Merge pull request #96 from paritytech/refactor/provider-trait

Refactor of ```Provider``` trait into ```Provider```, ```ProviderNamespace``` and ```ProviderNode```

.github/workflows/ci.yml

@@ -37,8 +37,12 @@ jobs:
       - name: Build
         run: cargo build
 
-      - name: Tests
-        run: cargo test
+      - name: Tests (except provider crate)
+        run: cargo test --workspace --exclude provider
+
+      - name: Tests (provider crate)
+        # there should be a unique test thread for native provider tests (asserting spawned processes count)
+        run: cargo test -p provider -- --test-threads 1
 
   coverage:
     name: Zombienet SDK - coverage
@@ -60,7 +64,8 @@ jobs:
         uses: taiki-e/install-action@cargo-llvm-cov
 
       - name: Collect coverage data
-        run: cargo llvm-cov nextest --lcov --output-path lcov.info
+        # there should be a unique test thread for native provider tests (asserting spawned processes count)
+        run: cargo llvm-cov nextest -j 1 --lcov --output-path lcov.info
 
       - name: Report code coverage
         uses: Nef10/[email protected]

.github/workflows/documentation.yml

@@ -35,13 +35,15 @@ jobs:
           cargo doc --no-deps
           echo "<meta http-equiv=\"refresh\" content=\"0; url=configuration\">" > target/doc/index.html
 
+
+
       - name: Move docs
         run: |
           mkdir -p ./doc
           mv ./target/doc/* ./doc
           git config user.email "[email protected]"
           git config user.name "GitHub Action"
-          git config user.password ${{ secrets.GH_PAGES_TOKEN }}
+          git config user.password "${{ secrets.GH_PAGES_TOKEN }}"
           git checkout --orphan gh-pages
           mkdir to_delete
           shopt -s extglob

Cargo.toml

@@ -23,3 +23,6 @@ regex = "1.8"
 lazy_static = "1.4"
 multiaddr = "0.18"
 url = "2.3"
+uuid = "1.4"
+nix = "0.27"
+procfs = "0.15"

crates/configuration/src/lib.rs

@@ -12,3 +12,5 @@ pub use hrmp_channel::{HrmpChannelConfig, HrmpChannelConfigBuilder};
 pub use network::{NetworkConfig, NetworkConfigBuilder};
 pub use parachain::{ParachainConfig, ParachainConfigBuilder};
 pub use relaychain::{RelaychainConfig, RelaychainConfigBuilder};
+// re-export shared
+pub use shared::{node::NodeConfig, types};

crates/provider/Cargo.toml

@@ -8,11 +8,23 @@ edition = "2021"
 [dependencies]
 support = { path = "../support" }
 configuration = { path = "../configuration" }
-async-trait = {workspace = true }
-futures = {workspace = true }
+async-trait = { workspace = true }
+futures = { workspace = true }
 #napi = { version="2.12.7", features=["async"]}
 #napi-derive = "2.12.5"
 serde = { workspace = true, features = ["derive"] }
-serde_json = {workspace = true}
-tokio = { workspace = true, features = ["process", "macros", "fs", "time", "rt"] }
-thiserror = {workspace = true}
+serde_json = { workspace = true }
+tokio = { workspace = true, features = [
+    "process",
+    "macros",
+    "fs",
+    "time",
+    "rt",
+] }
+thiserror = { workspace = true }
+anyhow = { workspace = true }
+uuid = { workspace = true, features = ["v4"] }
+nix = { workspace = true, features = ["signal"] }
+
+[dev-dependencies]
+procfs = { workspace = true }

crates/provider/src/lib.rs

@@ -1,94 +1,132 @@
-mod errors;
-mod native;
-mod shared;
+pub mod native;
+pub mod shared;
 
