Once llm fails to output json. Insert infinite loops

[hungrymonkey]

Whenever llm fails to output valid json, the insert infinite loops. We will need a way to skip the record and log it.

[GIN] 2025/07/27 - 11:04:25 | 200 | 11.068352935s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:04:25.445-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1311 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:04:36 | 200 |  11.09455642s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:04:36.545-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1312 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:04:47 | 200 | 11.044513869s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:04:47.595-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1313 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:04:58 | 200 |  11.05047549s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:04:58.655-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1314 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:05:09 | 200 | 11.089884062s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:05:09.755-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1315 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:05:20 | 200 | 11.065728696s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:05:20.907-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1316 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:05:31 | 200 | 11.070242225s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:05:31.983-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1317 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:05:43 | 200 | 11.080259857s |       127.0.0.1 | POST     "/api/chat"
time=2025-07-27T11:05:43.068-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1318 error="failed to parse llm result: unexpected end of JSON input"
[GIN] 2025/07/27 - 11:05:54 | 200 | 11.100380904s |       127.0.0.1 | POST     "/api/chat"

https://github.com/MegaGrindStone/go-light-rag/blob/main/insert.go#L227

Related #4

[hungrymonkey]

What do you recommend as the solution?

[hungrymonkey]

Possible solution

commit c840ad146f329075dcde81d479c3cf33a20de552
Author: Ted Chang <ted@Teds-MacBook-Pro.local>
Date:   Sun Jul 27 14:16:20 2025 -0400

    Resolve this issue
    
    [GIN] 2025/07/27 - 11:05:20 | 200 | 11.065728696s |       127.0.0.1 | POST     "/api/chat"
    time=2025-07-27T11:05:20.907-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1316 error="failed to parse llm result: unexpected end of JSON input"
    [GIN] 2025/07/27 - 11:05:31 | 200 | 11.070242225s |       127.0.0.1 | POST     "/api/chat"
    time=2025-07-27T11:05:31.983-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1317 error="failed to parse llm result: unexpected end of JSON input"
    [GIN] 2025/07/27 - 11:05:43 | 200 | 11.080259857s |       127.0.0.1 | POST     "/api/chat"
    time=2025-07-27T11:05:43.068-04:00 level=WARN msg="Retry parse result" package=golightrag function=Insert retry=1318 error="failed to parse llm result: unexpected end of JSON input"
    [GIN] 2025/07/27 - 11:05:54 | 200 | 11.100380904s |       127.0.0.1 | POST     "/api/chat"

diff --git a/insert.go b/insert.go
index a871f14..aa57262 100644
--- a/insert.go
+++ b/insert.go
@@ -139,10 +139,12 @@ func extractEntities(
 			defer func() { <-sem }()
 
 			// Extract entities and relationships for this source chunk
-			entities, relationships, err := llmExtractEntities(source.Content,
+			entities, relationships, err := llmExtractEntities(source,
 				extractPromptData, llmMaxRetries, llmMaxGleanCount, backoffDuration, llm, logger)
 			if err != nil {
 				return fmt.Errorf("failed to extract entities with LLM: %w", err)
+			} else if len(entities) == 0 && len(relationships) == 0 {
+				return nil
 			}
 
 			logger.Info("Done call LLM", "entities", len(entities), "relationships", len(relationships))
@@ -192,14 +194,14 @@ func removeMarkdownBackticks(input string) string {
 }
 
 func llmExtractEntities(
-	content string,
+	source Source,
 	data EntityExtractionPromptData,
 	maxRetries, maxGleanCount int,
 	backoffDuration time.Duration,
 	llm LLM,
 	logger *slog.Logger,
 ) (map[string][]GraphEntity, map[string][]GraphRelationship, error) {
-	data.Input = content
+	data.Input = source.Content
 	extractPrompt, err := promptTemplate("extract-entities", extractEntitiesPrompt, data)
 	if err != nil {
 		return nil, nil, fmt.Errorf("failed to generate extract entities prompt: %w", err)
@@ -249,6 +251,10 @@ func llmExtractEntities(
 		var sourceParsed llmResult
 		err = json.Unmarshal([]byte(sResult), &sourceParsed)
 		if err != nil {
+			if maxRetries < retry {
+				logger.Info("LLM failed to call source: %s, content: %s", source.ID, source.Content)
+				return map[string][]GraphEntity{}, map[string][]GraphRelationship{}, nil
+			}
 			nErr := fmt.Errorf("failed to parse llm result: %w", err)
 			retry++
 			logger.Warn("Retry parse result", "retry", retry, "error", nErr)
@@ -278,6 +284,10 @@ func llmExtractEntities(
 			var gleanParsed llmResult
 			err = json.Unmarshal([]byte(gResult), &gleanParsed)
 			if err != nil {
+				if maxRetries < retry {
+					logger.Info("LLM failed to call source: %s, content: %s", source.ID, source.Content)
+					return map[string][]GraphEntity{}, map[string][]GraphRelationship{}, nil
+				}
 				nErr := fmt.Errorf("failed to parse llm result: %w", err)
 				retry++
 				logger.Warn("Retry parse result", "retry", retry, "error", nErr)