Timeouts everywhere
If I could give one piece of advice to someone building their first distributed system, it would be to put a timeout on every network call.
Without timeouts, a slow dependency becomes a stuck thread, then a full connection pool, then an outage that looks nothing like its cause. With timeouts, it becomes a clear error you can retry, report or work around.
Choosing the right value is harder than adding one. Start generous, measure, and tighten later.