As I outlined in this overseer post, we aggregate all crashes so that we know about any serious issue the moment it happens, and we have some consumers that deal with them individually, one of them notifying us on Slack, others logging them, etc.
Because everyone can build and run a consumer, here's one that I 'll personally use: For each abort event (system.apps.abort), a key comprised of the source code file and line where it happened (filename:line) is created.
A lookup on a CloudDS 'table' checks if we have encountered that before; if not, then it's set there and a nice Wunderlist task, complete with stacktrace frames, node name, and more, is created and placed in a special list, via RPC.