Hello,
I am experiencing a reproducible issue with Gemini API File Search.
The problem occurs when registering valid files into an existing File Search Store.
Environment:
- Gemini API / File Search
- Existing File Search Store
- API key is valid and restricted to Gemini API
- Files API upload succeeds and the uploaded file reaches ACTIVE state
Observed behavior:
- Upload the file to Google Files API
- The file becomes ACTIVE
- Call fileSearchStores.importFile
- After about 80 seconds, the request fails with:
HTTP 503 UNAVAILABLE
{
“error”: {
“code”: 503,
“message”: “Failed to count tokens.”,
“status”: “UNAVAILABLE”
}
}
No Operation name is returned.
I also tested the official uploadToFileSearchStore route.
It also fails with HTTP 503 after a long wait.
The affected file is:
- text/plain
- UTF-8
- 113,962 bytes
- no invalid UTF-8
- no NULL bytes
- no obvious control-character corruption
Additional checks:
- countTokens succeeds for the same content
- gemini-embedding-001 embedding succeeds for the same content
- the same File Search Store has successfully accepted other DOCX files
- another File Search Store has successfully accepted a text/plain file of about 1.93 MB
- therefore this does not appear to be a simple file-size limitation
- the official per-document File Search size limit is far above the affected file size
This issue is blocking development and release of a production application because required Knowledge files cannot be registered.
I would like to know:
- Is this a known File Search ingestion issue?
- Is there any known condition that causes “Failed to count tokens” only for specific valid files?
- Is there any workaround that does not require splitting or rewriting the original document?
- Is there an estimated fix timeline if this is a known backend issue?
Thank you.