You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We should rely on the default number of CPUs, as the python os.get_cpus (or similar) will not correctly work on deployments where host and single instance have different number of dedicated cores (which is almost always the case, the single instance will always have only a fraction of the host resources)
The text was updated successfully, but these errors were encountered:
Hi @edan-bainglass, yes, it needs to be implemented. Is a fixing to the warning system. The issue is that now we estimate the available cores using python, but this detects the total cores of the host system, not the ones assigned to the container (this is an issue mainly in kubernetes clusters, e.g. demo server). We need to get the number of available cpus from the code setup parameters, defined when the qe default codes are installed (if I remember correctly, @unkcpz told me that this is done when the container is generated, right?).
I still need to check this, should happen this week
if I remember correctly, @unkcpz told me that this is done when the container is generated, right?).
Yes, it is set after container start bind to the computer node setup. By verdi computer show <localhost-hq> you can see the number of CPUs can be used (and by API as well).
We should rely on the default number of CPUs, as the python os.get_cpus (or similar) will not correctly work on deployments where host and single instance have different number of dedicated cores (which is almost always the case, the single instance will always have only a fraction of the host resources)
The text was updated successfully, but these errors were encountered: