Alex Chi Z.
Alex Chi Z.
Or we can use musl instead of glibc?
One feasible solution is to shrink the scope of generics. e.g. only have JoinSide on critical sections, instead of over full join executor. In this way, we can generate as...
After dropping support for TiKV and RocksDB state backend, 3m30s -> 2m37s 🤪
After changing all `start` functions' signature to: ``` pub fn start(opts: ComputeNodeOpts) -> Pin { ``` (previously, we use `pub async fn start` so that the async will be expanded...
After also dropping memory state backend support (only in release mode), 1m 30s :) The time is measured by: changing the `LOCAL_CHANNEL_OUTPUT` size and recompile using ``` cargo build -p...
Okay, so the remaining is for @yuhao-su to optimize hash join 🤣
Is it possible to have some inner parts of code to return `Box`, `Box` instead opaque type and async fn? So that this "boxed" part can be re-used by compilers.
I guess the main overhead of compile is expanding the async function body (which requires constructing a state machine). If we can make non critical sections to have `Box`, then...
ministant should fall back to something similar to coarsetime on arm64, as arm64 doesn't support TSO like x86 did. It would be okay to switch to coarsetime.