Skip to content

Add Django: the most popular Python web framework (~82k ⭐)#71

Merged
MDA2AV merged 4 commits intoMDA2AV:mainfrom
BennyFranciscus:add-django
Mar 24, 2026
Merged

Add Django: the most popular Python web framework (~82k ⭐)#71
MDA2AV merged 4 commits intoMDA2AV:mainfrom
BennyFranciscus:add-django

Conversation

@BennyFranciscus
Copy link
Copy Markdown
Collaborator

@BennyFranciscus BennyFranciscus commented Mar 18, 2026

Django — the web framework for perfectionists with deadlines

Adds Django (~82k stars) to HttpArena.

Why Django?

Django is THE most popular Python web framework — period. With ~82k GitHub stars, it's the backbone of Instagram, Pinterest, Mozilla, Disqus, and countless production applications. HttpArena already has Flask (micro, sync) and FastAPI (async, modern) — Django completes the Python trinity.

The matchup everyone wants to see:

  • Flask — minimalist WSGI micro-framework
  • FastAPI — modern async ASGI framework
  • Django — full-featured batteries-included framework

Django carries more weight than Flask (ORM, middleware, URL resolver, settings system) but is incredibly well-optimized after 20 years of production use. This benchmark strips Django to the essentials — no middleware, no ORM — to see how the core request/response pipeline performs.

Setup

  • Django 5.2 on Gunicorn with sync workers (same WSGI setup as Flask for a fair comparison)
  • Workers: 2× CPU cores
  • Minimal settings: no middleware, no installed apps, no debug
  • Pre-computed JSON + gzip for /json and /compression endpoints
  • Thread-local SQLite connections with mmap for /db

Endpoints

Endpoint Method Description
/pipeline GET Returns ok
/baseline11 GET/POST Sum query params (+ body for POST)
/baseline2 GET Sum query params (H2)
/json GET 50-item JSON with computed totals
/compression GET Gzip-compressed large JSON
/db GET SQLite range query
/upload POST Returns byte count

Tests subscribed

baseline, pipelined, noisy, limited-conn, json, upload, compression, mixed

cc @carltongibson @felixxm @sarahboyce — thought it'd be cool to see how Django stacks up in HttpArena! Flask vs FastAPI vs Django is the Python comparison everyone's been waiting for.

@BennyFranciscus BennyFranciscus requested a review from MDA2AV as a code owner March 18, 2026 02:05
@BennyFranciscus
Fix chunked POST body reading and add method restrictions
24413b6 
- Use request.read() for POST body to handle chunked Transfer-Encoding
- Add @require_GET and @require_http_methods decorators to reject unsupported methods with 405
@BennyFranciscus
Fix chunked POST: read wsgi.input directly when no Content-Length
2313c7d 
Django's request.body reads CONTENT_LENGTH bytes from the stream.
With chunked Transfer-Encoding, there's no Content-Length header,
so Django reads 0 bytes. Read from wsgi.input directly as fallback.
@MDA2AV
Copy link
Copy Markdown
Owner

MDA2AV commented Mar 23, 2026

/benchmark

@BennyFranciscus
Copy link
Copy Markdown
Collaborator Author

BennyFranciscus commented Mar 23, 2026

Benchmark run triggered for django — all profiles. Results incoming 🚀

Django's a big framework with a lot of middleware by default, so the numbers won't be in Flask territory — but that's kind of the point. Good to have the baseline for the most popular Python framework.

@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented Mar 23, 2026

🚀 Benchmark run triggered for django (all profiles). Results will be posted here when done.

