Make sure we can reach the user's requested FVM concurrency

We previously used the FVM's `ThreadedExecutor` to execute messages on separate threads because the FVM requires 64MiB of stack space. 1. The FVM v3 supported for 8 concurrent threads. 2. The FVM v4 supports up to the number of CPU threads available. Unfortunately, neither version was influenced by the `LOTUS_FVM_CONCURRENCY` environment variable. This patch fixes this by: 1. Moving the thread-pool to the FFI itself (sharing it between FVM versions). 2. Setting the thread-pool size equal to `LOTUS_FVM_CONCURRENCY`. It also defaults `LOTUS_FVM_CONCURRENCY` to the number of available CPU threads instead of the previous 4. NOTE: I've also tried increasing the stack size instead of using threads, but Go _does not_ like it when other foreign mess with the stack size of _its_ threads (but it has no problem if we create our own threads). fixes filecoin-project/lotus#11817
filecoin-project · Apr 6, 2024 · c6296ec · c6296ec
1 parent b715c94
commit c6296ec
Show file tree

Hide file tree

Showing 4 changed files with 80 additions and 66 deletions.
diff --git a/rust/Cargo.lock b/rust/Cargo.lock
diff --git a/rust/Cargo.toml b/rust/Cargo.toml
@@ -46,13 +46,14 @@ serde = "1.0.117"
 serde_tuple = "0.5"
 safer-ffi = { version = "0.0.7", features = ["proc_macros"] }
 filecoin-proofs-api = { version = "16.1", default-features = false }
+yastl = "0.1.2"
 
 [dev-dependencies]
 memmap2 = "0.5"
 tempfile = "3.0.8"
 
 [features]
-default = ["cuda", "multicore-sdr" ]
+default = ["cuda", "multicore-sdr"]
 blst-portable = ["bls-signatures/blst-portable", "blstrs/portable"]
 cuda = ["filecoin-proofs-api/cuda", "rust-gpu-tools/cuda", "fvm2/cuda", "fvm3/cuda", "fvm4/cuda"]
 cuda-supraseal = ["filecoin-proofs-api/cuda-supraseal", "rust-gpu-tools/cuda", "fvm3/cuda-supraseal", "fvm4/cuda-supraseal"]

diff --git a/rust/src/fvm/engine.rs b/rust/src/fvm/engine.rs
@@ -1,5 +1,4 @@
 use std::collections::HashMap;
-use std::ops::RangeInclusive;
 use std::sync::{Arc, Mutex};
 
 use anyhow::anyhow;
@@ -18,7 +17,7 @@ use super::externs::CgoExterns;
 use super::types::*;
 
 // Generic executor; uses the current (v3) engine types
-pub trait CgoExecutor {
+pub trait CgoExecutor: Send {
     fn execute_message(
         &mut self,
         msg: Message,
@@ -65,9 +64,6 @@ pub struct MultiEngineContainer {
     engines: Mutex<HashMap<EngineVersion, Arc<dyn AbstractMultiEngine + 'static>>>,
 }
 
-const LOTUS_FVM_CONCURRENCY_ENV_NAME: &str = "LOTUS_FVM_CONCURRENCY";
-const VALID_CONCURRENCY_RANGE: RangeInclusive<u32> = 1..=128;
-
 impl TryFrom<u32> for EngineVersion {
     type Error = anyhow::Error;
     fn try_from(value: u32) -> Result<Self, Self::Error> {
@@ -81,41 +77,6 @@ impl TryFrom<u32> for EngineVersion {
 }
 
 impl MultiEngineContainer {
-    /// Constructs a new multi-engine container with the default concurrency (4).
-    pub fn new() -> MultiEngineContainer {
-        Self::with_concurrency(4)
-    }
-
-    /// Constructs a new multi-engine container with the concurrency specified in the
-    /// `LOTUS_FVM_CONCURRENCY` environment variable.
-    pub fn new_env() -> MultiEngineContainer {
-        let valosstr = match std::env::var_os(LOTUS_FVM_CONCURRENCY_ENV_NAME) {
-            Some(v) => v,
-            None => return Self::new(),
-        };
-        let valstr = match valosstr.to_str() {
-            Some(s) => s,
-            None => {
-                log::error!("{LOTUS_FVM_CONCURRENCY_ENV_NAME} has invalid value");
-                return Self::new();
-            }
-        };
-        let concurrency: u32 = match valstr.parse() {
-            Ok(v) => v,
-            Err(e) => {
-                log::error!("{LOTUS_FVM_CONCURRENCY_ENV_NAME} has invalid value: {e}");
-                return Self::new();
-            }
-        };
-        if !VALID_CONCURRENCY_RANGE.contains(&concurrency) {
-            log::error!(
-                "{LOTUS_FVM_CONCURRENCY_ENV_NAME} must be in the range {VALID_CONCURRENCY_RANGE:?}, not {concurrency}"
-            );
-            return Self::new();
-        }
-        Self::with_concurrency(concurrency)
-    }
-
     pub fn with_concurrency(concurrency: u32) -> MultiEngineContainer {
         MultiEngineContainer {
             engines: Mutex::new(HashMap::new()),
@@ -146,12 +107,6 @@ impl MultiEngineContainer {
     }
 }
 
