Gpu hardening master (#2753)

* Add non interactive setup to golemsp

* Auto remove directory after Agreement finish

* Auto cleanup after activity finish

* Enable tests for gpu-hardening-master branch

* Add golemsp to deb (provider variant)

* Stabilize rust version in aarch64 builds

* Revert changes to enable tests on gpu-hardening child branches

---------

Co-authored-by: scx1332 <sieciech.czajka@golem.network>

.github/workflows/release.yml

@@ -181,7 +181,7 @@ jobs:
       - name: Setup toolchain
         uses: actions-rs/toolchain@v1
         with:
-          toolchain: stable
+          toolchain: ${{ env.rust_stable }}
           target: aarch64-unknown-linux-musl
           override: true
 

Cargo.toml

@@ -153,6 +153,11 @@ assets = [
         "usr/lib/yagna/plugins/",
         "755",
     ],
+    [
+        "target/release/golemsp",
+        "usr/bin/",
+        "755",
+    ],
     [
         "README.md",
         "usr/share/doc/yagna/",

agent/provider/src/execution/task_runner.rs

@@ -8,10 +8,10 @@ use humantime;
 use log_derive::{logfn, logfn_inputs};
 use std::collections::HashMap;
 use std::fs::{create_dir_all, File};
-use std::iter;
 use std::path::{Path, PathBuf};
 use std::sync::Arc;
 use std::time::Duration;
+use std::{fs, iter};
 use structopt::StructOpt;
 
 use ya_agreement_utils::{AgreementView, OfferTemplate};
@@ -106,7 +106,7 @@ pub struct TerminateActivity {
     pub message: String,
 }
 
-/// Called when process exited. There are 3 reasons for process to exit:
+/// Called when process exited. There are 2 reasons for process to exit:
 /// - We got DestroyActivity event and killed process.
 /// - ExeUnit crashed.
 #[derive(Message)]
@@ -128,6 +128,13 @@ pub struct TaskRunnerConfig {
     pub process_termination_timeout: Duration,
     #[structopt(long, env, parse(try_from_str = humantime::parse_duration), default_value = "10s")]
     pub exeunit_state_retry_interval: Duration,
+    /// Removes directory content after each `Activity` is destroyed.
+    #[structopt(long, env)]
+    pub auto_cleanup_activity: bool,
+    /// Removes directory content after `Agreement` is terminated.
+    /// Use this option to save disk space. Shouldn't be used when debugging.
+    #[structopt(long, env)]
+    pub auto_cleanup_agreement: bool,
     #[structopt(skip = "you-forgot-to-set-session-id")]
     pub session_id: String,
 }
@@ -370,6 +377,19 @@ impl TaskRunner {
             msg.activity_id
         );
 
+        if self.config.auto_cleanup_activity {
+            let workdir = self
+                .agreement_dir(&msg.agreement_id)
+                .secure_join(&msg.activity_id);
+            log::info!(
+                "Cleaning directory {} for agreement [{}], activity [{}].",
+                workdir.display(),
+                msg.agreement_id,
+                msg.activity_id
+            );
+            fs::remove_dir_all(workdir).ok();
+        }
+
         let destroy_msg = ActivityDestroyed {
             agreement_id: msg.agreement_id.to_string(),
             activity_id: msg.activity_id,
@@ -408,6 +428,10 @@ impl TaskRunner {
         })
     }
 
+    fn agreement_dir(&self, agreement_id: &str) -> PathBuf {
+        self.tasks_dir.secure_join(agreement_id)
+    }
+
     #[logfn(Debug, fmt = "Task created: {}")]
     fn create_task(
         &self,
@@ -416,10 +440,7 @@ impl TaskRunner {
         agreement_id: &str,
         requestor_pub_key: Option<&str>,
     ) -> Result<Task> {
-        let working_dir = self
-            .tasks_dir
-            .secure_join(agreement_id)
-            .secure_join(activity_id);
+        let working_dir = self.agreement_dir(agreement_id).secure_join(activity_id);
 
         create_dir_all(&working_dir).map_err(|error| {
             anyhow!(
@@ -795,6 +816,16 @@ impl Handler<AgreementClosed> for TaskRunner {
 
         self.active_agreements.remove(&agreement_id);
 
