Coverage for gco / inference_proxy_config.py: 100.00%

1"""Dependency-light inference proxy autoscaling defaults and rendering.""" 

2 

3from __future__ import annotations 

4 

5from collections.abc import Mapping 

6 

7INFERENCE_PROXY_TLS_CPU_REQUEST_MILLICORES_DEFAULT = 100 

8INFERENCE_PROXY_TLS_CPU_TARGET_UTILIZATION_DEFAULT = 70 

9#: HPA floor and ceiling for the shared proxy Deployment. The floor is also the 

10#: Deployment's create-time ``replicas`` (the HPA owns the count afterwards). 

11INFERENCE_PROXY_MIN_REPLICAS_DEFAULT = 3 

12INFERENCE_PROXY_MAX_REPLICAS_DEFAULT = 10 

13 

14 

15def compute_inference_proxy_tls_replacements( 

16 config: Mapping[str, object], 

17) -> dict[str, str]: 

18 """Render the typed inference-proxy placeholders without importing AWS CDK. 

19 

20 Two shapes matter to the applier and to kubeconform: the TLS CPU request 

21 is a quoted Kubernetes quantity (``"100m"``), while the HPA target and the 

22 replica bounds are bare integers in the manifest. 

23 """ 

24 request = config["tls_proxy_cpu_request_millicores"] 

25 target = config["tls_proxy_cpu_target_utilization_percentage"] 

26 min_replicas = config.get("min_replicas", INFERENCE_PROXY_MIN_REPLICAS_DEFAULT) 

27 max_replicas = config.get("max_replicas", INFERENCE_PROXY_MAX_REPLICAS_DEFAULT) 

28 if ( 

29 type(request) is not int 

30 or type(target) is not int 

31 or type(min_replicas) is not int 

32 or type(max_replicas) is not int 

33 ): 

34 raise ValueError("validated inference proxy TLS settings must be integers") 

35 return { 

36 "{{INFERENCE_PROXY_TLS_CPU_REQUEST}}": f"{request}m", 

37 "{{INFERENCE_PROXY_TLS_CPU_TARGET_UTILIZATION}}": str(target), 

38 "{{INFERENCE_PROXY_MIN_REPLICAS}}": str(min_replicas), 

39 "{{INFERENCE_PROXY_MAX_REPLICAS}}": str(max_replicas), 

40 }