We made sure to add a logger that would help us debug using Cloudwatch. We did all of our debugging using Cloudwatch and we also use Bugsnag to catch any runtime errors in production.

Most of the time, Cloudwatch is not very helpful when tracking down timeouts. We can see that a lambda was run, and eventually when it times out, but other than that, tracking down the strange callback quirks was a team effort of crawling through the code, and if necessary, we enabled our logger on production to see the console logs appear in Cloudwatch.