@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented Mar 23, 2026

Benchmark Results

Framework: django | Profile: all profiles

django / baseline / 512c (p=1, r=0, cpu=unlimited)
  Best: 113306 req/s (CPU: 5751.4%, Mem: 3.9GiB) ===

django / baseline / 4096c (p=1, r=0, cpu=unlimited)
  Best: 93810 req/s (CPU: 6564.3%, Mem: 7.6GiB) ===

django / baseline / 16384c (p=1, r=0, cpu=unlimited)
  Best: 36216 req/s (CPU: 1677.1%, Mem: 4.1GiB) ===

django / pipelined / 512c (p=16, r=0, cpu=unlimited)
  Best: 121307 req/s (CPU: 5619.2%, Mem: 3.8GiB) ===

django / pipelined / 4096c (p=16, r=0, cpu=unlimited)
  Best: 114552 req/s (CPU: 4966.7%, Mem: 3.5GiB) ===

django / pipelined / 16384c (p=16, r=0, cpu=unlimited)
  Best: 42010 req/s (CPU: 2526.2%, Mem: 7.3GiB) ===

django / limited-conn / 512c (p=1, r=10, cpu=unlimited)
  Best: 113431 req/s (CPU: 5706.7%, Mem: 3.6GiB) ===

django / limited-conn / 4096c (p=1, r=10, cpu=unlimited)
  Best: 109055 req/s (CPU: 5268.0%, Mem: 4.0GiB) ===

django / json / 4096c (p=1, r=0, cpu=unlimited)
  Best: 115456 req/s (CPU: 5049.6%, Mem: 3.8GiB) ===

django / json / 16384c (p=1, r=0, cpu=unlimited)
  Best: 36492 req/s (CPU: 1655.6%, Mem: 8.4GiB) ===

django / upload / 64c (p=1, r=0, cpu=unlimited)
  Best: 0 req/s (CPU: 0%, Mem: 0MiB) ===
Full log 
  Bandwidth:  845.68MB/s
  Status codes: 2xx=459482, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 459477 / 459482 responses (100.0%)
  Reconnects: 459664
  CPU: 6104.7% | Mem: 7.4GiB

[run 3/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/json
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   24.39ms   22.60ms   24.60ms   27.70ms   436.30ms

  927271 requests in 5.00s, 462392 responses
  Throughput: 92.44K req/s
  Bandwidth:  851.03MB/s
  Status codes: 2xx=462392, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 462392 / 462392 responses (100.0%)
  Reconnects: 462555
  Errors: connect 0, read 3, timeout 0
  CPU: 5762.7% | Mem: 8.4GiB

=== Best: 115456 req/s (CPU: 5049.6%, Mem: 3.8GiB) ===
[dry-run] Results not saved (use --save to persist)
httparena-bench-django
httparena-bench-django

==============================================
=== django / json / 16384c (p=1, r=0, cpu=unlimited) ===
==============================================
5c01feb0916817d322714743dd133836f56c6edd8a1146987ae9e3bc27002f36
[wait] Waiting for server...
[ready] Server is up

[run 1/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/json
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  1
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   72.81ms   46.90ms   74.90ms   484.60ms    3.40s

  363119 requests in 5.03s, 177301 responses
  Throughput: 35.27K req/s
  Bandwidth:  324.87MB/s
  Status codes: 2xx=177301, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 177301 / 177301 responses (100.0%)
  Reconnects: 177450
  Errors: connect 0, read 50, timeout 0
  CPU: 1526.9% | Mem: 3.8GiB

[run 2/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/json
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  1
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   29.68ms   21.80ms   37.00ms   149.20ms   355.80ms

  70073 requests in 5.00s, 33437 responses
  Throughput: 6.68K req/s
  Bandwidth:  61.52MB/s
  Status codes: 2xx=33437, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 33437 / 33437 responses (100.0%)
  Reconnects: 33437
  CPU: 838.6% | Mem: 7.4GiB

[run 3/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/json
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  1
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   68.47ms   53.00ms   91.70ms   508.50ms    1.73s

  358377 requests in 5.03s, 183559 responses
  Throughput: 36.53K req/s
  Bandwidth:  336.52MB/s
  Status codes: 2xx=183559, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 183559 / 183559 responses (100.0%)
  Reconnects: 183764
  Errors: connect 0, read 118, timeout 0
  CPU: 1655.6% | Mem: 8.4GiB

=== Best: 36492 req/s (CPU: 1655.6%, Mem: 8.4GiB) ===
[dry-run] Results not saved (use --save to persist)
httparena-bench-django
httparena-bench-django

