Remove outdated items from the catalog
Add collective.catalogcleanup to the eggs in your buildout (and to zcml on Plone 3.2 or earlier). This makes a browser view available on the Plone Site root: @@collective-catalogcleanup.
This goes through the portal_catalog and removes all catalog brains for which a getObject call does not work. In other words, it removes brains that no longer belong to an actual object in the site.
Similar cleanups are done for the uid_catalog and the reference_catalog.
The goal is to get rid of outdated brains that could otherwise cause problems, for example during an upgrade to Plone 4.
@@collective-catalogcleanup by default does a dry run, so it only reports problems. Call it with @@collective-catalogcleanup?dry_run=false to perform the actual cleanup.
So what does the catalog cleanup do?
It removes stuff! You must make a backup first!
It handles these catalogs: portal_catalog, uid_catalog, reference_catalog.
For each catalog it reports the number of catalog brains it contains.
It removes brains that have a UID of None.
It removes brains of which the object is broken. This can happen when the object belongs to a package that is no longer available in the Plone Site.
It removes brains of which the object cannot be found.
It looks for non unique uids. There can be some legitimate reasons why some brains may have the same UID, for example when they belong to comments: the UID is inherited from the parent object. Those items are kept. For other items we accept one object and we give the other objects a new UID.
References between objects that no longer exist or are broken, will be removed.
A simple report will be printed in the browser. For one catalog it may look like this:
Handling catalog portal_catalog. Brains in portal_catalog: 20148 portal_catalog: removed 25 brains without UID. portal_catalog: removed 100 brains with status broken. portal_catalog: removed 5 brains with status notfound. portal_catalog: 249 non unique uids found. portal_catalog: 249 items given new unique uids.
The instance log may contain more info, about individual items.
- A clear and rebuild of the portal_catalog should have partially the same effect, but it will likely take a lot longer and it will not solve some of the problems mentioned above. But this is definitely the most logical thing to try before giving collective.catalogcleanup a go.
I have tried this on Plone 3.3, Plone 4 and Plone 5.
It is automatically tested by Travis on Plone 4.3, 5.0, and 5.1, all on Python 2.7.
- Catch TypeError when getting object for brain. Can happen when an object that used to be referenceable is no longer referenceable. Fixes issue #19. [maurits]
- Disable CSRF protection. Fixes issue #17. [maurits]
- Abort any transaction changes in dry run mode. There should not be any changes here anyway, but this makes sure. [maurits]
- No longer test on Plone 4.1 and 4.2 and on Python 2.6. [maurits]
- Catch KeyError and AttributeError for getPath in more cases. Fixes issue #14. [maurits]
- Added traceback info to help in case of problems. [maurits]
- Tested for compatibility on Plone 4.0 through 5.1. [hvelarde]
- Ignore non existing catalogs. Plone 5 does not always have a uid_catalog or reference_catalog. Fixes issue #5. [maurits]
- Don’t look for non unique ids in the reference_catalog. It looks like it is normal there. At least, on one Plone 4.3 site the code keeps creating several new uids every time I run it. [maurits]
- Don’t complain about brains in reference_catalog where getObject returns None. This happens for content without apparent problems. [maurits]
- Do not complain about brains in uid_catalog that are references. When their path points to ...at_references/<uid of brain> then this is normal. I started wondering about a site that had more than 20 thousand problems reported this way. [maurits]
- Remove all items that have the portal_factory folder in their path. [maurits]
- Catch KeyErrors when getting the path of a brain. [maurits]
- Give less confusing message for comments that inherit the UID of their parent. It sounded too much like an error. [maurits]
- Improved the cleanup of non unique uids. [maurits]
- When doing an reindexObject, only reindex the UID. [maurits]
- Initial release [maurits]