WireBaron

7 ms·

1. 1.▲
   by WireBaron 5y ago

   This is essentially correct. It's a little more complicated inside, and the expansion's really more like: SELECT device_id, arrow_run_pipeline(timevector(ts, val), arrow_add_element(sort(), arrow_add_element(delta(), arrow_add_el
2. 2.▲
   by WireBaron 5y ago

   The UDDSketch (default) implementation will allow rolling percentiles, though we still need a bit of work on our end to support it. There isn't a way to do this with TDigest however.
3. 3.▲
   by WireBaron 7y ago

   We actually have done fairly extensive benchmarking of high cardinality data on our single-node product (we have a blog entry detailing at least our insert performance here: https://blog.timescale.com/blog/what-is-high-
4. 4.▲
   by WireBaron 7y ago

   We actually haven't been running against any limits here. One thing to keep in mind is that postgres remote-fetch operations aren't tuple-at-a-time, so this shouldn't be a bottleneck for our multi-node operations.