==============================================
=== django / upload / 64c (p=1, r=0, cpu=unlimited) ===
==============================================
34697be830360c991901b773915c3443a2d95f9770f34102e9cfda152b2993d7
[wait] Waiting for server...
[ready] Server is up

[run 1/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     64 (1/thread)
  Pipeline:  1
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   5.68ms   4.46ms   8.36ms   16.20ms   116.80ms

  81786 requests in 5.00s, 16663 responses
  Throughput: 3.33K req/s
  Bandwidth:  1.03MB/s
  Status codes: 2xx=0, 3xx=0, 4xx=16663, 5xx=0
  Latency samples: 16663 / 16663 responses (100.0%)
  Reconnects: 81797
  Errors: connect 0, read 80283, timeout 0

  WARNING: 16663/16663 responses (100.0%) had unexpected status (expected 2xx)
  CPU: 4648.2% | Mem: 3.6GiB

[run 2/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     64 (1/thread)
  Pipeline:  1
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   5.49ms   4.07ms   9.84ms   22.10ms   103.60ms

  58614 requests in 5.00s, 34655 responses
  Throughput: 6.92K req/s
  Bandwidth:  2.14MB/s
  Status codes: 2xx=0, 3xx=0, 4xx=34655, 5xx=0
  Latency samples: 34655 / 34655 responses (100.0%)
  Reconnects: 58626
  Errors: connect 0, read 53374, timeout 0

  WARNING: 34655/34655 responses (100.0%) had unexpected status (expected 2xx)
  CPU: 8948.2% | Mem: 7.4GiB

[run 3/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     64 (1/thread)
  Pipeline:  1
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   5.40ms   4.30ms   10.40ms   19.90ms   43.40ms

  53353 requests in 5.01s, 40048 responses
  Throughput: 8.00K req/s
  Bandwidth:  2.47MB/s
  Status codes: 2xx=0, 3xx=0, 4xx=40048, 5xx=0
  Latency samples: 40048 / 40048 responses (100.0%)
  Reconnects: 53361
  Errors: connect 0, read 47683, timeout 0

  WARNING: 40048/40048 responses (100.0%) had unexpected status (expected 2xx)
  CPU: 9703.6% | Mem: 8.5GiB

=== Best: 0 req/s (CPU: 0%, Mem: 0MiB) ===
httparena-bench-django
httparena-bench-django
[restore] Restoring CPU governor to powersave...

@BennyFranciscus
Copy link
Copy Markdown
Collaborator Author

BennyFranciscus commented Mar 23, 2026

Django results are in — and honestly, pretty respectable for a full-featured framework running on Gunicorn:

Highlights:

  • 113K baseline at 512c — not bad at all for Django with all its middleware stack. Gunicorn workers are doing solid work here
  • 121K pipelined — only a ~7% bump from pipelining, which makes sense since Gunicorn is process-based and each worker handles one request at a time
  • 115K JSON — serialization performance is close to baseline, so the JSON overhead is minimal
  • Memory: 3.9GB baseline — this is the multi-process model showing up. Each Gunicorn worker is a full Python process with its own memory space

Interesting patterns:

  • Steep drop at 16384c (113K → 36K) — that's the process-based concurrency model hitting its limits. With a fixed worker pool, high connection counts mean lots of waiting
  • Upload returning 4xx across all runs — looks like the upload endpoint config needs a look. Django's CSRF middleware and file upload handling probably need specific configuration for the benchmark's upload format
  • Limited-conn profile (113K at 512c, 109K at 4096c) shows Django handles connection reuse efficiently — almost no degradation when connections are reused

Context:
For comparison, Flask (also Python/WSGI) hits ~92K baseline. Django coming in 23% higher is interesting — could be the optimized ORM-less path or just Gunicorn tuning differences. Either way, Django's "batteries included" philosophy doesn't cost as much throughput as people assume.

The upload issue should probably be fixed before merging — want me to take a look at the Django config?

