add check that docker image supports compute architecture of cluster (#2)

Author: Panaetius

Cargo.lock

@@ -1,5 +1,5 @@
 [cscs]
-system = "daint"
+current_system = "daint"
 
 image = "ubuntu"
 
@@ -17,3 +17,11 @@ edf_file_template = """
 image = "{{edf_image}}"
 mounts = ["${SCRATCH}:/scratch"]
 """
+
+[cscs.systems]
+
+[cscs.systems.daint]
+architecture = ["arm64"]
+
+[cscs.systems.eiger]
+architecture = ["amd64"]

coman/.config/config.toml

@@ -65,6 +65,8 @@ tabled = { version = "0.20.0", features = ["macros"] }
 nom = "8.0.0"
 tera = "1.20.1"
 inquire = "0.9.1"
+oci-distribution = "0.11.0"
+docker_credential = "1.3.2"
 
 [build-dependencies]
 anyhow = "1.0.90"

coman/Cargo.toml

@@ -11,7 +11,6 @@ pub enum UserEvent {
     Cscs(CscsEvent),
     Error(String),
     Info(String),
-    None, // this is mainly used to return a nop result that keeps a port alive, as returning no Event stops the port
 }
 
 impl PartialEq for UserEvent {

coman/src/app/user_events.rs

@@ -1,6 +1,6 @@
 #![allow(dead_code)] // Remove this once you start using the code
 
-use std::{env, path::PathBuf};
+use std::{collections::HashMap, env, path::PathBuf};
 
 use color_eyre::Result;
 use directories::ProjectDirs;
@@ -9,6 +9,11 @@ use serde::{Deserialize, Serialize};
 
 const DEFAULT_CONFIG_TOML: &str = include_str!("../.config/config.toml");
 
+#[derive(Clone, Debug, Serialize, Deserialize, Default)]
+pub struct SystemDescription {
+    pub architecture: Vec<String>,
+}
+
 #[derive(Clone, Debug, Serialize, Deserialize, Default)]
 pub struct AppConfig {
     #[serde(default)]
@@ -20,7 +25,7 @@ pub struct AppConfig {
 #[derive(Clone, Debug, Serialize, Deserialize, Default)]
 pub struct CscsConfig {
     #[serde(default)]
-    pub system: String,
+    pub current_system: String,
     #[serde(default)]
     pub name: Option<String>,
     #[serde(default)]
@@ -31,7 +36,11 @@ pub struct CscsConfig {
     pub edf_file_template: String,
     #[serde(default)]
     pub command: Vec<String>,
+
+    #[serde(default)]
+    pub systems: HashMap<String, SystemDescription>,
 }
+
 #[derive(Clone, Debug, Default, Serialize, Deserialize)]
 pub struct Config {
     #[serde(default, flatten)]

coman/src/config.rs

@@ -74,6 +74,7 @@ impl From<CSCSFileSystem> for FileSystem {
 
 #[derive(Debug, Eq, Clone, PartialEq, PartialOrd, Ord, Display)]
 pub enum JobStatus {
+    Pending,
     Running,
     Finished,
     Cancelled,
@@ -96,6 +97,7 @@ impl From<JobModelOutput> for Job {
                 "FAILED" => JobStatus::Failed,
                 "COMPLETED" => JobStatus::Finished,
                 "CANCELLED" => JobStatus::Cancelled,
+                "PENDING" => JobStatus::Pending,
                 other => panic!("got job status: {}", other),
             },
             user: value.user.unwrap_or("".to_string()),

