Hi. I’m testing CrateDB and I was running a benchmark, but after trying twice I saw a sudden drop in performance after about 1hour of ingesting mock data. Is that behavior expected? I looked around a couldn’t find any limits on the trial or prod clusters. Can someone confirm that the reduced performance is in fact throttling and if it also happens in production?
Could you elaborate what scenarios (cluster sizes, ingest, disk sizes) you tested?
There is no specific throttling in CrateDB Cloud, however on the Azure based clusters the used PremiumSSDs (especially smaller variants) have a burstable behaviour.
I checked back with the cloud team. The provisioned 32GiB disks are Azure Premium SSDs with a default IOPS of 120 and throughput of 25 MB/s. Those disks can burst to 3 500 IOPS and 170 MB/s for a limited amount of time. So you probably can experience this especially with longer running ingest tests.
This credit based bursting is only relevant for disk 512GiB and smaller. Bigger disks have higher provisioned IOPS and throughout. e.g. a 1 TiB disk comes with 5000 provisioned IOPS and 200 MB/s throughput.
On request we can also provide deviating disk setup.