Comments for post About the new INFO command, Diskstore and Cluster ETA
Navigatore Anonimo writes:
Alaric Snell-Pym writes: Where I work at GenieDB, we expose a heap of stats by putting them in a C struct stored in mmap-ed memory. This restricts you to same-node access, and means you need to control it with UNIX filesystem permissions in all their glory, and you have to be very careful about atomicity of updates, but it does mean that the time taken to sample stats is miniscule so we do crazy things like having a tool that samples fine-grained counters a thousand times a second and drawing histograms thereof. It's been invaluable for our profiling... but it is quite a low-level trick :-)
antirez writes: @Theo: thanks for your feedback. We believe that there is at least the same number of people waiting for diskstore, but it is a much faster project to ship into stable form, this is why it's IMHO better to start from diskstore. Same amount of happy users but in littler time. Also delaying cluster 6 weeks is not as bad as delaying diskstore for ... 6 months. Given that the time to provide stable versions of the two projects is so different.
Theo writes: I think that prioritizing diskstore over cluster is a great idea. From my own perspective diskstore would solve 90% of the problems I have with scaling Redis (I'm not using it for speed, but for storing and calculating the unions of large sets). Cluster would solve it too, but in a less ideal way. I can see that cluster is more important for other people, though. Diskstore was a big and very welcome surprise.