getgrav/grav 2.2.29 on GitHub

Log files in logs/ now rotate when they reach 10 MB: grav.log becomes grav.log.1, older copies are gzipped (grav.log.2.gz and on), and five are kept. grav.log, security.log and scheduler.log rotate as they are written, an hourly log-rotate scheduler job catches plugin logs, and bin/grav logrotate runs it by hand. Change or turn it off under system.log.rotate (Configuration → System → Error handler)
A page can unpublish everything below it along with itself. Set cascade_unpublish: true on a page (Options → "Unpublish child pages too"), and while that page is unpublished, by published: false or by its publish_date / unpublish_date, every page below it is unpublished too: left out of menus, collections, the sitemap and search, and not served. The pages below keep their own published setting, so they are back as they were once the parent is published again, and a section given a publish_date goes live all at once. system.pages.cascade_unpublish (Configuration → System → Content) sets the default for every page, and is off. $page->unpublishedBy() returns the route of the page that hides a page

A backup profile's Scheduled Backup switch now decides on its own whether the backup runs. An old Enabled or Disabled entry for the backup job in user/config/scheduler.yaml, which classic admin wrote whenever its Scheduler page was saved, used to override it, so backups could stop with the switch on #4397
A scheduler job that fails now keeps what it printed to standard output as well as its error output, so the reason shows in the job's history, user/data/scheduler/status.yaml and the output file. Most console commands print their errors to standard output, so a failed run used to be recorded with no output at all
A page with a link or image whose address is longer than any file path, such as an inline image written as /data:image/png;base64,..., no longer turns into an error page on servers with open_basedir set
Several Grav sites on one server no longer try to load each other's files. The class loader's APCu cache is now kept separate for each site, so a class found in one site's vendor/ folder is never looked up at another site's path #4392
The Page IDs option and the GFM and table Markdown options in Configuration → System use translation keys instead of English written into the blueprint, so they can be translated. The English comes from Admin2 2.2.13 and Admin 1.10.63