Page tree
Skip to end of metadata
Go to start of metadata

How Personalization and Cache work, in a few words


  • Out of the box with 5.3: cache is bypassed for pages with variants. (we removed the contention issues that can occur when doing header negotiation).
    • When installing the Personalization module, your cache configuration will be automatically adapted (if it is using the default cache policy) to use a new CachePolicy and Store CacheExecutor, which simply result in bypassing the cache for pages which have variants or are personalized. (info.magnolia.personalization.cache.BypassUncacheableEntriesPolicy and info.magnolia.personalization.cache.BypassVariantsCacheStore respectively)
  • Possible: customize cache key to include your traits
    • We can’t make that work for everyone without a lot of effort (would need to be configurable, magic, or dumbly add all traits), 
    • We don’t believe this is a good solution: it only works if a majority of pages are personalized AND you have a small amount of traits AND those traits have a small amount of possible values. (It would be completely pointless with a “date” trait, and even just 2 traits with 2 possible values would increase the cache size by 4)
  • Future: we will implement another caching mechanism which should allow caching pages with variations while maintaining a reasonably sized cache.
    • It will require for trait detectors to be executed before the cache filter, which means performance of these detectors will be paramount to this mechanism working properly.
    • Tech details: cache policy will work with an extra indirection. See sequence diagram below.
  • Future: we are investigating component-level caching and Edge Side Includes. See Concept - partial caching.
  • No labels

4 Comments

  1. if resolving variants before accessing cache, we should also avoid touching underlying repo as much as possible and will probably have to keep in-memory cache for most info that variants resolver needs, similar to what we do in MultiSiteFilter for resolving sites.

    1. That's exactly the purpose of VariantRegistry, and/or another cache. (see the note in the diagram)

      1. Hmm, I see. Wasn't that clear when I first looked at the diagram. I guess it's the VariantRegistry (Cache') name that confused me. Would it be cache-backed-VariantRegistry or is VariantRegistry shown in the diagram the actual cache of variants?

        1. It should read "Variant Registry or Cache-prime (e.g another cache) depending on what we find out while implementing" (i.e I'm not sure the VariantRegistry's API will be adequate to get the List<Rules> from the current AggState, because AggState doesn't know about a page's uuid yet, so we might have to use a cache on top of it)