coman/src/cscs/api_client.rs

@@ -6,8 +6,8 @@ use crate::{
     cscs::{
         api_client::{CscsApi, FileSystemType, Job, JobDetail, System},
         oauth2::{
-            CLIENT_ID_SECRET_NAME, CLIENT_SECRET_SECRET_NAME,
-            client_credentials_login, finish_cscs_device_login, start_cscs_device_login,
+            CLIENT_ID_SECRET_NAME, CLIENT_SECRET_SECRET_NAME, client_credentials_login,
+            finish_cscs_device_login, start_cscs_device_login,
         },
     },
     util::{
@@ -61,17 +61,22 @@ pub async fn cscs_job_list() -> Result<Vec<Job>> {
         Ok(access_token) => {
             let api_client = CscsApi::new(access_token.0).unwrap();
             let config = Config::new().unwrap();
-            api_client.list_jobs(&config.cscs.system, Some(true)).await
+            api_client
+                .list_jobs(&config.cscs.current_system, Some(true))
+                .await
         }
         Err(e) => Err(e),
     }
 }
+
 pub async fn cscs_job_details(job_id: i64) -> Result<Option<JobDetail>> {
     match get_access_token().await {
         Ok(access_token) => {
             let api_client = CscsApi::new(access_token.0).unwrap();
             let config = Config::new().unwrap();
-            api_client.get_job(&config.cscs.system, job_id).await
+            api_client
+                .get_job(&config.cscs.current_system, job_id)
+                .await
         }
         Err(e) => Err(e),
     }
@@ -86,9 +91,9 @@ pub async fn cscs_start_job(
         Ok(access_token) => {
             let api_client = CscsApi::new(access_token.0).unwrap();
             let config = Config::new().unwrap();
-            let user_info = api_client.get_userinfo(&config.cscs.system).await?;
-            let system = api_client.get_system(&config.cscs.system).await?;
-            let scratch = match system {
+            let user_info = api_client.get_userinfo(&config.cscs.current_system).await?;
+            let current_system = api_client.get_system(&config.cscs.current_system).await?;
+            let scratch = match current_system {
                 Some(system) => PathBuf::from(
                     system
                         .file_systems
@@ -101,7 +106,7 @@ pub async fn cscs_start_job(
                 None => {
                     return Err(eyre!(
                         "couldn't get system description for {}",
-                        config.cscs.system
+                        config.cscs.current_system
                     ));
                 }
             };
@@ -113,21 +118,43 @@ pub async fn cscs_start_job(
             let environment_path = base_path.join("environment.toml");
             let environment_template = config.cscs.edf_file_template;
             tera.add_raw_template("environment.toml", &environment_template)?;
+
+            let docker_image = image.unwrap_or(config.cscs.image.try_into()?);
+            let meta = docker_image.inspect().await?;
+            if let Some(system_info) = config.cscs.systems.get(&config.cscs.current_system) {
+                let mut compatible = false;
+                for sys_platform in system_info.architecture.iter() {
+                    if meta.platforms.contains(&sys_platform.clone().into()) {
+                        compatible = true;
+                    }
+                }
+
+                if !compatible {
+                    return Err(eyre!(
+                        "System {} only supports images with architecture(s) '{}' but the supplied image is for architecture(s) '{}'",
+                        config.cscs.current_system,
+                        system_info.architecture.join(","),
+                        meta.platforms
+                            .iter()
+                            .map(|p| p.to_string())
+                            .collect::<Vec<String>>()
+                            .join(",")
+                    ));
+                }
+            }
+
             let mut context = tera::Context::new();
-            context.insert(
-                "edf_image",
-                &image.unwrap_or(config.cscs.image.try_into()?).to_edf(),
-            );
+            context.insert("edf_image", &docker_image.to_edf());
             let environment_file = tera.render("environment.toml", &context)?;
             api_client
-                .mkdir(&config.cscs.system, base_path.clone())
+                .mkdir(&config.cscs.current_system, base_path.clone())
                 .await?;
             api_client
-                .chmod(&config.cscs.system, base_path.clone(), "700")
+                .chmod(&config.cscs.current_system, base_path.clone(), "700")
                 .await?;
             api_client
                 .upload(
-                    &config.cscs.system,
+                    &config.cscs.current_system,
                     environment_path.clone(),
                     environment_file.into_bytes(),
                 )
@@ -150,15 +177,15 @@ pub async fn cscs_start_job(
             let script = tera.render("script.sh", &context)?;
             api_client
                 .upload(
-                    &config.cscs.system,
+                    &config.cscs.current_system,
                     script_path.clone(),
                     script.into_bytes(),
                 )
                 .await?;
 
             // start job
             api_client
-                .start_job(&config.cscs.system, &name, script_path)
+                .start_job(&config.cscs.current_system, &name, script_path)
                 .await?;
             Ok(())
         }