in reply to Crash report collection package

There are many systems for monitoring, alerting and event analysis and consolidation: both commercial packages and free software. For a large scale distributed system, I would use such a package rather than writing my own.

You need to consider how to send a reasonable number of alerts that come as close to identifying the root cause of the fault as practical. For example, thousands of alerts about every component in your large scale distributed system is not what you want when the problem is one core router or switch that has failed. Letting through the significant alerts and suppressing alerts about consequential faults is a complex problem made easier by the better tools.

You also need to record enough information to be able to find the root cause from your logs.

The better tools allow you to automate response to select events. For example, restarting servers or services.

If you are concerned about security, you might also want indelible logs mirrored to one or more secure systems.