I did a different implementation in my own rest service where I instantiated not only a fresh cas, but fresh pipeline for each threadpool object. It's memory hungry but safe - running for weeks with 35 simultaneous threads & zero errors. I wasn't convinced that all the AEs were thread-safe to let them be shared. So have a look at how the REST server sets up its threadpool and what it does after the service has completed a request in terms of releasing resources. On RHEL, you may also need to increase your ulimit if you're getting connection refused errors under heavy load. I discovered that....
Peter On Thu, Mar 25, 2021 at 3:26 PM Miller, Timothy < timothy.mil...@childrens.harvard.edu> wrote: > Just wondering what the logistics of this are. The REST interface has a > CAS pool of 10, and when it gets a new request, it grabs a CAS and > sends it into a pipeline. So what happens if the REST endpoint is > getting hit by tons of different requests at the same time? I'm > experimenting with this in python and getting hard to understand errors > (best as I can tell it looks like it's complainin that the output is > None). Just wondering if anyone has any insight about what's going on > on the server side and whether a) this _should_ work, b) it _could_ > work if done properly. > > Thanks > Tim > >