Recently I had issue with Upstart service. By incident service generated plenty (dozens of megabytes per minute) of logs on STDOUT. After few hours of execution we have seen 100% Init process CPU consumption and memory usage approaching to 2GB! Then application frozen in arbitrary state. It was interesting that last log file size in /var/log/upstart was 2147483646 bytes, that is MAX_INT-1.
Command dmesg returned following:
[73216.335000] init: Failed to write to log file /var/log/upstart/myservice.log [73216.335019] init: Failed to write to log file /var/log/upstart/myservice.log ...
It seems problem is described in https://paperairoplane.net/?p=725. For some reason it froze during log rotation and since then all messages was buffered in init process memory.
So check amount of logs your service produces on STDOUT/STDERR. We just had for debug purposes logs generated by log4j redirected to console too. So we disabled console output for production and problem disappeared.