What happens when cron jobs fail depends on the type of failure. Errors occurring within the module pointed to by the cron job will propagate to the cron job. If the module referred to by the URL of the cron job is unavailable or otherwise unresponsive, I believe a 5xx HTTP response is returned.
Regarding your questions of Datastore records processed during a cron job, this is not connected to cron jobs but rather the module processing those tasks, the speed/size of Datastore retrievals, the concurrency of requests being made, etc. If the cron jobs are outlasting the 10-minute limitation described in the '
Scheduled Tasks' article, you may want to study the performance of the module being used to find some possible optimizations. To help with this analysis, you may want to consider using
Cloud Trace. Though in Beta, it can reveal great detail on where your bottleneck can be found.
For instance,
bulk operations may be used to consolidate Datastore operations to reduce retrieve more records per request. If this is already being used, you could consider using
memcache instead to contain this data temporarily until sent to Cloud Monitoring. The linked
Best Practices article can be very helpful in this case showing how to handle checks, sets and batch operations. Otherwise, you may consider configuring the cron job module to scale with more instances or more powerful tiers. An introduction to scaling for App Engine can be found with the article '
How Apps Scale'.
Hope this helps.