Chris

I don't have a 4.0 or the time to work on it, but happy to take a patch. On 6 Mar 2021, 18:14, at 18:14, netmindz wrote: >Any plans to...

The first part is likely to be correct. I'd fix it, but if you'd prefer to have your commit in there please send a pull request with your changes. On...

I think this simply needs the lame_set_VBR method exposing and the enum to pass into it.

I don't think this is the best solution. I think waiting for EOF on both RX in the worker makes more sense. The ownership of the worker is by itself...

*[`nativelink-util/src/connection_manager.rs` line 225 at r1](https://reviewable.io/reviews/TraceMachina/nativelink/866#-NvpB6uG6r7IQ-wlcYXD:-NvpB6uHAxtVol_gqtko:bqvjqia) ([raw file](https://github.com/TraceMachina/nativelink/blob/7319ba758c9156832ea1eb1cd9bc0319e77a4d90/nativelink-util/src/connection_manager.rs#L225)):* > ```XML > // The ConnectionManager was dropped, shut down the > // worker. > break; > ``` We should add a loop...

Will be Tuesday before I can try. Will give time for the jemalloc version to soak.

I have a branch with an initial implementation of this, just waiting on a few more PRs before it is possible to bring it in.

I like it from a configuration point of view, but I wonder if the implementation and testing might be difficult with so many permutations?

Tricky implementation detail: when paired with a fast-slow store this would need to know if the original request is a read or write operation. Perhaps this might be better served...

@MarcusSorealheis I've brought this back up to date.