Allow max requests per minute in Gemini

[dtvc87]

It is very easy to reach the max requests per minute. It would be good to have a setting to change this. Another option, which I'm not sure if it is the current behaviour, is to keep retrying. I can see some settings about retries, but in the UI it says that translation failed. So I'm not sure if it is actually still trying to make it --just waiting to retry again and continue where it left it in the previous max request error-- or if it actually failed.

[rowanfuchs]

Currently there is a Delay between retries and a Retry delay multiplier with that setting you can estimate the retry per minute. I do am looking into enhancing this functionality with more options.

[rowanfuchs]

This error occurs when it indeed exceeds the "free" quota, if it still is there you should try chunking it in batches but this gets unreliable quickly.

[dtvc87]

Right, but looking at the tiers, Gemini 2.5 Flash-Lite on tier 1 has a quota of 15 requests per minute. It looks like the error is because I was exceeding the 15 requests per minute. So I guess requesting again in 1 minute should work, until it reaches the 1000 requests per day. This might be a WAI, but looking at the error message, I thought it would retry again doing exponential backoff.

[swavans]

It doesn't seem to respect the retry delay/interval, I have set max_concurrent_jobs=1 in environmental variable and the following settings for the interval/retry:

But still the job fails after a couple of seconds.

Ideally the retry would parse/respect the retryinseconds from the 429 response.

[swavans]

The retry loop isn't triggering for gemini specifically as the

lingarr/Lingarr.Server/Services/Translation/GoogleGeminiService.cs

Line 170 in 7154d9a

private async Task<string> TranslateWithGeminiApi(string? message, CancellationToken cancellationToken)

throws a TranslateException while the catch expects an HttpException with statuscode 409.

I could try to create a merge request tomorrow if you'd like? Same for the batch

[rowanfuchs]

The retry loop isn't triggering for gemini specifically as the

lingarr/Lingarr.Server/Services/Translation/GoogleGeminiService.cs

Line 170 in 7154d9a

private async Task<string> TranslateWithGeminiApi(string? message, CancellationToken cancellationToken)

throws a TranslateException while the catch expects an HttpException with statuscode 409.
I could try to create a merge request tomorrow if you'd like?

Good catch! feel free to create one when you have time, thank you for looking into this.

[swavans]

#292 Merge for the bug is ready, I'll see if I can add a retry based on the response later this week ;)

[swavans]

From the logs it seems to be working as intended now

[rowanfuchs]

From the logs it seems to be working as intended now

cheers! thank you very much

[swavans]

Dove a bit further into this, for the free tier it might be better to increase the timeout so the whole sub file can be processed in one go, as the daily limit per model seems to be 20 requests. I see there is already a setting for it, is somebody working on exposing that to the UI?

[rowanfuchs]

hasn't been a priority, but if it's exposed to the ui it's important it works on all services that have such a feature.