Language for specifying Predicate Group mapping

Users
1

I was thinking about how we could specify this. Initially, a JSON file could work for us. We specify comma separated predicate prefixes with a group number.

{
     "type.object.name, film.actor.film" : 2,
     "film.performance.film" : 3
}

Internally, these can be mapped to a map[string]int. We can reserve group no. 0 and group no. 1 for the cluster-wide and the default group respectively.

Then we can have the following APIs

  1. getPredicates(group int) - To know which predicates to transfer to a new node which is part of the group.
  2. getGroup(predicate string) - To know which group to send queries and mutations to.

Would love to get your advise on this @mrjn.

2

We’d also need to have the modulos, where we say fp(predicate) % N + k = groupId, for some N and some k. I think for such cases JSON might not be ideal.

Given that we’d be using modulos, we won’t be able to get predicates given a group id. We can only go from predicate → group id, not the other way round.

3

So do we have both the options for the user here? Like he can specify the groupIds or we calculate them using modulos?

4

Yes, we do have both the options. And we process the rules in the order they’re specified. So, if some predicate matches a condition listed higher, that’s the group id it would belong to.

5

Okay, if I understand correctly the N and k would be some sort of globals that he’d specify as part of the config?

A JSON format could still work here, instead of supplying an int, we just need a way for the user to specify that he wants the group to be calculated using modulos. Maybe he can use % .

{
     "default" : {
       "numGroups" : 100,
       "offset": 2
     },
     "predicates" : {
       "type.object.name, film.actor.film" : "2",
       "film.performance.film" : "3",
     }
}
6

More like,

0: _uid_
1: type.object.name.en
2: type.object.name*, film.performance.*
default: fp % 100 + 3

So, we use star as a way to depict prefixes, and use exact matches otherwise. Yeah, I think a JSON would still work.

1 Like
7

Thanks, I will get this done in a bit.

1 Like
8

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.