Journal

Notes

• RPM: request per minute.
• /v1/admissions/cohort/all: a light endpoint cached.
• /v1/admissions/syllabus/version: a heavy endpoint cached.

01/12/2024

• [dev] switch WEB_WORKER_CLASS from gevent to uvicorn.workers.UvicornWorker.
• [dev] switch WEB_WORKERS from 1 to 2.
• [dev] switch WEB_WORKER_CONNECTION from 120 to 160.

Side effects:

• Support to async in Django.
• Memory up from 265MB to 437MB.
• /v1/admissions/cohort/all and /v1/admissions/syllabus/version changed from 35 RPM both to 133 RPM #1 and 31 RPM #2, this should change with each attempt.

01/15/2024

• [all] django_minify_html was added to the middlewares.
• [prod] make_charges frequency changed from every 10 minutes to Daily at 12:00 AM UTC.

01/16/2024

• [prod] set CELERY_DYNOS to 2.
• [prod] set CELERY_MAX_WORKERS to 3.
• [dev] switch CELERY_MAX_WORKERS from 1 to 2.

01/17/2024

• [all] STATIC_BUCKET setted.

Side effects:

• /v1/admissions/cohort/all changed from 133 RPM both to 145 - 170 (one time) RPM, this should change with each attempt.

01/24/2024

• [prod] preboot was enabled.
• [prod] switch WEB_WORKER_CLASS from gevent to uvicorn.workers.UvicornWorker.
• [prod] switch LOG_LEVEL from DEBUG to WARNING.
• [prod] set WEB_MAX_REQUESTS to 6000.
• [prod] set WEB_MAX_REQUESTS_JITTER to 3000.
• [dev] set WEB_MAX_REQUESTS to 100.
• [dev] set WEB_MAX_REQUESTS_JITTER to 50.

Side effects:

• In dev, this prevents the server could getting out of memory.
• In prod, this helps to have sufficient memory to add a third worker.
• If the server would have 1000 requests each 10 minutes, this would restart the worker each 60-90 minutes.

01/25/2024

• [prod] switch WEB_MAX_REQUESTS from 6000 to 3000.
• [prod] switch WEB_MAX_REQUESTS_JITTER from 3000 to 1500.

Reasons for the change:

• Web worker was reaching 841 MB ram.

-09/09/2024

• [all] GOOGLE_SECRET setted.
• [dev] GOOGLE_CLIENT_ID setted.
• [all] GOOGLE_REDIRECT_URL setted.

11/09/2024

• [all] GOOGLE_CLIENT_SECRET setted.

Why:

• To enable the oauth flow for the Google Meet API.

30/09/2024

• [prod] set GOOGLE_WEBHOOK_TOPIC to webhook.
• [dev] set GOOGLE_WEBHOOK_TOPIC to webhook-test.

Why:

• Google doesn't support webhooks directly, it uses its Pub/Sub service.

06/10/2024

• [dev] HEROKU_POSTGRESQL_TEAL was replaced by HEROKU_POSTGRESQL_GOLD.

29/01/2025

• [dev] switch CELERY_CONCURRENCY from 8 to 4.

Why:

• To reduce the memory usage of the celery workers.