retry-amplification · unknown · also request retry amplification

Retry amplification

Retry amplification is the additional request-processing load created by repeated attempts to complete logical operations.

Related Observatory article: Retry amplification on hesela.com

Delayed replies can induce more attempts and further delay, creating a sustaining feedback loop. This is distinct from NAND write amplification. [1]

Repeated storage-server membership queries can saturate the metadata service that supplies them. This is a service recovery mechanism, not evidence that every retry causes an outage. [2]

A chain of retrying callers can multiply total attempts. Count the initial attempt separately from retries and name the layer where attempts are measured. A per-operation limit differs from an aggregate retry budget. [3]

Relations

related

Metastable failure, Retry deduplication, I/O timeout, Retry budget

Citations

1. Bronson et al. Metastable Failures in Distributed Systems. HotOS 2021; DOI 10.1145/3458336.3465286, sections 1-3. (accessed 2026-10-20)
2. AWS. DynamoDB service disruption and related impacts in US-East: 20 September 2015 incident. (accessed 2026-10-20)
3. Mike Ulrich. Google SRE, chapter 22: Addressing Cascading Failures, retry guidance. (accessed 2026-10-20)

JSON: /data/terms/retry-amplification.json