Multi-hop query support

[em3s]

Background

EdgeIndex uses narrow rows, requiring N scan API calls for the 2nd hop in multi-hop queries. A wide row structure is introduced to enable multi-get and resolve this. A process to prevent fat rows is also introduced alongside.

Extending from independent per-feature edge queries to cross-feature graph traversal (Unified Graph).

Task

• Phase 0: Design alignment with core model via skeleton E2E
  [0/4] Multi-hop query: skeleton E2E #217
• Phase 1: Mutation + multi-get query to verify (not n-hop yet)
  [1/4] Multi-hop query: Mutation + multi-get query #226
• Phase 2: Query - 2-hop or more
  [2/4] Multi-hop query: Query (2-hop or more) #243
• Phase 3: Server with E2E from Phase 0
  [2/4] Multi-hop query: Query (2-hop or more) #243
• Phase 4: Pruner (fat row prevention)

Done When

• Design alignment based on skeleton
• 2-hop query works in 1+1 API calls
• E2E tests pass

Notes

• Implement n-hop generalization

[zipdoki]

@em3s
I think I can start with a PR from Phase 0. I will let you know once the PR is open!

[zipdoki]

@em3s
The Phase 0 PR is up: #217
It introduces the skeleton E2E for the multi-hop query path along with the EdgeCache model.

[zipdoki]

@em3s
In #217 , I used the term Cache for a wide row structure. However, as you may know, EdgeCache was already being used as a model, so I'd like to rename it to avoid confusion. I'm planning to replace Cache with Memoize.

I'll start working on this. If you have a different opinion, feel free to chime in!

[em3s]

@zipdoki

To clarify, I wasn't referring to the model name. I was talking about the variable name within ActionbaseQuery that determines whether or not to use the cache.

actionbase/engine/src/main/kotlin/com/kakao/actionbase/v2/engine/query/ActionbaseQuery.kt

Lines 25 to 29 in 8b013f4

	sealed class Item {
	abstract val name: String
	abstract val include: Boolean
	abstract val cache: Boolean
	abstract val post: List<PostProcessor>

[zipdoki]

@em3s
Ah, got it. That was my misunderstanding. Thanks for clarifying!

[em3s]

Sharing the results from our offline meeting. @zipdoki

1. Align requirements based on E2E tests — [0/4] Multi-hop query: skeleton E2E #217
2. Migrate ActionbaseQuery to the engine package
   • 1st version: current ActionbaseQuery in the v2 package
   • 2nd version: com.kakao.actionbase.core.java.query.Query
   • Target: implement the union of both versions and move to the engine package
   • No existing usages, so backward compatibility is not a concern
3. Pruner as an API (2 PRs)
   • PR 1: Pruner API implementation
   • PR 2: Pipeline code — extracts pruning candidates via WAL/CDC and calls the Pruner API

[zipdoki]

@em3s
Phase 2 is up for review in #243, including what we discussed offline.