-use std::{net::IpAddr, path::PathBuf};
+use std::{
+    collections::HashMap, net::IpAddr, path::PathBuf, process::ExitStatus, sync::Arc,
+    time::Duration,
+};
 
 use async_trait::async_trait;
-use errors::ProviderError;
-use shared::types::{FileMap, NativeRunCommandOptions, PodDef, Port, RunCommandResponse};
+use shared::types::{
+    GenerateFileCommand, GenerateFilesOptions, ProviderCapabilities, RunCommandOptions,
+    RunScriptOptions, SpawnNodeOptions,
+};
+use support::fs::FileSystemError;
+
+use crate::shared::types::Port;
+
+#[derive(Debug, thiserror::Error)]
+#[allow(missing_docs)]
+pub enum ProviderError {
+    #[error("Failed to spawn node '{0}': {1}")]
+    NodeSpawningFailed(String, anyhow::Error),
+
+    #[error("Error running command '{0}': {1}")]
+    RunCommandError(String, anyhow::Error),
+
+    #[error("Duplicated node name: {0}")]
+    DuplicatedNodeName(String),
+
+    #[error(transparent)]
+    FileSystemError(#[from] FileSystemError),
+
+    #[error("Invalid script path for {0}")]
+    InvalidScriptPath(anyhow::Error),
+
+    #[error("Script with path {0} not found")]
+    ScriptNotFound(PathBuf),
+
+    #[error("File generation failed: {0}")]
+    FileGenerationFailed(anyhow::Error),
+
+    #[error("Failed to retrieve process ID for node '{0}'")]
+    ProcessIdRetrievalFailed(String),
+
+    #[error("Failed to pause node '{0}'")]
+    PauseNodeFailed(String),
+
+    #[error("Failed to resume node '{0}'")]
+    ResumeNodeFaied(String),
+
+    #[error("Failed to kill node '{0}'")]
+    KillNodeFailed(String),
+}
 
