Page MenuHomePhabricator

Request increase to Google TTS rate limits
Closed, ResolvedPublic1 Estimated Story Points

Description

From https://cloud.google.com/text-to-speech/quotas:

Type of LimitUsage Limit
Total characters per request5,000
Requests per minute1,000
Characters per minute500,000

We can request an increase to the request/character rate limits β€” we should probably rate limit requests ourselves & queue requests which would go over?

Event Timeline

TheresNoTime renamed this task from Google TTS rate limits to Google TTS rate limits [tbd].Aug 23 2022, 1:39 PM
TheresNoTime renamed this task from Google TTS rate limits [tbd] to Request increase to Google TTS rate limits.Aug 28 2022, 11:53 PM
TheresNoTime updated the task description. (Show Details)

Do we want to proactively request an increase now, independent of T316006: Handle missing files?

Sounds like we should! Moving it into sprint 33

Estimating a "requests per minute" limit in order to provide a new quota limit is fairly difficult β€” if, on deployment, x Phonos audio render requests are generated and added to the job queue:

  • how quickly will the Job Queue attempt to run through them? [0]
  • when the limit is hit, what is the starting retry delay? [1]
  • how aggressively does the retry delay ramp?

Depending on the outcome of T317131: [8 hours] Investigate JobQueue and the job processing rate, I'm leaning towards picking a high number (10k?) and letting the retries handle propagating out the initial generations. I think @MusikAnimal suggested we reach out to Google and see if they'd consider giving us a "burst" quota to handle these initial spikes in (re)generation.


[0] https://grafana.wikimedia.org/d/LSeAShkGz/jobqueue?orgId=1&viewPanel=20 suggests ~220/min averaged over 5m
[1] https://wikitech.wikimedia.org/wiki/Kafka_Job_Queue#Architecture_overview

The quota increase request includes a text field marked "Your description will be sent to your service provider and is used to evaluate your request. It's useful to include the intent of the quota usage, future growth plans, region or zone spread, and any additional requirements or dependencies. " β€” I'm happy to write this, and would primarily be giving an overview of the Phonos project (linking to https://meta.wikimedia.org/wiki/Community_Wishlist_Survey_2022/Generate_Audio_for_IPA ?) and touching on the possibility of reaching out separately to discuss burst quotas. Is there anything specific I should mention and/or should someone else be writing such a request? Perhaps @NRodriguez or @KSiebert?

The quota increase request includes a text field marked "Your description will be sent to your service provider and is used to evaluate your request. It's useful to include the intent of the quota usage, future growth plans, region or zone spread, and any additional requirements or dependencies. " β€” I'm happy to write this, and would primarily be giving an overview of the Phonos project (linking to https://meta.wikimedia.org/wiki/Community_Wishlist_Survey_2022/Generate_Audio_for_IPA ?) and touching on the possibility of reaching out separately to discuss burst quotas. Is there anything specific I should mention and/or should someone else be writing such a request? Perhaps @NRodriguez or @KSiebert?

Maybe the partnership team could/should handle the request? They might have done similar things in the past.

Maybe the partnership team could/should handle the request? They might have done similar things in the past.

Happy to hand it over to whomever 😸

I wrote Nicholas from Partnerships and will report back.

TheresNoTime changed the task status from Open to Stalled.Sep 20 2022, 11:02 AM

Request made for an increase from 1000 to 10,000 β€” ref #30176934

Last update from Google was 2022-10-02 β€” the request was still in review

Just received this from Google: "Your quota request has been approved, and your project quota has been adjusted accordingly."

Logging in, I can confirm that our quota is now:

  • Count of characters in all requests per minute: 500,000
  • All requests per minute: 10,000
TheresNoTime set the point value for this task to 1.Oct 5 2022, 8:21 AM