.

2025-12-27 23:42:13 +05:00
parent e8b99c1fe5
commit dd1bed2536
2 changed files with 54 additions and 20 deletions
@@ -44,6 +44,9 @@ public class OpenAIAnalyticsService {
    @Value("${openai.timeoutMs:90000}")
    private long timeoutMs;

+    @Value("${openai.timeoutMs.instruction:180000}")
+    private long timeoutMsInstruction;
+
    @Value("${openai.retry.maxAttempts:5}")
    private int maxRetryAttempts;

@@ -76,7 +79,8 @@ public class OpenAIAnalyticsService {
    private String openaiUrl;

    /**
-     * Helper for building OpenAI chat-completions request body (extracted for unit testing).
+     * Helper for building OpenAI chat-completions request body (extracted for unit
+     * testing).
     */
    Map<String, Object> buildChatCompletionsRequestBody(
            String model,
@@ -92,7 +96,8 @@ public class OpenAIAnalyticsService {
    }

    /**
-     * Helper for building chat messages list (system + user) (extracted for unit testing).
+     * Helper for building chat messages list (system + user) (extracted for unit
+     * testing).
     */
    List<Map<String, Object>> buildChatMessages(String userPrompt, String systemPrompt) {
        List<Map<String, Object>> messages = new ArrayList<>();
@@ -237,6 +242,7 @@ public class OpenAIAnalyticsService {
                        .retrieve()
                        .bodyToMono(new ParameterizedTypeReference<Map<String, Object>>() {
                        })
+                        .timeout(Duration.ofMillis(timeoutMs))
                        .retryWhen(createRetrySpec("generate"))
                        .onErrorResume(err -> {
                            logger.error("OpenAI request failed after retries: {} - {}", err.getMessage(),
@@ -251,10 +257,12 @@ public class OpenAIAnalyticsService {
                                } else if (wcre.getStatusCode().value() >= 500) {
                                    logger.error("OpenAI API server error after all retry attempts.");
                                }
+                            } else if (err instanceof java.util.concurrent.TimeoutException) {
+                                logger.error("OpenAI API request timed out after {}ms", timeoutMs);
                            }
                            return Mono.empty();
                        })
-                        .block(Duration.ofMillis(timeoutMs));
+                        .block(Duration.ofMillis(timeoutMs + 10000)); // Add buffer for block timeout

                if (response == null) {
                    return null;
@@ -282,7 +290,12 @@ public class OpenAIAnalyticsService {
            logger.warn("Interrupted while waiting for OpenAI request slot: {}", e.getMessage());
            return null;
        } catch (Exception e) {
-            logger.warn("Error calling OpenAI generate: {}", e.getMessage());
+            // .block() throws IllegalStateException on timeout, not TimeoutException
+            if (e.getMessage() != null && e.getMessage().contains("Timeout")) {
+                logger.warn("Timeout waiting for OpenAI response (block timeout): {}", e.getMessage());
+            } else {
+                logger.warn("Error calling OpenAI generate: {} - {}", e.getMessage(), e.getClass().getSimpleName());
+            }
            return null;
        }
    }
@@ -312,12 +325,15 @@ public class OpenAIAnalyticsService {
    /**
     * Генерирует текст с поддержкой system prompt и override для max_tokens.
     *
-     * @param text              Текст (контекст), который будет добавлен к instruction
+     * @param text              Текст (контекст), который будет добавлен к
+     *                          instruction
     * @param instruction       Основная инструкция для модели
-     * @param language          Язык ответа (исторически передается, но не влияет на протокол)
+     * @param language          Язык ответа (исторически передается, но не влияет на
+     *                          протокол)
     * @param modelName         Название модели OpenAI
     * @param systemPrompt      System prompt (если null/blank — не добавляется)
-     * @param maxTokensOverride Override для max_tokens (если null — используется дефолт)
+     * @param maxTokensOverride Override для max_tokens (если null — используется
+     *                          дефолт)
     * @return Сгенерированный текст
     */
    public String generateWithInstructionWithModel(
@@ -346,7 +362,8 @@ public class OpenAIAnalyticsService {
            }

            try {
-                logger.debug("Calling OpenAI API with model: {}", modelName);
+                logger.debug("Calling OpenAI API with model: {} (timeout: {}ms)", modelName, timeoutMsInstruction);
+                // Use timeout operator in the reactive chain so timeouts can be retried
                Map<String, Object> response = this.webClient.post()
                        .contentType(MediaType.APPLICATION_JSON)
                        .accept(MediaType.APPLICATION_JSON)
@@ -354,6 +371,7 @@ public class OpenAIAnalyticsService {
                        .retrieve()
                        .bodyToMono(new ParameterizedTypeReference<Map<String, Object>>() {
                        })
+                        .timeout(Duration.ofMillis(timeoutMsInstruction))
                        .retryWhen(createRetrySpec("generateWithInstruction"))
                        .onErrorResume(err -> {
                            logger.error("OpenAI request failed after retries: {} - {}", err.getMessage(),
@@ -368,10 +386,12 @@ public class OpenAIAnalyticsService {
                                } else if (wcre.getStatusCode().value() >= 500) {
                                    logger.error("OpenAI API server error after all retry attempts.");
                                }
+                            } else if (err instanceof java.util.concurrent.TimeoutException) {
+                                logger.error("OpenAI API request timed out after {}ms", timeoutMsInstruction);
                            }
                            return Mono.empty();
                        })
-                        .block(Duration.ofMillis(timeoutMs));
+                        .block(Duration.ofMillis(timeoutMsInstruction + 10000)); // Add buffer for block timeout

