Closed khsrali closed 22 hours ago
Trial attempts may be ineffective in certain transport plugins, such as aiida-firecrest.
Why is this? Isn't it possible that your internet connection temporarily drops, or that the FireCREST service is (temporarily) unavailable? These would be transient problems that would be helped from the expontentiall backoff mechanism just as it does for the SSH transport, doesn't it?
I thought the main idea of exponential backup was to avoid being banned from HPC by sending too many requests.
In FirecREST
they have a similar concept, rate-limit, but it's handled internally by pyfirecrest
which aiida-firecrest
depends on.
I understand and agree that still there are other things could fail, like internet connection, FirecREST server being done temporarily. However, this doesn't need to be exponential.
So I think it make sense if the backoff mechanism (series of seconds 1, 2, 4, 8, 16 or 1,1,1,1) would be set from the transport plugin / computer settings.
I thought the main idea of exponential backup was to avoid being banned from HPC by sending too many requests.
No, that is the safe_interval
setting which is set on the transport. The EBM is done for all CalcJob
s and is to ensure that an exception in a transport operation that is due to a transient problem (such as connection problems) doesn't except the entire process. So this has nothing to do with rate limiting
Alright, I think we already set safe_interval=0
in aiida-firecrest
,
So I think we can close here.
Thanks a lot @sphuber for clearing that up,
Trial attempts may be ineffective in certain transport plugins, such as
aiida-firecrest
. If I understood correctly this option can be set to zero for the profile insrc/aiida/manage/configuration/config
.However, it would be preferable to leave it in the hands of the transport plugin, as multiple transport plugins (including SSH) may coexist in the same profile.