?

You don't control :db/id. In some cases, like migrations/restores, it can change, so you should not use it in things like URL's

I see. But Datomic doesn’t have anything like a long sequence generator, right?

I considered that. What I ended up doing was using tuples for session key / value pairs.

random UUIDs are usually good

if you really need a sequence, you could create a tx fn to do it and use an attribute with nohistory

but usually opaque identifiers are best

Yes, I think that an UUID is enough, really. My context here is that I am trying to model using Datomic the (relational) db model of the app that I maintain, and some of its tables has this weird combination of auto-increment long key plus an UUID. Thinking about it now, the pair doesn’t make much sense, the UUID would be enough. I think that the long key exists because of some legacy stuff.

If you end up needing both, you can use a unique tuple

Yes, or you can update this value yourself as a denormalization

Ie make it a heterogeneous tuple value instead of a composite one

I'll try that! The relation never changes in my case, so duping the relation the other way around is not too bad. Was hoping for something automatic, so it wouldn't get out of sync. but oh well

I just started maintaining this system, I think that both keys are unnecessary… I would drop the long key and keep only the UUID if I started it in Clojure/Datomic. 🙂

If I'm deploying an api service using datomic.lambda.api-gateway/ionize can I start a long running listener (like a kafka consumer) in the handler I pass in?