PoC: Fractal Service (load balancer)

[BobdenOs]

Fractal Service

By tracking database query targets. Create short lived caches (using @cap-js/sqlite) which all instances redirect their queries to. Creating a live cache which can be queried reliably. Greatly reducing the load on the underlying database. While also greatly increasing query response times overall.

With the capability to integrate the load tracking with the BTP load balancer custom metrics. Allowing the creation of new application instances when heavy application loads are detected. Removing the local cache limits inherited from the application memory allocation.

Example

Local SQLite cache vs pure locally hosted Postgres.

queries	fractal(sqlite cache)	postgres
1 cold	5ms	5ms
10 cold	11ms	28ms
10 warm	4ms	~
100 cold	42ms	256ms
100 cold	30ms	~
1000 cold	338ms	2397ms
1000 warm	302ms	~

Diagram

Diagram demonstrating how the queries are tracked and the cache is utilized by multiple application instances.

sequenceDiagram
    Clients->>+CAP1: SELECT
    Clients->>+CAP1: INSERT
    CAP1->>+CAP0: Request Token (2)
    Clients->>+CAP0: SELECT
    CAP0->>CAP0: Request Token (1)
    CAP0->>-CAP1: Grant Ownership
    CAP1->>+DB: SELECT
    DB->>-CAP1: return
    CAP1->>+CAP1: CACHE
    CAP1->>-Clients: return
    CAP0->>+CAP1: SELECT
    CAP1->>-CAP0: return from cache
    CAP1->>-Clients: return
    CAP0->>-Clients: return
    CAP1->>+CAP1: CACHE TIMEOUT
    CAP1->>+DB: FLUSH
    DB->>-CAP1: return
    CAP0->>+CAP0: CACHE TIMEOUT

Loading