fix: prevent Ollama indexing from freezing at 69% #6850
Conversation
- Add dynamic timeout for batch embedding operations based on batch size - Improve error handling and recovery in batch processing - Ensure errors are properly thrown to prevent silent failures - Add immediate error state updates when batch processing fails - Add better logging for debugging timeout and retry scenarios Fixes #6849
dosubot
bot
added
size:M
![ellipsis-dev[bot]](https://avatars.githubusercontent.com/in/64358?s=60&v=4){kind=small}
| texts.length > 1 | ||
| ? t("embeddings:ollama.batchTimeoutError", { | ||
| count: texts.length, | ||
| timeout: Math.round(Math.max(OLLAMA_EMBEDDING_TIMEOUT_MS, texts.length * 2000) / 1000), |
There was a problem hiding this comment.
Suggestion: Instead of recalculating the batch timeout for the error message (using Math.max again), reuse the 'batchTimeout' variable calculated earlier to ensure consistency.
![roomote[bot]](https://avatars.githubusercontent.com/in/1546624?s=60&v=4){kind=small}
|
|
||
| // Add timeout to prevent indefinite hanging | ||
| // Use a longer timeout for batch operations as they can take more time | ||
| const batchTimeout = Math.max(OLLAMA_EMBEDDING_TIMEOUT_MS, texts.length * 2000) // At least 2 seconds per text |
There was a problem hiding this comment.
Is the 2-second-per-text multiplier intentional? This seems like a reasonable heuristic, but could we make this configurable or at least document why 2 seconds was chosen? Some models or systems might need different timeout scaling.
| "modelNotEmbeddingCapable": "Ollama model is not embedding capable: {{modelId}}", | ||
| "hostNotFound": "Ollama host not found: {{baseUrl}}" | ||
| "hostNotFound": "Ollama host not found: {{baseUrl}}", | ||
| "batchTimeoutError": "Ollama embedding timed out after {{timeout}} seconds while processing {{count}} texts. Consider reducing batch size or increasing timeout." |
There was a problem hiding this comment.
The new 'batchTimeoutError' translation key is only added to the English locale. Should we add placeholder translations to all other locale files (fr, de, es, ca, hi, id, it, ja, ko, nl, pl, pt-BR, ru, tr, vi) to prevent missing translation errors?
| if (embeddingError.message?.includes("timed out") || embeddingError.message?.includes("timeout")) { | ||
| throw new Error( | ||
| `Embedding timeout for batch of ${batchTexts.length} texts. This may indicate the Ollama service is overloaded or the model is too slow. ${embeddingError.message}`, | ||
| { cause: embeddingError }, |
There was a problem hiding this comment.
When wrapping errors with the 'cause' option, could we lose stack traces in some environments? Consider preserving the original error more explicitly, perhaps by including the original stack in the message or using a custom error class that better preserves debugging information.
| texts.length > 1 | ||
| ? t("embeddings:ollama.batchTimeoutError", { | ||
| count: texts.length, | ||
| timeout: Math.round(Math.max(OLLAMA_EMBEDDING_TIMEOUT_MS, texts.length * 2000) / 1000), |
There was a problem hiding this comment.
The timeout calculation logic appears both here (line 135) and earlier (line 73). Could we extract this to a helper function like 'calculateBatchTimeout(textCount: number)' for better maintainability?
|
|
||
| if (attempts < MAX_BATCH_RETRIES) { | ||
| const delay = INITIAL_RETRY_DELAY_MS * Math.pow(2, attempts - 1) | ||
| console.log(`[DirectoryScanner] Retrying batch processing in ${delay}ms...`) |
There was a problem hiding this comment.
For consistency with other error messages in this file, should this retry log message include the workspace context? Something like: '[DirectoryScanner] Retrying batch processing in ms for workspace ...'
hannesrudolph
added
the
Issue/PR - Triage
|
No repro steps or scope |
Summary
This PR fixes an issue where codebase indexing would freeze at 69% when using Ollama with the mxbai-embed-large model.
Problem
The indexing process was hanging indefinitely due to:
Solution
Changes
Ollama Embedder (
src/services/code-index/embedders/ollama.ts):Scanner (
src/services/code-index/processors/scanner.ts):Orchestrator (
src/services/code-index/orchestrator.ts):Localization (
src/i18n/locales/en/embeddings.json):Testing
Fixes #6849
Important
Fixes indexing freeze at 69% by adding dynamic timeout handling and improving error propagation in Ollama embedder.
ollama.ts.scanner.tsandorchestrator.ts.orchestrator.ts.ollama.ts.ollama.tsandscanner.ts.embeddings.json.scanner.tsandorchestrator.ts.This description was created by
for 2eb26fc. You can customize this summary. It will automatically update as commits are pushed.