Shared Cache for TileServer Proxy with TensorStore

[unidesigner]

I'm implementing a tile server backed by tensorstore using FastAPI with multiple workers. I'd like to use a single common memory cache across the workers and I figured that this may be possible with the Tensorstore Context Framework.
However, I'm unclear on how to configure such a shared context when opening a dataset. Could you provide any guidance or example code to help? Thank you!

[jbms]

There is no way to share the cache across processes, unfortunately.

[unidesigner]

Also not with tmpfs ?

[jbms]

Certainly there are ways in principle to share a cache across processes, but the cache in tensorstore does not support that. An alternative would be to use a single-process server, but I don't know if the GIL overhead will be too great. TensorStore does not yet support nogil Python, but in the future it will, and that will presumably allow for improved multi-threaded server performance.

[unidesigner]

I'll look into using a single-process server, and perform benchmarks for different request situations, or deploy single-process servers for different scale levels with background prefetching. It'd be great to be able to also compare performance with nogil Python later.