Puppet runs on tools nodes are taking much longer than they should. I think this is a result of growth in tools (adding the Stretch grid) with no accompanying growth in puppetmaster size. tools-puppetmaster-01 is an m1.medium, only 2 CPUs. At any given time it seems to be handling about 10 concurrent puppet runs.
We can ignore this and assume it will go away when we pare down the Trusty grid, or we can build a new bigger puppetmaster, or we can figure out how to properly scale this and have multiple puppetmasters. (I'm not especially in favor of the last option since it will require keeping local hacks in sync)