 #[async_trait]
 pub trait Provider {
-    async fn create_namespace(&mut self) -> Result<(), ProviderError>;
-    async fn get_node_ip(&self) -> Result<IpAddr, ProviderError>;
-    async fn get_port_mapping(
-        &mut self,
-        port: Port,
-        pod_name: String,
-    ) -> Result<Port, ProviderError>;
-    async fn get_node_info(&mut self, pod_name: String) -> Result<(IpAddr, Port), ProviderError>;
+    fn capabilities(&self) -> &ProviderCapabilities;
+
+    async fn namespaces(&self) -> HashMap<String, DynNamespace>;
+
+    async fn create_namespace(&self) -> Result<DynNamespace, ProviderError>;
+}
+
+pub type DynProvider = Arc<dyn Provider>;
+
+#[async_trait]
+pub trait ProviderNamespace {
+    fn id(&self) -> &str;
+
+    fn base_dir(&self) -> &PathBuf;
+
+    async fn nodes(&self) -> HashMap<String, DynNode>;
+
+    async fn spawn_node(&self, options: SpawnNodeOptions) -> Result<DynNode, ProviderError>;
+
+    async fn generate_files(&self, options: GenerateFilesOptions) -> Result<(), ProviderError>;
+
+    async fn destroy(&self) -> Result<(), ProviderError>;
+
+    async fn static_setup(&self) -> Result<(), ProviderError>;
+}
+
+pub type DynNamespace = Arc<dyn ProviderNamespace>;
+
+type ExecutionResult = Result<String, (ExitStatus, String)>;
+
+#[async_trait]
+pub trait ProviderNode {
+    fn name(&self) -> &str;
+
+    fn base_dir(&self) -> &PathBuf;
+
+    fn config_dir(&self) -> &PathBuf;
+
+    fn data_dir(&self) -> &PathBuf;
+
+    fn scripts_dir(&self) -> &PathBuf;
+
+    fn log_path(&self) -> &PathBuf;
+
+    async fn endpoint(&self) -> Result<(IpAddr, Port), ProviderError>;
+
+    async fn mapped_port(&self, port: Port) -> Result<Port, ProviderError>;
+
+    async fn logs(&self) -> Result<String, ProviderError>;
+
+    async fn dump_logs(&self, local_dest: PathBuf) -> Result<(), ProviderError>;
+
     async fn run_command(
         &self,
-        args: Vec<String>,
-        opts: NativeRunCommandOptions,
-    ) -> Result<RunCommandResponse, ProviderError>;
-    async fn run_script(
-        &mut self,
-        identifier: String,
-        script_path: String,
-        args: Vec<String>,
-    ) -> Result<RunCommandResponse, ProviderError>;
-    async fn spawn_from_def(
-        &mut self,
-        pod_def: PodDef,
-        files_to_copy: Vec<FileMap>,
-        keystore: String,
-        chain_spec_id: String,
-        db_snapshot: String,
-    ) -> Result<(), ProviderError>;
-    async fn copy_file_from_pod(
-        &mut self,
-        pod_file_path: PathBuf,
-        local_file_path: PathBuf,
-    ) -> Result<(), ProviderError>;
-    async fn create_resource(
-        &mut self,
-        resource_def: PodDef,
-        scoped: bool,
-        wait_ready: bool,
-    ) -> Result<(), ProviderError>;
-    async fn wait_node_ready(&mut self, node_name: String) -> Result<(), ProviderError>;
-    async fn get_node_logs(&mut self, node_name: String) -> Result<String, ProviderError>;
-    async fn dump_logs(&mut self, path: String, pod_name: String) -> Result<(), ProviderError>;
-    fn get_pause_args(&mut self, name: String) -> Vec<String>;
-    fn get_resume_args(&mut self, name: String) -> Vec<String>;
-    async fn restart_node(&mut self, name: String, timeout: u64) -> Result<bool, ProviderError>;
-    async fn get_help_info(&mut self) -> Result<(), ProviderError>;
-    async fn destroy_namespace(&mut self) -> Result<(), ProviderError>;
-    async fn get_logs_command(&mut self, name: String) -> Result<String, ProviderError>;
-    async fn put_local_magic_file(
+        options: RunCommandOptions,
+    ) -> Result<ExecutionResult, ProviderError>;
+
+    async fn run_script(&self, options: RunScriptOptions)
+        -> Result<ExecutionResult, ProviderError>;
+
+    async fn copy_file_from_node(
         &self,
-        _name: String,
-        _container: Option<String>,
-    ) -> Result<(), ProviderError> {
-        Ok(())
-    }
-    fn is_pod_monitor_available() -> Result<bool, ProviderError> {
-        Ok(false)
-    }
-    async fn spawn_introspector() -> Result<(), ProviderError> {
-        Ok(())
-    }
-
-    async fn static_setup() -> Result<(), ProviderError> {
-        Ok(())
-    }
-    async fn create_static_resource() -> Result<(), ProviderError> {
-        Ok(())
-    }
-    async fn create_pod_monitor() -> Result<(), ProviderError> {
-        Ok(())
-    }
-    async fn setup_cleaner() -> Result<(), ProviderError> {
-        Ok(())
-    }
-
-    #[allow(clippy::diverging_sub_expression)]
-    async fn upsert_cron_job() -> Result<(), ProviderError> {
-        unimplemented!();
-    }
+        remote_src: PathBuf,
+        local_dest: PathBuf,
+    ) -> Result<(), ProviderError>;
+
+    async fn pause(&self) -> Result<(), ProviderError>;
+
+    async fn resume(&self) -> Result<(), ProviderError>;
+
+    async fn restart(&self, after: Option<Duration>) -> Result<(), ProviderError>;
+
+    async fn destroy(&self) -> Result<(), ProviderError>;
 }
 
-// re-exports
-pub use native::NativeProvider;
+pub type DynNode = Arc<dyn ProviderNode + Send + Sync>;