                if (response == null) {
                    return null;
@@ -399,7 +419,13 @@ public class OpenAIAnalyticsService {
            logger.warn("Interrupted while waiting for OpenAI request slot: {}", e.getMessage());
            return null;
        } catch (Exception e) {
-            logger.warn("Error calling OpenAI generate (generic): {}", e.getMessage());
+            // .block() throws IllegalStateException on timeout, not TimeoutException
+            if (e.getMessage() != null && e.getMessage().contains("Timeout")) {
+                logger.warn("Timeout waiting for OpenAI response (block timeout): {}", e.getMessage());
+            } else {
+                logger.warn("Error calling OpenAI generate (generic): {} - {}", e.getMessage(),
+                        e.getClass().getSimpleName());
+            }
            return null;
        }
    }
@@ -407,11 +433,11 @@ public class OpenAIAnalyticsService {
    private RetryBackoffSpec createRetrySpec(String operation) {
        // Store Retry-After duration for 429 errors
        AtomicReference<Duration> retryAfterDuration = new AtomicReference<>(null);
-        
+
        // Use the higher max attempts to handle rate limit errors better
        // This will apply to all retries, but rate limit errors need more attempts
        int effectiveMaxAttempts = Math.max(maxRetryAttempts, rateLimitMaxAttempts);
-        
+
        return Retry.backoff(effectiveMaxAttempts, Duration.ofMillis(initialRetryDelayMs))
                .maxBackoff(Duration.ofMillis(maxRetryDelayMs))
                .multiplier(retryMultiplier)
@@ -419,7 +445,7 @@ public class OpenAIAnalyticsService {
                    if (throwable instanceof WebClientResponseException) {
                        WebClientResponseException wcre = (WebClientResponseException) throwable;
                        int statusCode = wcre.getStatusCode().value();
-                        
+
                        if (statusCode == 429) {
                            // Extract Retry-After header for 429 errors
                            String retryAfterHeader = wcre.getHeaders().getFirst("Retry-After");
@@ -442,22 +468,28 @@ public class OpenAIAnalyticsService {
                        } else {
                            retryAfterDuration.set(null);
                        }
-                        
+
                        // Only retry on 429 (rate limit) and 5xx (server errors)
                        return statusCode == 429 || statusCode >= 500;
                    }
                    // Retry on network errors (timeouts, connection issues)
                    retryAfterDuration.set(null);
-                    return throwable instanceof java.util.concurrent.TimeoutException
+                    boolean isTimeout = throwable instanceof java.util.concurrent.TimeoutException
+                            || (throwable.getMessage() != null && throwable.getMessage().contains("Timeout"));
+                    if (isTimeout) {
+                        logger.warn("OpenAI API timeout detected for operation '{}'. Will retry.", operation);
+                    }
+                    return isTimeout
                            || throwable instanceof java.net.ConnectException
                            || throwable instanceof java.io.IOException;
                })
                .doBeforeRetry(retrySignal -> {
                    long attempt = retrySignal.totalRetries() + 1;
                    Throwable failure = retrySignal.failure();
-                    
+
                    // If we have a Retry-After duration for 429 errors, sleep for that duration
-                    // This ensures we respect the API's requested delay before Reactor applies its backoff
+                    // This ensures we respect the API's requested delay before Reactor applies its
+                    // backoff
                    Duration customDelay = retryAfterDuration.get();
                    if (customDelay != null && failure instanceof WebClientResponseException) {
                        WebClientResponseException wcre = (WebClientResponseException) failure;
@@ -474,14 +506,15 @@ public class OpenAIAnalyticsService {
                            }
                        }
                    }
-                    
+
                    // Log retry attempt
                    if (failure instanceof WebClientResponseException) {
                        WebClientResponseException wcre = (WebClientResponseException) failure;
                        int statusCode = wcre.getStatusCode().value();
                        int maxAttempts = statusCode == 429 ? rateLimitMaxAttempts : maxRetryAttempts;
                        if (statusCode == 429 && customDelay == null) {
-                            logger.warn("OpenAI API returned 429 for operation '{}'. Will retry with exponential backoff (attempt {}/{})",
+                            logger.warn(
+                                    "OpenAI API returned 429 for operation '{}'. Will retry with exponential backoff (attempt {}/{})",
                                    operation, attempt, maxAttempts);
                        } else if (statusCode != 429) {
                            logger.warn("OpenAI API returned {} for operation '{}'. Will retry (attempt {}/{})",
@@ -506,5 +539,5 @@ public class OpenAIAnalyticsService {
                    return retrySignal.failure();
                });
    }
-    
+
 }
@@ -64,6 +64,7 @@ openai.api.url=https://api.openai.com/v1/chat/completions
 openai.model.name=gpt-4o-mini
 openai.model.name.text=gpt-4o
 openai.timeoutMs=90000
+openai.timeoutMs.instruction=180000
 openai.retry.maxAttempts=3
 openai.retry.initialDelayMs=2000
 openai.retry.maxDelayMs=60000