Scale annotations are leaking

[hvelarde]

I was checking annotations on one cover object and I found the following:

>>> len([k for k in IAnnotations(obj) if k.startswith('plone.tiles.data')])
85
>>> len([k for k in IAnnotations(obj) if k.startswith('plone.tiles.configuration')])
61
>>> len([k for k in IAnnotations(obj) if k.startswith('plone.tiles.scale')])
200

that means we have annotations for 200 scales but only 28 of them seem to be related with existing tiles:

>>> scales = {k.split('.')[3] for k in IAnnotations(obj) if k.startswith('plone.tiles.scale')}
>>> tiles = {k.split('.')[3] for k in IAnnotations(obj) if k.startswith('plone.tiles.data')}
>>> len(scales & tiles)
28

[hvelarde]

I did the following test using collective.revisionmanager:

• calculate statistics and getting information about the object
• purge all but 5 revisions
• recalculate statistics

object size is reduced, but we have the same number of annotations present; this seems to me as an issue with versioning.

[rafaelbco]

The code in the first comment only gets the annotations for the current object. After erasing revisions the object and it's annotations are expected to remain intact. Does that makes sense or am I missing some detail?

[idgserpro]

This is just another example of why #742 is not that trivial...

[hvelarde]

that's right; I think the problem is versioning: versions store their data in annotations on the same base object and, when they are removed, annotations remain in place.

[idgserpro]

Since we don't have a properly CMFEditions configuration, is https://github.com/collective/collective.cover/blob/f66adb50d2834301ff7023a4f1bfe23bd0438039/src/collective/cover/profiles/default/repositorytool.xml really necessary? Do you think it may be the reason we're having this leakage?

[hvelarde]

I have no idea :)

[rafaelbco]

I have an idea, which is only tangentially related to this issue, but it would solve it if
implemented.

I think setups where scales are heavily used (think site front pages) and where a cache frontend is used (Varnish), image scales could be stored only in a RAM cache without losing much performance.

In a RAM cache things expire and are evicted from the cache. So the scales storage should be able to rebuild a requested scale when needed.

I have an idea on how to implement it. If you like this approach I'd appreciate feedback
on this: plone/plone.scale#32

For now I wrote a CMFEditions modifier which prevents image scales from being versioned. I can provide the code if you need it.