docs: README /metrics + MAX_TOKENS, smoke-test health/metrics/timing checks

README.md

@@ -10,7 +10,23 @@ OpenAI-compatible LLM inference server using `llama-cpp-2` (Rust).
 
 ### `GET /health`
 ```json
-{"status": "ok", "model": "minicpm5-1b-fable5-v2-thinking"}
+{"status": "ok", "model": "minicpm5-1b-fable5-v2-thinking", "uptime_s": 1234, "n_ctx": 8192, "version": "0.1.0"}
+```
+
+### `GET /metrics`
+Prometheus text exposition (no auth) — request counters, token usage, generation
+latency/throughput, process uptime:
+
+```
+llm_api_requests_total            # total /v1/chat/completions
+llm_api_errors_total              # errored requests
+llm_api_streaming_requests_total  # stream: true requests
+llm_api_aborted_requests_total    # aborted generations (client disconnect)
+llm_api_prompt_tokens_total       # prompt tokens accepted
+llm_api_completion_tokens_total   # tokens generated
+llm_api_generation_ms_total       # generation time (ms)
+llm_api_tokens_per_second         # lifetime throughput gauge
+llm_api_build_info{version,model} # identity
 ```
 
 ### `GET /v1/models`
@@ -53,6 +69,7 @@ MODEL_PATH=/path/to/model.gguf ./target/release/llm-api
 | `SERVER_PORT` | `4010` | Listen port |
 | `RUST_LOG` | `info` | Log level |
 | `N_CTX` / `N_BATCH` / `N_THREADS` | `8192` / `512` / `4` | llama.cpp context/batch/threads |
+| `MAX_TOKENS` | `2048` | Hard cap untuk `max_tokens` request (0 = unlimited) |
 
 ### Smoke test (setelah deploy)
 

scripts/smoke-test.sh

@@ -32,16 +32,25 @@ check() {
   fi
 }
 
-echo "== 1. Health =="
+echo "== 1. Health ==="
 health=$(curl -sf "$BASE_URL/health") || { echo "FAIL: /health unreachable"; exit 1; }
 echo "$health" | jq -e ".status == \"ok\" and .model == \"$MODEL_ID\"" >/dev/null
 check "health melaporkan $MODEL_ID" $?
+echo "$health" | jq -e '.uptime_s >= 0 and (.version | type == "string")' >/dev/null
+check "health punya uptime + version" $?
 
 echo "== 2. Models =="
 models=$(curl -sf "$BASE_URL/v1/models")
 echo "$models" | jq -e ".data[0].id == \"$MODEL_ID\"" >/dev/null
 check "models mencantumkan $MODEL_ID" $?
 
+echo "== 2b. Metrics =="
+metrics=$(curl -sf "$BASE_URL/metrics") || { echo "FAIL: /metrics unreachable"; exit 1; }
+echo "$metrics" | grep -q "llm_api_requests_total"
+check "metrics punya llm_api_requests_total" $?
+echo "$metrics" | grep -q "llm_api_build_info{"
+check "metrics punya llm_api_build_info" $?
+
 echo "== 3. Chat non-streaming =="
 resp=$(curl -sf "${AUTH[@]}" -H "Content-Type: application/json" \
   -d "{\"model\":\"$MODEL_ID\",\"messages\":[{\"role\":\"user\",\"content\":\"Say hi\"}],\"max_tokens\":64}" \
@@ -50,6 +59,8 @@ echo "$resp" | jq -e '.choices[0].message.content | type == "string"' >/dev/null
 check "non-streaming mengembalikan content" $?
 echo "$resp" | jq -e '.usage.total_tokens > 0' >/dev/null
 check "non-streaming berisi usage" $?
+echo "$resp" | jq -e '.usage.duration_ms > 0 and .usage.tokens_per_second > 0' >/dev/null
+check "non-streaming berisi timing (duration_ms + tok/s)" $?
 
 echo "== 4. Chat streaming =="
 stream=$(curl -sfN "${AUTH[@]}" -H "Content-Type: application/json" \