You are designing for a write-heavy application. During testing, you discover that the write workloads are performant in a regional Cloud Spanner instance but slow down by an order of magnitude in a…

Question

Options

• APlace the bulk of the read and write workloads closer to the default leader region.
• BUse staleness of at least 15 seconds.
• CAdd more read-write replicas.
• DKeep the total CPU utilization under 45% in each region.

• A
  84% (27)
• B
  9% (3)
• C
  3% (1)
• D
  3% (1)

Explanation

In a multi-regional Cloud Spanner instance, all write transactions must achieve a quorum commit, which requires the default leader region to participate. If write requests originate far from the leader region, they incur significant network round-trip latency before being committed. Placing the application (and thus the bulk of read/write workloads) in the same or a nearby region as the default leader dramatically reduces this latency. Using read staleness (B) helps reads avoid the leader but has no effect on write latency. Adding more read-write replicas (C) increases replication overhead and can actually slow down writes. Keeping CPU under 45% (D) is a general best-practice threshold to avoid hotspots but does not address the fundamental cross-region write latency problem.

No community discussion yet for this question.