The GH cache is a compressed tar, hence it's not possible to just-in-time (JIT) deliver only those artifacts which are actually required.
On an average size big mono-repo this amounts to 7min lead time just downloading and decompressing the GH cache (and same for populating in the first place).
An S3 interface to the cache would be enough to let nix do it's thing and only fetch what's necessary by the actual job at hand.
We should implement the run action inputs in a way that they could seemlessly make use of a s3-like cache should it be made available.
The GH cache is a compressed tar, hence it's not possible to just-in-time (JIT) deliver only those artifacts which are actually required.
On an average size big mono-repo this amounts to 7min lead time just downloading and decompressing the GH cache (and same for populating in the first place).
An S3 interface to the cache would be enough to let
nixdo it's thing and only fetch what's necessary by the actual job at hand.We should implement the run action inputs in a way that they could seemlessly make use of a s3-like cache should it be made available.