@BennyFranciscus
fix(django): stream upload from wsgi.input, remove body size limit
a0b33cd 
- Upload endpoint now reads from wsgi.input in 64KB chunks instead of
  using request.body which fails on chunked Transfer-Encoding (no
  CONTENT_LENGTH header)
- Set DATA_UPLOAD_MAX_MEMORY_SIZE = None to allow 20MB benchmark uploads
- Same streaming pattern used in baseline11 for POST bodies
@BennyFranciscus
Copy link
Copy Markdown
Collaborator Author

BennyFranciscus commented Mar 24, 2026

Fixed the upload 4xx issue! Two problems:

  1. request.body fails on chunked transfers — the benchmark sends uploads without Content-Length, and Django's request.body checks CONTENT_LENGTH meta before reading. Switched to streaming from wsgi.input directly in 64KB chunks (same pattern as the baseline11 POST handler).

  2. DATA_UPLOAD_MAX_MEMORY_SIZE default is 2.5MB — the benchmark sends 20MB files, which would hit Django's built-in limit and return 400. Set to None to remove the cap.

Push is up (a0b33cd) — should be ready for a re-benchmark whenever you get a chance!

@MDA2AV
Copy link
Copy Markdown
Owner

MDA2AV commented Mar 24, 2026

/benchmark

@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented Mar 24, 2026

🚀 Benchmark run triggered for django (all profiles). Results will be posted here when done.

@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented Mar 24, 2026

Benchmark Results

Framework: django | Profile: all profiles

django / baseline / 512c (p=1, r=0, cpu=unlimited)
  Best: 111648 req/s (CPU: 5470.6%, Mem: 3.5GiB) ===

django / baseline / 4096c (p=1, r=0, cpu=unlimited)
  Best: 103106 req/s (CPU: 4567.3%, Mem: 3.9GiB) ===

django / baseline / 16384c (p=1, r=0, cpu=unlimited)
  Best: 39940 req/s (CPU: 2273.2%, Mem: 3.8GiB) ===

django / pipelined / 512c (p=16, r=0, cpu=unlimited)
  Best: 115630 req/s (CPU: 5336.9%, Mem: 3.7GiB) ===

django / pipelined / 4096c (p=16, r=0, cpu=unlimited)
  Best: 106666 req/s (CPU: 4273.0%, Mem: 3.7GiB) ===

django / pipelined / 16384c (p=16, r=0, cpu=unlimited)
  Best: 35090 req/s (CPU: 1635.4%, Mem: 3.6GiB) ===

django / limited-conn / 512c (p=1, r=10, cpu=unlimited)
  Best: 110381 req/s (CPU: 5588.6%, Mem: 3.9GiB) ===

django / limited-conn / 4096c (p=1, r=10, cpu=unlimited)
  Best: 105221 req/s (CPU: 5165.5%, Mem: 3.8GiB) ===

django / json / 4096c (p=1, r=0, cpu=unlimited)
  Best: 102649 req/s (CPU: 4348.9%, Mem: 3.6GiB) ===

django / json / 16384c (p=1, r=0, cpu=unlimited)
  Best: 34133 req/s (CPU: 1618.3%, Mem: 3.9GiB) ===

django / upload / 64c (p=1, r=0, cpu=unlimited)
  Best: 813 req/s (CPU: 7547.6%, Mem: 8.4GiB) ===

django / upload / 256c (p=1, r=0, cpu=unlimited)
  Best: 873 req/s (CPU: 11433.9%, Mem: 8.4GiB) ===

django / upload / 512c (p=1, r=0, cpu=unlimited)
  Best: 862 req/s (CPU: 11306.4%, Mem: 8.4GiB) ===

django / compression / 4096c (p=1, r=0, cpu=unlimited)
  Best: 50295 req/s (CPU: 4264.2%, Mem: 3.5GiB) ===

django / compression / 16384c (p=1, r=0, cpu=unlimited)
  Best: 28275 req/s (CPU: 2105.8%, Mem: 8.7GiB) ===