-impl Default for MultiEngineContainer {
-    fn default() -> MultiEngineContainer {
-        MultiEngineContainer::new()
-    }
-}
-
 // fvm v4 implementation
 mod v4 {
     use anyhow::anyhow;
@@ -160,10 +115,7 @@ mod v4 {
 
     use fvm4::call_manager::DefaultCallManager as DefaultCallManager4;
     use fvm4::engine::{EnginePool as EnginePool4, MultiEngine as MultiEngine4};
-    use fvm4::executor::{
-        ApplyKind, ApplyRet, DefaultExecutor as DefaultExecutor4,
-        ThreadedExecutor as ThreadedExecutor4,
-    };
+    use fvm4::executor::{ApplyKind, ApplyRet, DefaultExecutor as DefaultExecutor4};
     use fvm4::kernel::filecoin::DefaultFilecoinKernel as DefaultFilecoinKernel4;
     use fvm4::machine::{DefaultMachine as DefaultMachine4, NetworkConfig};
     use fvm4_shared::{chainid::ChainID, clock::ChainEpoch, message::Message};
@@ -175,14 +127,13 @@ mod v4 {
     use super::Config;
 
     type CgoMachine4 = DefaultMachine4<CgoBlockstore, CgoExterns>;
-    type BaseExecutor4 = DefaultExecutor4<DefaultFilecoinKernel4<DefaultCallManager4<CgoMachine4>>>;
-    type CgoExecutor4 = ThreadedExecutor4<BaseExecutor4>;
+    type CgoExecutor4 = DefaultExecutor4<DefaultFilecoinKernel4<DefaultCallManager4<CgoMachine4>>>;
 
     fn new_executor(
         engine_pool: EnginePool4,
         machine: CgoMachine4,
     ) -> anyhow::Result<CgoExecutor4> {
-        Ok(ThreadedExecutor4(BaseExecutor4::new(engine_pool, machine)?))
+        CgoExecutor4::new(engine_pool, machine)
     }
 
     impl CgoExecutor for CgoExecutor4 {
@@ -254,8 +205,7 @@ mod v3 {
     };
     use fvm3::engine::{EnginePool as EnginePool3, MultiEngine as MultiEngine3};
     use fvm3::executor::{
-        ApplyFailure as ApplyFailure3, ApplyKind as ApplyKind3,
-        DefaultExecutor as DefaultExecutor3, ThreadedExecutor as ThreadedExecutor3,
+        ApplyFailure as ApplyFailure3, ApplyKind as ApplyKind3, DefaultExecutor as DefaultExecutor3,
     };
     use fvm3::machine::{DefaultMachine as DefaultMachine3, NetworkConfig as NetworkConfig3};
     use fvm3::trace::ExecutionEvent as ExecutionEvent3;
@@ -284,14 +234,13 @@ mod v3 {
     use super::Config;
 
     type CgoMachine3 = DefaultMachine3<CgoBlockstore, CgoExterns>;
-    type BaseExecutor3 = DefaultExecutor3<DefaultKernel3<DefaultCallManager3<CgoMachine3>>>;
-    type CgoExecutor3 = ThreadedExecutor3<BaseExecutor3>;
+    type CgoExecutor3 = DefaultExecutor3<DefaultKernel3<DefaultCallManager3<CgoMachine3>>>;
 
     fn new_executor(
         engine_pool: EnginePool3,
         machine: CgoMachine3,
     ) -> anyhow::Result<CgoExecutor3> {
-        Ok(ThreadedExecutor3(BaseExecutor3::new(engine_pool, machine)?))
+        CgoExecutor3::new(engine_pool, machine)
     }
 
     impl CgoExecutor for CgoExecutor3 {
@@ -533,8 +482,7 @@ mod v2 {
         backtrace::Cause as Cause2, DefaultCallManager as DefaultCallManager2,
     };
     use fvm2::executor::{
-        ApplyFailure as ApplyFailure2, ApplyKind as ApplyKind2,
-        DefaultExecutor as DefaultExecutor2, ThreadedExecutor as ThreadedExecutor2,
+        ApplyFailure as ApplyFailure2, ApplyKind as ApplyKind2, DefaultExecutor as DefaultExecutor2,
     };
     use fvm2::machine::{
         DefaultMachine as DefaultMachine2, MultiEngine as MultiEngine2,
@@ -565,11 +513,10 @@ mod v2 {
     use super::Config;
 
