Failure-policy fingerprinting
Failure-policy fingerprinting uses targeted faults and controlled workloads to identify a filesystem's detection and recovery behavior.
Method: IRON's 2005 framework varies the operation, block type and fault, then compares faulted and fault-free runs using API results, logs and I/O traces. One failed block cannot characterize every error path. [1]
Injection boundary: Its pseudo-device driver returns read or write errors, or alters read data, with transient and persistent faults. Injection below the filesystem does not evaluate retries or mitigation in lower layers. [1]
Interpretation: Detection and recovery are separate observations. Returning an error, stopping writes, retrying and reconstructing data have different consequences; a detected fault does not imply restored content. [1]
Later methodology: A 2019 study targets individual structures and accesses on Linux 4.17, reinitializes the image for each trial, and checks both immediate behavior and offline recovery. Mountability with lost data, restoration of the pre-operation state and full expected-state recovery are separate outcomes. [2]
Relations
- related
- Fault injection, Fail-partial failure model, Semantic block-level analysis, Intra-disk redundancy, Crash-test oracle
Limitations
- Observed behavior is bounded by the tested workload, implementation version and selected fault cases, not an estimate of production failure frequency. [1, 2]
Citations
- Prabhakaran et al. IRON File Systems. SOSP 2005, sections 3-4.
- Jaffer et al. Evaluating File System Reliability on Solid State Drives. USENIX ATC 2019, sections 2.3-2.5 and table 4.
Sources & caveats
- Source-based methodology synthesis; no execution of either historical fault-injection framework or evaluation of current filesystem versions.