Introduce safeguards for the image processing

[amureki]

Greetings and thanks for the library!

We encountered memory issues when django-pictures processed large images in production with our setup:

# settings.py
PICTURES = {
    "BREAKPOINTS": {
        "mobile": 576,
        "desktop": 1024,
    },
    "FILE_TYPES": ["AVIF", "WEBP"],
}

# models.py
logo = PictureField(
    "logo",
    blank=True,
    aspect_ratios=[None],
    validators=[
        FileExtensionValidator(["avif", "png", "webp"]),
    ],
    height_field="logo_height",
    width_field="logo_width",
)

A 15000x8000px image (the file was only 481KB PNG) generated 180+ variants, consuming ~1.5GB+ memory and causing our Heroku 2x dynos (1GB RAM) to crash.

We also noticed PIL's DecompressionBombWarning in the logs.

I am preparing some safe-guards within my codebase to prevent uploading such images, but I wonder if we can make it friendlier within the library. For example:

1. Add optional built-in validators that could be opt-in by default:

• Maximum pixel count (e.g., 10MP limit)
• Maximum dimensions (e.g., 4096×4096)
• File size limits

2. Instead of failing on large images, offer auto-resize:

• Resize source image to reasonable dimensions before processing
• Maintain aspect ratio
• Configurable resize thresholds

3. Processing Configuration

• Timeout settings for long-running tasks
• Memory-aware processing options

Also, maybe we could document the server requrements better (not sure if this makes sense, thus asking):

• Memory requirements based on expected image sizes
• CPU recommendations for different workloads
• Use dedicated workers for image processing
• Configure isolated queues to prevent blocking other operations

Cheers,
Rust

[codingjoe]

Hi @amureki,

This package actually does ship with a set of validators, like a MaxSizeValidator.
However, I just noticed that they are not documented yet 😳
Do you care to have the honors and vide code a little documentation for them?

Additionally, we could consider a byte-size validator. However, the compressed file size and the memory required to process said image are not closely related.

Pre-processing doesn't make sense. This package is already pretty memory efficient, and it would be fairly unintuitive to add edge case handling above a certain image dimension.

Best
Joe