Broken image tags when image is changed with rendition caching enabled

Issue Summary

With rendition caching enabled, changing an image results in broken image tags (old file paths remain cached).

Steps to Reproduce

1. Start a new project with wagtail start myproject
2. Enable rendition caching
3. Edit home/models.py as follows

from django.db import models

from wagtail.core.models import Page
from wagtail.images.edit_handlers import ImageChooserPanel


class HomePage(Page):
    image = models.ForeignKey(
        'wagtailimages.Image', on_delete=models.SET_NULL, related_name='+', null=True, blank=True
    )

    content_panels = Page.content_panels + [
        ImageChooserPanel('image')
    ]

1. Edit home/templates/home/home_page.html as follows

{% extends "base.html" %}
{% load static wagtailimages_tags %}

{% block content %}
{% image page.image width-200 %}
{% endblock content %}

1. Open the page in the frontend, ensure the rendition is cached
2. Change the image in the Wagtail admin (Change image file)
3. Reload the page from step 5, a broken image tag is now rendered because an old image path is still cached

I would expect the cache to be cleared when the image is changed.

• I have confirmed that this issue can be reproduced as described on a fresh Wagtail project: no

Technical details

• Python version: 3.8.2
• Django version: 3.0.5
• Wagtail version: 2.9.0

[tomdyson]

Thank you very much for the report and investigations, @pascalw. I’m sorry for introducing this bug.

I think the cache key can be made robust enough by including the file hash, i.e.:

# images/models.py, line 286
rendition_cache_key = "image-{}-{}-{}-{}".format(
    self.id,
    self.file_hash,
    cache_key,
    filter.spec
)

I’ve tested this successfully with file and focal point changes.

[pascalw]

@tomdyson no need to apologize for bugs! (https://blog.danslimmon.com/2019/08/02/stop-apologizing-for-bugs/)

I’m not sure though that only adding the file_hash to the cache key is a full solution. It indeed will work most of the time, but for example it doesn’t in this case:

1. Upload an image
2. Make sure renditions are cached
3. Change the image (upload a new file)
4. Change the image back to the previous file → Cache is poisoned

Now granted, this is a bit of an edge case but it’s annoying to recover from. Also if in the future new attributes or functionality are added to Images or Renditions, new scenario’s might be introduced where the cache is poisoned in a similar way.

What do you think about using signals to purge the cache when renditions are destroyed or updated? This seems pretty robust and cover all current and future cases, and we don’t have to add a column to the image table.

Something like this:

def post_delete_purge_cache(instance, **kwargs):
    transaction.on_commit(lambda: purge_rendition_cache(instance))

def register():
    Image = get_image_model()
    Rendition = Image.get_rendition_model()

    post_delete.connect(post_delete_purge_cache, sender=Rendition)