Merge pull request #15 from avnik/avnik/fix-restart

fix restart crashed applications
tiiuae · Sep 11, 2024 · 6a85012 · 6a85012
2 parents d3291cc + 06f5849
commit 6a85012
Show file tree

Hide file tree

Showing 3 changed files with 31 additions and 12 deletions.
diff --git a/common/src/types.rs b/common/src/types.rs
@@ -149,7 +149,12 @@ pub struct UnitStatus {
 
 impl UnitStatus {
     pub fn is_running(&self) -> bool {
-        self.active_state == "active"
+        self.active_state == "active" && self.load_state == "loaded" && self.sub_state == "running"
+    }
+    pub fn is_exitted(&self) -> bool {
+        self.active_state == "inactive"
+            && self.load_state == "not-found"
+            && self.sub_state == "dead"
     }
 }
 

diff --git a/nixos/tests/admin.nix b/nixos/tests/admin.nix
@@ -200,7 +200,10 @@ in
                   admin = adminSettings;
                   tls = mkTls "chromium-vm";
                   applications = lib.mkForce (
-                    builtins.toJSON { "foot" = "/run/current-system/sw/bin/run-waypipe ${pkgs.foot}/bin/foot"; }
+                    builtins.toJSON {
+                      "foot" = "/run/current-system/sw/bin/run-waypipe ${pkgs.foot}/bin/foot";
+                      "clearexit" = "/run/current-system/sw/bin/sleep 5";
+                    }
                   );
                 };
               };
@@ -288,10 +291,23 @@ in
                   swaymsg("exec ssh -R /tmp/vsock:/tmp/vsock -f -N ${addrs.appvm}")
                   time.sleep(5) # Give ssh some time to setup remote socket
 
-              #swaymsg("exec run-waypipe foot")
-              print(hostvm.succeed("${cli} --addr ${nodes.adminvm.config.givc.admin.addr} --port ${nodes.adminvm.config.givc.admin.port} --cacert ${nodes.hostvm.givc.host.tls.caCertPath} --cert ${nodes.hostvm.givc.host.tls.certPath} --key ${nodes.hostvm.givc.host.tls.keyPath} ${if tls then "" else "--notls"} --name ${nodes.adminvm.config.givc.admin.name} start foot"))
-              time.sleep(10) # Give few seconds to application to spin up
-              wait_for_window("ghaf@appvm")
+              with subtest("Clean run"):
+                  print(hostvm.succeed("${cli} --addr ${nodes.adminvm.config.givc.admin.addr} --port ${nodes.adminvm.config.givc.admin.port} --cacert ${nodes.hostvm.givc.host.tls.caCertPath} --cert ${nodes.hostvm.givc.host.tls.certPath} --key ${nodes.hostvm.givc.host.tls.keyPath} ${if tls then "" else "--notls"} --name ${nodes.adminvm.config.givc.admin.name} start foot"))
+                  time.sleep(10) # Give few seconds to application to spin up
+                  wait_for_window("ghaf@appvm")
+
+              with subtest("crash and restart"):
+                  # Crash application
+                  appvm.succeed("pkill foot")
+                  time.sleep(10)
+                  # .. then ask to restart
+                  print(hostvm.succeed("${cli} --addr ${nodes.adminvm.config.givc.admin.addr} --port ${nodes.adminvm.config.givc.admin.port} --cacert ${nodes.hostvm.givc.host.tls.caCertPath} --cert ${nodes.hostvm.givc.host.tls.certPath} --key ${nodes.hostvm.givc.host.tls.keyPath} ${if tls then "" else "--notls"} --name ${nodes.adminvm.config.givc.admin.name} start foot"))
+                  wait_for_window("ghaf@appvm")
+
+              with subtest("clear exit and restart"):
+                  print(hostvm.succeed("${cli} --addr ${nodes.adminvm.config.givc.admin.addr} --port ${nodes.adminvm.config.givc.admin.port} --cacert ${nodes.hostvm.givc.host.tls.caCertPath} --cert ${nodes.hostvm.givc.host.tls.certPath} --key ${nodes.hostvm.givc.host.tls.keyPath} ${if tls then "" else "--notls"} --name ${nodes.adminvm.config.givc.admin.name} start --vm foot-vm clearexit"))
+                  time.sleep(20) # Give few seconds to application to spin up, exit, then start it again
+                  print(hostvm.succeed("${cli} --addr ${nodes.adminvm.config.givc.admin.addr} --port ${nodes.adminvm.config.givc.admin.port} --cacert ${nodes.hostvm.givc.host.tls.caCertPath} --cert ${nodes.hostvm.givc.host.tls.certPath} --key ${nodes.hostvm.givc.host.tls.keyPath} ${if tls then "" else "--notls"} --name ${nodes.adminvm.config.givc.admin.name} start --vm foot-vm clearexit"))
             '';
         };
       };

diff --git a/src/admin/server.rs b/src/admin/server.rs
@@ -160,7 +160,9 @@ impl AdminServiceImpl {
     pub async fn handle_error(&self, entry: RegistryEntry) -> anyhow::Result<()> {
         match (entry.r#type.vm, entry.r#type.service) {
             (VmType::AppVM, ServiceType::App) => {
-                self.registry.deregister(&entry.name)?;
+                if entry.status.is_exitted() {
+                    self.registry.deregister(&entry.name)?;
+                }
                 Ok(())
             }
             (VmType::AppVM, ServiceType::Mgr) | (VmType::SysVM, ServiceType::Mgr) => {
@@ -184,11 +186,7 @@ impl AdminServiceImpl {
             debug!("Monitoring {}...", &entry.name);
             match self.get_remote_status(&entry).await {
                 Err(err) => {
-                    error!(
-                        "could not get status of unit {}: {}",
-                        entry.name.clone(),
-                        err
-                    );
+                    error!("could not get status of unit {}: {}", &entry.name, err);
                     self.handle_error(entry)
                         .await
                         .with_context(|| "during handle error")?