Engineers are well-tuned with debugging issues on a single machine. However, when the architecture scales out to possibly hundreds or thousands of machines with components 10+ layers deep, debugging doesn't look the same anymore. The concept of looking at logs becomes 'collective' in nature and looking for patterns in logs is the only viable way of associating them with the problems you are trying to solve.
We will walk through motivation for building such a system and how it differs from traditional monitoring and debugging. A system designed this way collects all needed artifacts, identifies known/unknown patterns in error messages, correlates with infrastructure serving these errors, and allows outlier service components to be exposed within 10-15 minutes of a developing problem trend.