django / noisy / 512c (p=1, r=0, cpu=unlimited)
  Best: 20 req/s (CPU: 338.3%, Mem: 3.6GiB) ===

django / noisy / 4096c (p=1, r=0, cpu=unlimited)
  Best: 20 req/s (CPU: 253.4%, Mem: 7.6GiB) ===

django / noisy / 16384c (p=1, r=0, cpu=unlimited)
  Best: 51 req/s (CPU: 43.2%, Mem: 8.4GiB) ===

django / mixed / 4096c (p=1, r=5, cpu=unlimited)
  Best: 66962 req/s (CPU: 10165.5%, Mem: 7.2GiB) ===

django / mixed / 16384c (p=1, r=5, cpu=unlimited)
  Best: 39279 req/s (CPU: 4128.7%, Mem: 8.5GiB) ===
Full log 
  Per-template-ok: 129,130,0,0,0

  WARNING: 130/389 responses (33.4%) had unexpected status (expected 2xx)
  CPU: 43.2% | Mem: 8.4GiB

=== Best: 51 req/s (CPU: 43.2%, Mem: 8.4GiB) ===
  Input BW: 5.28KB/s (avg template: 106 bytes)
[dry-run] Results not saved (use --save to persist)
httparena-bench-django
httparena-bench-django

==============================================
=== django / mixed / 4096c (p=1, r=5, cpu=unlimited) ===
==============================================
a34537fc629843e96f2d13dabbff03d40b158f5e3f9cb5f2063dd3522798f613
[wait] Waiting for server...
[ready] Server is up

[run 1/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  5
  Templates: 10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   37.29ms   30.80ms   45.90ms   120.90ms   853.30ms

  596611 requests in 5.00s, 297488 responses
  Throughput: 59.46K req/s
  Bandwidth:  3.15GB/s
  Status codes: 2xx=297488, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 297506 / 297488 responses (100.0%)
  Reconnects: 298390
  Errors: connect 0, read 22455, timeout 0
  Per-template: 29720,29795,29855,29932,29993,30043,30074,28906,29556,29612
  Per-template-ok: 29720,29795,29855,29932,29993,30043,30074,28906,29556,29612
  CPU: 5973.5% | Mem: 3.8GiB

[run 2/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  5
  Templates: 10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   34.23ms   31.40ms   36.60ms   47.70ms   452.40ms

  673091 requests in 5.00s, 334813 responses
  Throughput: 66.93K req/s
  Bandwidth:  3.38GB/s
  Status codes: 2xx=334813, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 334813 / 334813 responses (100.0%)
  Reconnects: 335504
  Errors: connect 0, read 18156, timeout 0
  Per-template: 33434,33511,33565,33612,33701,33708,33751,33015,33263,33253
  Per-template-ok: 33434,33511,33565,33612,33701,33708,33751,33015,33263,33253
  CPU: 10165.5% | Mem: 7.2GiB

[run 3/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  5
  Templates: 10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   37.58ms   34.20ms   39.30ms   49.40ms   855.70ms

  628445 requests in 5.00s, 313569 responses
  Throughput: 62.68K req/s
  Bandwidth:  3.16GB/s
  Status codes: 2xx=313569, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 313569 / 313569 responses (100.0%)
  Reconnects: 314155
  Errors: connect 0, read 17299, timeout 0
  Per-template: 31310,31368,31408,31449,31512,31573,31627,30990,31206,31126
  Per-template-ok: 31310,31368,31408,31449,31512,31573,31627,30990,31206,31126
  CPU: 9471.7% | Mem: 8.4GiB

=== Best: 66962 req/s (CPU: 10165.5%, Mem: 7.2GiB) ===
  Input BW: 6.54GB/s (avg template: 104924 bytes)
[dry-run] Results not saved (use --save to persist)
httparena-bench-django
httparena-bench-django

==============================================
=== django / mixed / 16384c (p=1, r=5, cpu=unlimited) ===
==============================================
7ee08a5211e37e72ec634b93c0f566a2698b1c30d86054ac77e18ae4afb095b5
[wait] Waiting for server...
[ready] Server is up

