Добавил логи

internal/llm/openai.go

@@ -6,6 +6,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
+	"log/slog"
 	"net/http"
 	"net/url"
 	"strings"
@@ -26,10 +27,11 @@ type openAICompat struct {
 	apiKey    string
 	model     string
 	retryWait time.Duration // базовая пауза между ретраями (0 в тестах)
+	log       *slog.Logger
 }
 
 // newOpenAICompat собирает клиент из конфига.
-func newOpenAICompat(cfg Config) (*openAICompat, error) {
+func newOpenAICompat(cfg Config, logger *slog.Logger) (*openAICompat, error) {
 	if cfg.BaseURL == "" {
 		return nil, fmt.Errorf("llm: empty base_url")
 	}
@@ -51,12 +53,16 @@ func newOpenAICompat(cfg Config) (*openAICompat, error) {
 		transport = &http.Transport{Proxy: http.ProxyURL(proxyURL)}
 	}
 
+	if logger == nil {
+		logger = slog.Default()
+	}
 	return &openAICompat{
 		endpoint:  strings.TrimRight(cfg.BaseURL, "/") + "/chat/completions",
 		hc:        &http.Client{Timeout: timeout, Transport: transport},
 		apiKey:    cfg.APIKey,
 		model:     cfg.Model,
 		retryWait: baseRetryWait,
+		log:       logger,
 	}, nil
 }
 
@@ -119,15 +125,30 @@ func (c *openAICompat) Complete(ctx context.Context, req Request) (Response, err
 			}
 		}
 
+		c.log.Debug("llm: request",
+			"endpoint", c.endpoint, "model", c.model,
+			"attempt", attempt, "max_attempts", maxAttempts)
+		start := time.Now()
 		resp, retryable, err := c.do(ctx, body)
 		if err == nil {
+			c.log.Debug("llm: response ok",
+				"model", resp.Model, "attempt", attempt,
+				"duration", time.Since(start),
+				"total_tokens", resp.Usage.TotalTokens, "cost", resp.Usage.Cost)
 			return resp, nil
 		}
 		lastErr = err
 		if !retryable {
+			c.log.Error("llm: request failed (non-retryable)",
+				"model", c.model, "attempt", attempt, "duration", time.Since(start), "err", err)
 			return Response{}, err
 		}
+		c.log.Warn("llm: request failed, will retry",
+			"model", c.model, "attempt", attempt, "max_attempts", maxAttempts,
+			"duration", time.Since(start), "err", err)
 	}
+	c.log.Error("llm: all attempts exhausted",
+		"model", c.model, "max_attempts", maxAttempts, "err", lastErr)
 	return Response{}, fmt.Errorf("llm: exhausted %d attempts: %w", maxAttempts, lastErr)
 }