-
Notifications
You must be signed in to change notification settings - Fork 2.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Reuse VM memory across executions #1888
Conversation
pub(crate) fn get_vm_memory() -> Memory { | ||
POOL.with(|pool| { | ||
let mut pool = pool.borrow_mut(); | ||
pool.pop().unwrap_or_else(Memory::new) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should probably have some kind constraint on the max number of items allowed in each pool? Otherwise it should wait if there are none available.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We didn't do this earlier, and I don't see why it would make sense to track the items here, instead of just limiting concurrency of the executor itself. That said, keeping a pool per thread is by itself problematic and not useful, as:
- The threads are not CPU-pinned anyway
- The async runtime might decide to create new threads arbitarily, which would essentially cause a memory leak here
So I' changed the code to not keep separate pools in 5caec42
…o/reuse-vm-memory
## Version v0.28.0 ### Changed - [#1934](#1934): Updated benchmark for the `aloc` opcode to be `DependentCost`. Updated `vm_initialization` benchmark to exclude growing of memory(It is handled by VM reuse). - [#1916](#1916): Speed up synchronisation of the blocks for the `fuel-core-sync` service. - [#1888](#1888): optimization: Reuse VM memory across executions. #### Breaking - [#1934](#1934): Changed `GasCosts` endpoint to return `DependentCost` for the `aloc` opcode via `alocDependentCost`. - [#1934](#1934): Updated default gas costs for the local testnet configuration. All opcodes became cheaper. - [#1924](#1924): `dry_run_opt` has new `gas_price: Option<u64>` argument - [#1888](#1888): Upgraded `fuel-vm` to `0.51.0`. See [release](https://github.com/FuelLabs/fuel-vm/releases/tag/v0.51.0) for more information. ### Added - [#1939](#1939): Added API functions to open a RocksDB in different modes. - [#1929](#1929): Added support of customization of the state transition version in the `ChainConfig`. ### Removed - [#1913](#1913): Removed dead code from the project. ### Fixed - [#1921](#1921): Fixed unstable `gossipsub_broadcast_tx_with_accept` test. - [#1915](#1915): Fixed reconnection issue in the dev cluster with AWS cluster. - [#1914](#1914): Fixed halting of the node during synchronization in PoA service. ## What's Changed * Removed dead code by @xgreenx in #1913 * Added backward and forward compatibility integration tests for forkless upgrades by @xgreenx in #1895 * Fixed halting of the node in rare conditions by @xgreenx in #1914 * Weekly `cargo update` by @github-actions in #1928 * Fixed logging of the WASM executor by @xgreenx in #1930 * Added support of customization of the state transition version in the `ChainConfig` by @xgreenx in #1929 * Document wasm toolchain installation, add rust-toolchain.toml by @Dentosal in #1932 * Add optional `gas_price` argument to `dry_run_opt` by @hal3e in #1924 * Reuse VM memory across executions by @Dentosal in #1888 * Fixed reconnection issue in the dev cluster with AWS cluster by @xgreenx in #1915 * Speeds up synchronisation of the blocks for the `fuel-core-sync` service by @xgreenx in #1916 * Fixed unstable `gossipsub_broadcast_tx_with_accept` test by @xgreenx in #1921 * Added API functions to open a RocksDB in different modes by @xgreenx in #1939 * Use `DependentCost` for `aloc` opcode by @xgreenx in #1934 ## New Contributors * @hal3e made their first contribution in #1924 **Full Changelog**: v0.27.0...v0.28.0
Closes #1878. Requires FuelLabs/fuel-vm#732 and a new fuel-vm release.
Introduces a per-thread pool for allocated VM memory instances. Whenever a new VM instance is needed for a tx, the memory is taken from this pool. After the tx has been executed, the memory is reset to an empty state without deallocating, and returned to the pool.
A per-thread pool was chosen to no pass allocations between threads unnecessarily.
Checklist
Before requesting review