Sampling: Trading Completeness for Feasibility¶

At scale, capturing 100% of detailed data is economically impossible; sampling strategies determine which subset provides representative insight at acceptable cost.

Tension Addressed¶

Coverage vs. Cost - sampling is the direct mechanism for managing this trade-off

Recognizable As¶

Any trace/log system not capturing everything; head-based vs. tail-based sampling decisions; sample rates as configuration; "why don't we see all requests"

Canonical Sources¶

Dapper paper (2010); Google SRE Book on sampling strategies

Transfers To¶

Statistics (sample design), Quality assurance (sampling inspection), Scientific research (representative sampling), A/B testing (traffic allocation)

Sub-concepts: - Head-based sampling: Decide at request start (simple, but misses interesting failures) - Tail-based sampling: Decide after completion (captures failures, but requires buffering)