ngl just stumbled onto why tail-based sampling is such a game changer for otel collectors. head-based sampling makes the call way too early, basically deciding to drop a trace b4 any errors or slow calls even happen. its like trying to decide if a movie is bad during the opening credits w/o seeing the plot. with tail-based, you wait until the full context is available so you dont lose those
critical error traces that actually matter for debugging.
it saves so much time on friday nights because you arent hunting for data that was
randomly deleted discarded at the start. has anyone else moved away from head-based entirely, or are you still dealing with the gaps?
full read:
https://dzone.com/articles/tail-based-opentelemetry-collector-sampling