project
Embedded key-value store used as the local state store holding running per-user aggregates with window expiry in the streaming feature pipeline.
Moving authz-gateway's key checks from Postgres to a Kafka Streams store cut p99 from 40ms to 4ms, yet a revoked key kept working for four days. Three pods cut off from Kafka for 30 hours rejoined at zero lag and kept honoring it, a dev.to write-up says.
Reality
Perplexity moved its search serving tier from DynamoDB to CobbleDB, an in-house Rust store, and cut median batch reads from 31.4ms to 5.60ms. The cost case carries to other retrieval teams only at similar payload sizes and request rates.
Reality
An air-gapped cluster of 10 servers runs every workload on NFS. The write-up blames that for Elasticsearch and Zeebe I/O failures, and its own numbers rule out Longhorn and local-path on ~100GB node disks.
Reality
The partition count was frozen at checkpoint creation because each partition has its own RocksDB instance on disk. In Databricks Runtime 18, a separate config resizes the store and moves the state to match.
Reality
ParaQL logs every modifying statement to a rebased multi-writer oplog, because conflicting writes corrupt the file fast. Benchmarking found 1MiB of data taking up to 200MiB on disk.
Reality
A p99 of 200 milliseconds from Kafka to the online store beats batch cadence by a wide margin. It also stops before your model reads the feature.
Reality