Druid Historical Startup I/O

Hi all,

I am curious about what I/O is being done during the startup of a historical node. I understand that all of the segments on a node need to be memory mapped, but I’m not understanding why so much I/O needs to occur… Is each segment being read in its entirety? During startup our HW is maxing out read throughput throughout the whole startup process. Is there any configuration, etc. to avoid having to do so much I/O during startup, or is this always going to be an I/O bound activity.

Druid: 0.11.0

HW: 4 x SATA SSD RAID 0 configuration. Max read ~2GB/s.

Data Size: ~ 5TB / node

Thanks in advance for your input!