[lstamour]

This reminds me of https://docs.google.com/a/gravitant.com/document/d/199PqyG3U... on how you should only wake up engineers when there really is a problem. I'd suggest logging based on error messages -- though I get it, if a problem occurs upstream, you wouldn't know it unless you'd polled for it too, as a data point. HN comments on that doc at: https://news.ycombinator.com/item?id=8450147