     type CgoMachine2 = DefaultMachine2<CgoBlockstore, CgoExterns>;
-    type BaseExecutor2 = DefaultExecutor2<DefaultKernel2<DefaultCallManager2<CgoMachine2>>>;
-    type CgoExecutor2 = ThreadedExecutor2<BaseExecutor2>;
+    type CgoExecutor2 = DefaultExecutor2<DefaultKernel2<DefaultCallManager2<CgoMachine2>>>;
 
     fn new_executor(machine: CgoMachine2) -> CgoExecutor2 {
-        ThreadedExecutor2(BaseExecutor2::new(machine))
+        CgoExecutor2::new(machine)
     }
 
     fn bytes_to_block(bytes: RawBytes) -> Option<IpldBlock> {

diff --git a/rust/src/fvm/machine.rs b/rust/src/fvm/machine.rs
@@ -1,5 +1,6 @@
 use std::borrow::Cow;
 use std::convert::{TryFrom, TryInto};
+use std::ops::RangeInclusive;
 
 use anyhow::{anyhow, Context};
 use cid::Cid;
@@ -27,8 +28,54 @@ use super::types::*;
 use crate::destructor;
 use crate::util::types::{catch_panic_response, catch_panic_response_no_default, Result};
 
+const STACK_SIZE: usize = 64 << 20; // 20MiB
+
 lazy_static! {
-    static ref ENGINES: MultiEngineContainer = MultiEngineContainer::new_env();
+    static ref CONCURRENCY: u32 = get_concurrency();
+    static ref ENGINES: MultiEngineContainer = MultiEngineContainer::with_concurrency(*CONCURRENCY);
+    static ref THREAD_POOL: yastl::Pool = yastl::Pool::with_config(
+        *CONCURRENCY as usize,
+        yastl::ThreadConfig::new()
+            .prefix("fvm")
+            .stack_size(STACK_SIZE)
+    );
+}
+
+const LOTUS_FVM_CONCURRENCY_ENV_NAME: &str = "LOTUS_FVM_CONCURRENCY";
+const VALID_CONCURRENCY_RANGE: RangeInclusive<u32> = 1..=256;
+
+fn available_parallelism() -> u32 {
+    std::thread::available_parallelism()
+        .map(usize::from)
+        .unwrap_or(8) as u32
+}
+
+fn get_concurrency() -> u32 {
+    let valosstr = match std::env::var_os(LOTUS_FVM_CONCURRENCY_ENV_NAME) {
+        Some(v) => v,
+        None => return available_parallelism(),
+    };
+    let valstr = match valosstr.to_str() {
+        Some(s) => s,
+        None => {
+            log::error!("{LOTUS_FVM_CONCURRENCY_ENV_NAME} has invalid value");
+            return available_parallelism();
+        }
+    };
+    let concurrency: u32 = match valstr.parse() {
+        Ok(v) => v,
+        Err(e) => {
+            log::error!("{LOTUS_FVM_CONCURRENCY_ENV_NAME} has invalid value: {e}");
+            return available_parallelism();
+        }
+    };
+    if !VALID_CONCURRENCY_RANGE.contains(&concurrency) {
+        log::error!(
+            "{LOTUS_FVM_CONCURRENCY_ENV_NAME} must be in the range {VALID_CONCURRENCY_RANGE:?}, not {concurrency}"
+        );
+        return available_parallelism();
+    }
+    concurrency
 }
 
 #[allow(clippy::too_many_arguments)]
@@ -181,14 +228,32 @@ fn create_fvm_debug_machine(
     )
 }
 
+fn with_new_stack<F, T>(name: &str, pool: &yastl::Pool, callback: F) -> repr_c::Box<Result<T>>
+where
+    T: Sized + Default + Send,
+    F: FnOnce() -> anyhow::Result<T> + std::panic::UnwindSafe + Send,
+{
+    let mut res = None;
+    pool.scoped(|scope| scope.execute(|| res = Some(catch_panic_response(name, callback))));
+
+    res.unwrap_or_else(|| {
+        repr_c::Box::new(Result::err(
+            format!("failed to schedule {name}")
+                .into_bytes()
+                .into_boxed_slice(),
+        ))
+    })
+}
+
 #[ffi_export]
 fn fvm_machine_execute_message(
     executor: &'_ InnerFvmMachine,
     message: c_slice::Ref<u8>,
     chain_len: u64,
     apply_kind: u64, /* 0: Explicit, _: Implicit */
 ) -> repr_c::Box<Result<FvmMachineExecuteResponse>> {
-    catch_panic_response("fvm_machine_execute_message", || {
+    // Execute in the thread-pool because we need a 64MiB stack.
+    with_new_stack("fvm_machine_execute_message", &THREAD_POOL, || {
         let apply_kind = if apply_kind == 0 {
             ApplyKind::Explicit
         } else {