There was a problem with a caching service which ran out of memory. Now we have increased the memory for this service and will improve our monitoring to avoid this problem in the future.