eva
eva copied to clipboard
A distributed database-system implementing an entity-attribute-value data-model that is time-aware, accumulative, and atomically consistent
We should drop/reject transactions enqueued past a certain (small) limit.
When indexing gets behind (following periods of sustained heavy write loads) we should apply backpressure in the transactor to give indexing an opportunity to catch up.
In the event that two transactor nodes simultaneously handle the same tenant, indexes can potentially stop flushing. Example scenario - a K8s pod spins up before the failing one is...
Index caching tests to ensure correctness for both transactor and peer use cases.
Setting `EVA_STORAGE_SQL_MAX_RETRIES` to `30000` (default `5`) prevents the code which repairs the database in situations where concurrent modifications occur from running.
Features for moving from hot storage to colder and colder storage.
Our existing metrics capturing `get-node` operations make up only a subset of an entire storage operation. It would be nice if we had a metric that encompassed all of the...
It would be good to have metrics around the hit and miss rates per instance so we can identify eg. noisy neighbor / cache bullying sorts of behavior.
GraalVM is a spiffy new vm/compiler for several different languages, including the JVM, which performs static analysis / optimization at the byte-code level. https://www.graalvm.org/ Folks have reported being able to...
On kubernetes we have the option of specifying new liveness and readiness probes. Our current http status endpoint is largely independent of our actual system state. We should design/build more...