Hi
I am currently investigating the feasibility of booting many AiiDA RESTAPIs under one service.
The old implementation uses a supervisord PID spawner for each profile, which has considerable overhead, (50MB per instance) which is resulting in large memory concerns when spinning up over 100 profiles >5Gb+.
I have noticed that hte AiiDA docs seems to imply the following is supported:
http://127.0.0.1:5000/api/v4/computers?profile=some-profile-name
https://aiida.readthedocs.io/projects/aiida-core/en/stable/howto/share_data.html?utm_source=chatgpt.com
However, this seems to be a syntax i would not keep, (putting this sort of routing semantics in searchparams seems undesired)
I would propose an alternative:
http://127.0.0.1:5000/api/v4/some-profile-name/computers
I think this feature also appears to be a little buggy / may not work with QueryBuilder or other things.
I can in principle have a hypervisord that spins up restapis only on request to avoid having many PIDs open at once and the large memory footprint, https://github.com/Bud-Macaulay/aiida-hyperserver ... but perhaps this feels like a feature we may want to support at the rest-api level.
Hi
I am currently investigating the feasibility of booting many AiiDA RESTAPIs under one service.
The old implementation uses a supervisord PID spawner for each profile, which has considerable overhead, (50MB per instance) which is resulting in large memory concerns when spinning up over 100 profiles >5Gb+.
I have noticed that hte AiiDA docs seems to imply the following is supported:
http://127.0.0.1:5000/api/v4/computers?profile=some-profile-namehttps://aiida.readthedocs.io/projects/aiida-core/en/stable/howto/share_data.html?utm_source=chatgpt.com
However, this seems to be a syntax i would not keep, (putting this sort of routing semantics in searchparams seems undesired)
I would propose an alternative:
http://127.0.0.1:5000/api/v4/some-profile-name/computersI think this feature also appears to be a little buggy / may not work with QueryBuilder or other things.
I can in principle have a hypervisord that spins up restapis only on request to avoid having many PIDs open at once and the large memory footprint, https://github.com/Bud-Macaulay/aiida-hyperserver ... but perhaps this feels like a feature we may want to support at the rest-api level.