+        if self.config.auto_cleanup_agreement {
+            let workdir = self.agreement_dir(&agreement_id);
+            log::info!(
+                "Cleaning directory {} for agreement [{}].",
+                workdir.display(),
+                agreement_id
+            );
+            fs::remove_dir_all(workdir).ok();
+        }
+
         // All activities should be destroyed by now, so it is only sanity call.
         let remove_future = async move {
             remove_remaining_tasks(activities, agreement_id, myself).await;
@@ -806,21 +837,19 @@ impl Handler<AgreementClosed> for TaskRunner {
 }
 
 impl Handler<AgreementBroken> for TaskRunner {
-    type Result = ActorResponse<Self, Result<(), Error>>;
+    type Result = ResponseFuture<Result<(), Error>>;
 
     fn handle(&mut self, msg: AgreementBroken, ctx: &mut Context<Self>) -> Self::Result {
-        let agreement_id = msg.agreement_id;
-        let myself = ctx.address();
-        let activities = self.list_activities(&agreement_id);
-
-        self.active_agreements.remove(&agreement_id);
-
-        let remove_future = async move {
-            remove_remaining_tasks(activities, agreement_id, myself).await;
-            Ok(())
-        };
-
-        ActorResponse::r#async(remove_future.into_actor(self))
+        // We don't distinguish between `AgreementClosed` and `AgreementBroken`.
+        let addr = ctx.address();
+        async move {
+            addr.send(AgreementClosed {
+                agreement_id: msg.agreement_id,
+                send_terminate: false,
+            })
+            .await?
+        }
+        .boxed_local()
     }
 }
 

golem_cli/src/main.rs

@@ -125,15 +125,20 @@ async fn my_main() -> Result</*exit code*/ i32> {
     }
 }
 
-pub fn banner() {
-    terminal::fade_in(&format!(
+pub fn banner(interactive: bool) {
+    let banner = format!(
         include_str!("banner.txt"),
         version = ya_compile_time_utils::semver_str!(),
         git_commit = ya_compile_time_utils::git_rev(),
         date = ya_compile_time_utils::build_date(),
         build = ya_compile_time_utils::build_number_str().unwrap_or("-"),
-    ))
-    .unwrap();
+    );
+
+    if interactive {
+        terminal::fade_in(&banner).unwrap();
+    } else {
+        println!("{banner}")
+    }
 }
 
 #[actix_rt::main]

golem_cli/src/setup.rs

@@ -42,11 +42,15 @@ pub struct RunConfig {
         set = clap::ArgSettings::Global
     )]
     pub log_dir: Option<PathBuf>,
+    /// Don't prompt user for configuration, but use defaults instead.
+    #[structopt(long)]
+    pub no_interactive: bool,
 }
 
 pub async fn setup(run_config: &RunConfig, force: bool) -> Result<i32> {
+    let interactive = !run_config.no_interactive;
     if force {
-        super::banner();
+        super::banner(interactive);
         eprintln!("Initial node setup");
         let _ = clear_stdin().await;
     }
@@ -71,13 +75,20 @@ pub async fn setup(run_config: &RunConfig, force: bool) -> Result<i32> {
     }
 
     if config.node_name.is_none() || force {
-        let node_name = promptly::prompt_default(
-            "Node name ",
-            config
-                .node_name
-                .clone()
-                .unwrap_or_else(|| names::Generator::default().next().unwrap_or_default()),
-        )?;
+        let node_name = if interactive {
+            promptly::prompt_default(
+                "Node name ",
+                config
+                    .node_name
+                    .clone()
+                    .unwrap_or_else(|| names::Generator::default().next().unwrap_or_default()),
+            )?
+        } else {
+            let name = names::Generator::default().next().unwrap_or_default();
+            println!("Node name (default={name})");
+            name
+        };
+
         let account_msg = &config
             .account
             .map(|n| n.to_string())
@@ -87,14 +98,18 @@ pub async fn setup(run_config: &RunConfig, force: bool) -> Result<i32> {
             run_config.account.network, account_msg
         );
 
-        while let Some(account) = promptly::prompt_opt::<String, _>(&message)? {
-            match account.parse::<NodeId>() {
-                Err(e) => eprintln!("Invalid ethereum address, is should be 20-byte hex (example 0xB1974E1F44EAD2d22bB995167A709b89Fc466B6c): {}", e),
-                Ok(account) => {
-                    config.account = Some(account);
-                    break;
+        if interactive {
+            while let Some(account) = promptly::prompt_opt::<String, _>(&message)? {
+                match account.parse::<NodeId>() {
+                    Err(e) => eprintln!("Invalid ethereum address, is should be 20-byte hex (example 0xB1974E1F44EAD2d22bB995167A709b89Fc466B6c): {}", e),
+                    Ok(account) => {
+                        config.account = Some(account);
+                        break;
+                    }
                 }
             }
+        } else {
+            println!("{message}");
         }
 
         config.node_name = Some(node_name);
@@ -132,7 +147,12 @@ pub async fn setup(run_config: &RunConfig, force: bool) -> Result<i32> {
             .collect();
 
         let default_glm_per_h = 0.025;
-        let glm_per_h = promptly::prompt_default("Price GLM per hour", default_glm_per_h)?;
+        let glm_per_h = if interactive {
+            promptly::prompt_default("Price GLM per hour", default_glm_per_h)?
+        } else {
+            println!("Price GLM per hour (default={default_glm_per_h})");
+            default_glm_per_h
+        };
 
         let mut usage = UsageDef::new();
         usage.insert("CPU".into(), glm_per_h / 3600.0);