Better Django worker
This article is based fundamentally on which option is better in efficiency and features, we are covering three libraries because they were popular choices by the community.
Collected workers
Gevent
Gevent patches many system libraries to become cooperative and it does that the worker works as if this were asynchronous but over WSGI, but this does not allow running asynchronous code.
Uvicorn
Uvicorn is an ASGI implementation for Python, it is known as a faster worker until now.
Hypercorn
Hypercorn is an ASGI implementation for Python, it is a good option compared with Uvicorn.
Source of Truth
I needed to see how should impact the blocking code to the web worker, I wrote Gevent vs Uvicorn vs Hypercorn to cover this purpose, this is the next charset of Requests vs HTTPX vs AIOHTTP.
Source
Result
Which was the conclusions
Should be fine read this first, it was the results:
Gevent:
- Stability: good
- Blocking code: it is bad, it gets worse the longer it lasts.
- Non-blocking code: it is the better option.
- Async support: no.
- Best HTTP sync client: requests, HTTPX performs bad, maybe its performance is degraded by the patch.
- Best HTTP async client: not compatible.
Uvicorn:
- Statibility: good
- Blocking code: Good.
- Non-blocking code: regular.
- Async support: yes.
- Best HTTP sync client: requests, HTTPX performs bad, maybe its performance is degraded by the patch.
- Best HTTP async client: AIOHTTP is good.
Hypercorn + Asyncio:
- Stability: bad, I could configure it properly.
- Blocking code: weak.
- Non-blocking code: regular.
- Async support: yes.
- Best HTTP sync client: requests, HTTPX performs bad, maybe its performance is degraded by the patch.
- Best HTTP async client: AIOHTTP is good.
Hypercorn + Evloop:
- Stability: bad, I could configure it properly.
- Blocking code: Best performance, with fails.
- Non-blocking code: regular.
- Async support: yes.
- Best HTTP sync client: requests, HTTPX perform bad, maybe its performance is degraded by the patch.
- Best HTTP async client: AIOHTTP is good.
Hypercorn + Trio:
- Stability: no supported yet
Notes
- RPS are requests per second.
- Requests do not support asynchronous code, so you cannot use wait with it.
Making requests within the web worker
| Library | Gevent | Uvicorn | Hypercorn + Asyncio | Hypercorn + Evloop |
|---|---|---|---|---|
| Sync JSON | 30913.89 RPS | 4061.19 RPS | 3753.90 RPS | 4929.08 RPS |
| Async JSON | 3464.25 RPS | 3270.73 RPS | 4533.80 RPS | |
| Sync query JSON | 11498.22 RPS | 1845.01 RPS | 1691.03 RPS | 1969.74 RPS |
| Async query JSON | 1997.32 RPS | 1565.80 RPS | 2148.56 RPS | |
| Sync query HTML | 7113.50 RPS | 1590.24 RPS | 1189.28 RPS | 1577.94 RPS |
| Async query HTML | 1749.29 RPS | 1382.19 RPS | 1792.08 RPS | |
| Sync blocking 1s | 1861.80 RPS | 1819.99 RPS | 1747.01 RPS | 1769.52 RPS |
| Async blocking 1s | 1805.06 RPS | 1676.20 RPS | 1712.17 RPS | |
| Sync blocking 3s | 635.14 RPS | 631.95 RPS | 585.55 RPS | 578.60 RPS |
| Async blocking 3s | 1592.22 RPS | 1476.54 RPS | 1395.52 RPS | |
| Sync blocking 10s | 181.38 RPS | 181.37 RPS | 181.43 RPS | 152.51 RPS |
| Async blocking 10s | 90.72 RPS | 37.97 RPS | 29.12 RPS | |
| Sync Brotli | 414.45 RPS | 295.72 RPS | 314.35 RPS | 324.34 RPS |
| Async Brotli | 158.04 RPS | 142.45 RPS | 193.18 RPS | |
| Sync Request | 38.50 RPS | 25.08 RPS | 22.04 RPS | 16.01 RPS |
| Async Request | 0.00 RPS | 0.00 RPS | 0.00 RPS | |
| Sync HTTPX | 0.50 RPS | 0.00 RPS | 0.00 RPS | 0.00 RPS |
| Async HTTPX | 42.81 RPS | 17.74 RPS | 17.42 RPS | |
| Async AIOHTTP | 35.07 RPS | 23.40 RPS | 45.29 RPS | |
| Sync Fake Redis hit | 26658.00 RPS | 5654.22 RPS | 3251.49 RPS | 4327.30 RPS |
| Async Fake Redis hit | 6295.06 RPS | 3735.66 RPS | 4035.76 RPS | |
| Sync Fake Redis set | 9669.09 RPS | 4366.79 RPS | 3236.40 RPS | 3676.91 RPS |
| Async Fake Redis set | 3806.02 RPS | 3399.52 RPS | 3713.74 RPS |