[run 1/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  1
  Req/conn:  5
  Templates: 10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   89.82ms   71.80ms   99.60ms   486.70ms    1.73s

  267884 requests in 5.02s, 132164 responses
  Throughput: 26.34K req/s
  Bandwidth:  1.25GB/s
  Status codes: 2xx=132164, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 132164 / 132164 responses (100.0%)
  Reconnects: 132400
  Errors: connect 0, read 5128, timeout 0
  Per-template: 13163,13187,13242,13251,13286,13256,13354,13180,13184,13061
  Per-template-ok: 13163,13187,13242,13251,13286,13256,13354,13180,13184,13061
  CPU: 2498.8% | Mem: 4.0GiB

[run 2/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  1
  Req/conn:  5
  Templates: 10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   70.07ms   50.70ms   88.10ms   467.50ms    3.36s

  389292 requests in 5.02s, 192310 responses
  Throughput: 38.29K req/s
  Bandwidth:  1.76GB/s
  Status codes: 2xx=192310, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 192310 / 192310 responses (100.0%)
  Reconnects: 193782
  Errors: connect 0, read 5370, timeout 0
  Per-template: 19177,19222,19265,19324,19348,19389,19485,18843,19221,19036
  Per-template-ok: 19177,19222,19265,19324,19348,19389,19485,18843,19221,19036
  CPU: 5017.1% | Mem: 7.6GiB

[run 3/3]
gcannon — io_uring HTTP load generator
  Target:    localhost:8080/
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  1
  Req/conn:  5
  Templates: 10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   68.15ms   52.50ms   89.50ms   448.30ms    1.77s

  395661 requests in 5.02s, 197181 responses
  Throughput: 39.27K req/s
  Bandwidth:  1.83GB/s
  Status codes: 2xx=197181, 3xx=0, 4xx=0, 5xx=0
  Latency samples: 197181 / 197181 responses (100.0%)
  Reconnects: 198034
  Errors: connect 0, read 5938, timeout 0
  Per-template: 19566,19687,19776,19839,19856,19868,19985,19593,19635,19376
  Per-template-ok: 19566,19687,19776,19839,19856,19868,19985,19593,19635,19376
  CPU: 4128.7% | Mem: 8.5GiB

=== Best: 39279 req/s (CPU: 4128.7%, Mem: 8.5GiB) ===
  Input BW: 3.84GB/s (avg template: 104924 bytes)
[dry-run] Results not saved (use --save to persist)
httparena-bench-django
httparena-bench-django
[skip] django does not subscribe to baseline-h2
[skip] django does not subscribe to static-h2
[skip] django does not subscribe to baseline-h3
[skip] django does not subscribe to static-h3
[skip] django does not subscribe to unary-grpc
[skip] django does not subscribe to unary-grpc-tls
[skip] django does not subscribe to echo-ws
[restore] Restoring CPU governor to powersave...

@BennyFranciscus
Copy link
Copy Markdown
Collaborator Author

BennyFranciscus commented Mar 24, 2026

New results look good! The upload fix is working — went from 0 req/s (all 4xx) to ~870 req/s across concurrency levels. Not blazing fast, but Django is doing real work there with the streaming read + chunked transfer handling.

Baseline numbers are consistent with the first run (~111K vs ~113K at 512c — well within normal variance).

Quick summary of where Django lands:

Profile Best
Baseline 512c 111K
Pipelined 512c 115K
JSON 4096c 102K
Upload 256c 873
Compression 4096c 50K
Mixed 4096c 67K

The noisy profile is rough (20 req/s) but that's Django's middleware stack + GIL fighting the background CPU load — kind of expected for a sync WSGI framework. Flask will probably look similar there.

I think this is ready for a proper review whenever you get a chance 👍

@MDA2AV MDA2AV merged commit f853cff into MDA2AV:main Mar 24, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@MDA2AV MDA2AV Awaiting requested review from MDA2AV MDA2AV is a code owner

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@BennyFranciscus @MDA2AV