hyf

Context-aware query service for Radroots
git clone https://radroots.dev/git/hyf.git
Log | Files | Refs | README | LICENSE

client.mojo (4096B)

      1 from std.collections import Optional
      2 from std.time import perf_counter_ns
      3 
      4 from json import Value
      5 from flare.http import HttpClient
      6 
      7 from hyf_provider.config import MaxLocalProviderConfig
      8 
      9 
     10 @fieldwise_init
     11 struct MaxLocalTransportResponse(Copyable, Movable):
     12     var status: Int
     13     var body_text: String
     14     var latency_ms: Int
     15 
     16 
     17 @fieldwise_init
     18 struct MaxLocalTransportFailure(Copyable, Movable):
     19     var kind: String
     20     var reason: String
     21 
     22 
     23 @fieldwise_init
     24 struct MaxLocalTransportOutcome(Copyable, Movable):
     25     var response: Optional[MaxLocalTransportResponse]
     26     var failure: Optional[MaxLocalTransportFailure]
     27 
     28 
     29 def _trim_trailing_slash(url: String) -> String:
     30     if url.endswith("/") and url.byte_length() > 1:
     31         return String(url[byte = 0 : url.byte_length() - 1])
     32     return String(url)
     33 
     34 
     35 def _http_url(url: String) -> Bool:
     36     return url.startswith("http://") or url.startswith("https://")
     37 
     38 
     39 def _elapsed_ms_since(start_ns: UInt) -> Int:
     40     return Int((perf_counter_ns() - start_ns) // 1_000_000)
     41 
     42 
     43 def _transport_response_outcome(
     44     status: Int, body_text: String, latency_ms: Int
     45 ) -> MaxLocalTransportOutcome:
     46     return MaxLocalTransportOutcome(
     47         response=Optional[MaxLocalTransportResponse](
     48             MaxLocalTransportResponse(
     49                 status=status,
     50                 body_text=String(body_text),
     51                 latency_ms=latency_ms,
     52             )
     53         ),
     54         failure=Optional[MaxLocalTransportFailure](None),
     55     )
     56 
     57 
     58 def _transport_failure_outcome(
     59     kind: String, reason: String
     60 ) -> MaxLocalTransportOutcome:
     61     return MaxLocalTransportOutcome(
     62         response=Optional[MaxLocalTransportResponse](None),
     63         failure=Optional[MaxLocalTransportFailure](
     64             MaxLocalTransportFailure(
     65                 kind=String(kind), reason=String(reason)
     66             )
     67         ),
     68     )
     69 
     70 
     71 def _transport_exception_reason(
     72     start_ns: UInt, request_timeout_ms: Int
     73 ) -> String:
     74     if _elapsed_ms_since(start_ns) >= request_timeout_ms:
     75         return "timeout"
     76     return "unknown_transport"
     77 
     78 
     79 def make_max_local_http_client(config: MaxLocalProviderConfig) -> HttpClient:
     80     return HttpClient(timeout_ms=config.request_timeout_ms)
     81 
     82 
     83 def max_local_chat_completions_url(config: MaxLocalProviderConfig) -> String:
     84     return _trim_trailing_slash(config.base_url) + "/chat/completions"
     85 
     86 
     87 def get_max_local_health(
     88     config: MaxLocalProviderConfig,
     89 ) -> MaxLocalTransportOutcome:
     90     if not _http_url(config.health_url):
     91         return _transport_failure_outcome("transport", "invalid_url")
     92 
     93     var start_ns = perf_counter_ns()
     94     try:
     95         with make_max_local_http_client(config) as client:
     96             var response = client.get(config.health_url)
     97             var latency_ms = _elapsed_ms_since(start_ns)
     98             if not response.ok():
     99                 return _transport_failure_outcome("http_status", "non_2xx")
    100             return _transport_response_outcome(
    101                 response.status, response.text(), latency_ms
    102             )
    103     except:
    104         return _transport_failure_outcome(
    105             "transport",
    106             _transport_exception_reason(start_ns, config.request_timeout_ms),
    107         )
    108 
    109 
    110 def post_max_local_chat_completion(
    111     config: MaxLocalProviderConfig, body: Value
    112 ) -> MaxLocalTransportOutcome:
    113     var url = max_local_chat_completions_url(config)
    114     if not _http_url(url):
    115         return _transport_failure_outcome("transport", "invalid_url")
    116 
    117     var start_ns = perf_counter_ns()
    118     try:
    119         with make_max_local_http_client(config) as client:
    120             var response = client.post(url, body)
    121             var latency_ms = _elapsed_ms_since(start_ns)
    122             if not response.ok():
    123                 return _transport_failure_outcome(
    124                     "http_status", "provider_non_2xx"
    125                 )
    126             return _transport_response_outcome(
    127                 response.status, response.text(), latency_ms
    128             )
    129     except:
    130         return _transport_failure_outcome(
    131             "transport",
    132             _transport_exception_reason(start_ns, config.request_timeout_ms),
    133         )