Observation

On nodes running out-of-cache workloads, the WiredTiger cache miss rate is exceeding the cache hit rate. This indicates that the working set is not being served from the in-memory cache, causing frequent disk reads and elevated read latency.

Investigation Questions

• Is the cache simply undersized for this workload — i.e. the working set is larger than the configured cache and the behavior is expected?
• Or is WiredTiger doing something wrong — e.g. evicting hot pages prematurely, miscounting cache hits/misses, or a bug in the cache management path?

Areas to Check

• Cache size vs working set size ratio for the affected cluster.
• Eviction patterns — are frequently accessed pages being evicted before they can be reused?
• Whether the cache miss rate correlates with eviction pressure or a specific query pattern.
• Whether the ratio is stable (expected out-of-cache behavior) or spiking (indicative of a problem).

Expected Outcome

Determine whether the cache miss rate is a consequence of an undersized cache for the workload, or whether there is a WiredTiger-side issue causing avoidable cache misses, and recommend corrective action accordingly.