1 / 5
Deciding whether to keep a trace at the moment it starts, before knowing its outcome, is ___ sampling.
-
-
-
-
Head-based sampling decides up front (e.g. keep 10% of traces) cheaply, but it can't preferentially keep traces that later turn out to be errors.
2 / 5
Deciding which traces to keep after they complete, so you can retain slow or failed ones, is ___ sampling.
-
-
-
-
Tail-based sampling buffers spans and decides at the end, letting you keep all error/high-latency traces while dropping boring ones.
3 / 5
The fraction of traces retained out of all generated is the ___.
-
-
-
-
The sample rate (e.g. 1%) controls cost and data volume; lower rates save money but reduce visibility into rare events.
4 / 5
A policy that always keeps traces containing an error regardless of base rate is called error ___.
-
-
-
-
Keeping all error traces ensures you never sample away the failures you most need to debug, even at a low overall sample rate.
5 / 5
Keeping each trace with a fixed independent probability is ___ sampling.
-
-
-
-
Probabilistic sampling keeps traces with a set chance (e.g. p=0.1), giving a statistically representative subset at predictable volume.