Hi, Luke!

a nice write up. At certain degree, a simple Linux “side-car” process could put you closer to your ideal solution: launch it at cold start, use local IPC to send your log messages and let this Lambda side car to Kinesis or whatever you need. Your main stream lambda might finish the processing even before all logs are delivered. Since Lambda will stay for a while before being scraped for idle it gives enough time to finish the job. We used this technique for another purpose. Not ideal, but does its job. I’m not sure btw who pays for that side car time while main lambda is out, but so far did not get any eye popping bills, so this direction might be worth further exploration.