Benedict Elliott Smith
Benedict Elliott Smith
This patch refactors CommandsForKey to a single efficient special-purpose collection, that can answer all queries without occupying too much memory. There remain further efficiencies to be had. In addition, this...
Introduce faster per-key linearizable reads (suitable for any single key transactional read, or weaker multi-key reads). These guarantee 1 WAN RT, require minimal state, and offer single replica consistent execution....
While messages are submitted to a CommandStore asynchronously, the processing of a message is then expected to be synchronous. However, it would be beneficial particularly for large range transactions, but...
It occurs to me that a lot of the variance between runs we are seeing could be explained by random variation in vnode token allocation fairness. If this were copied...
Along with improved naming, it would be great if jobs themselves could be tagged after the fact by users, to note important / useful runs and help finding them.
It would be great if we could either rename jobs after they've run or, even better, that the job names were partially auto-generated based on the settings of the stress...
The URLs for github gists are really quite long, and corroborating which one you've pasted (and making sure you don't screw up the rest of the command) is somewhat tedious....
Right now (especially close to release) we're having problems with patches with different baselines being compared to each other. It would be great if cstar presented the most recent SHA...
Our performance tests are never like-for-like comparisons, because compaction makes no guarantees about consistent end state. Separately, working with huge datasets is extremely costly now, as populating them takes far...