Bulk UID creation?

[John Humphreys]

We have 40,000 servers, hundreds of metrics, and hundreds of analytics on each metric.

If I up the UID bytes for metrics to 4 bytes, I can push the server name into the metric (metric = server.metric).  Right now I have to use the server-id as the metric and the metric as a tag to make things performant which makes things like Grafana frustrating to use.

The UID creation is incredibly slow though.  Right now, 150-million metric IDs are taking like 36 hours to make (with a multi-threaded app making rest calls).  Is there any way to create metrics in bulk that is vastly faster than the REST API?  Why is it so slow?

[ManOLamancha]

On Thursday, September 7, 2017 at 8:17:14 AM UTC-7, John Humphreys wrote:

The UID creation is incredibly slow though.  Right now, 150-million metric IDs are taking like 36 hours to make (with a multi-threaded app making rest calls).  Is there any way to create metrics in bulk that is vastly faster than the REST API?  Why is it so slow?

http://opentsdb.net/docs/build/html/api_http/uid/assign.html

In real-time it's slow due to the need to lock on the UID row, unfortunately. We're exploring some options around hashing to get rid of UIDs as an option in 3.x