Three days debugging a missing trace

Last week we had a deploy on Suga that was taking longer than usual. It was hard to reproduce, so the only obvious thing to do was try and trace it end to end. We immediately ran into roadblocks because the trace only covered the first 140 milliseconds. For context: the bulk of our deployment code runs in a background worker and is triggered via a queue, and the worker's trace had no connection back to the request that kicked it off. Without a complete trace, we were forced to look at everything