Merge pull request #3306 from reubenmiller/fix-in-progress-workflows-ignored-after-self-update

fix: process non-versioned in-progress commands on startup

Author: reubenmiller

crates/core/tedge_api/src/workflow/supervisor.rs

@@ -76,7 +76,25 @@ impl WorkflowSupervisor {
     }
 
     /// Update on start the set of pending commands
-    pub fn load_pending_commands(&mut self, commands: CommandBoard) -> Vec<GenericCommandState> {
+    pub fn load_pending_commands(
+        &mut self,
+        mut commands: CommandBoard,
+    ) -> Vec<GenericCommandState> {
+        // If the resumed commands have been triggered by an agent without workflow version management
+        // then these commands are assigned the current version for the operation.
+        // These currents versions have also to be marked as in use.
+        for (_, ref mut command) in commands.iter_mut() {
+            if command.workflow_version().is_none() {
+                if let Some(versions) = command
+                    .operation()
+                    .and_then(|operation| self.workflows.get_mut(&operation.as_str().into()))
+                {
+                    if let Some(current_version) = versions.use_current_version() {
+                        *command = command.clone().set_workflow_version(current_version);
+                    }
+                }
+            }
+        }
         self.commands = commands;
         self.commands
             .iter()
@@ -279,8 +297,15 @@ impl WorkflowSupervisor {
         timestamp: &Timestamp,
         command: &GenericCommandState,
     ) -> Option<GenericCommandState> {
-        let Ok(action) = self.get_action(command) else {
-            return None;
+        let action = match self.get_action(command) {
+            Ok(action) => action,
+            Err(err) => {
+                return Some(
+                    command
+                        .clone()
+                        .fail_with(format!("Fail to resume on start: {err:?}")),
+                );
+            }
         };
 
         let epoch = format!("{}.{}", timestamp.unix_timestamp(), timestamp.millisecond());
@@ -464,6 +489,10 @@ impl CommandBoard {
         self.commands.values()
     }
 
+    pub fn iter_mut(&mut self) -> impl Iterator<Item = &mut (Timestamp, GenericCommandState)> {
+        self.commands.values_mut()
+    }
+
     /// Insert a new operation request into the [CommandBoard]
     ///
     /// Reject the request if there is already an entry with the same command id, but in a different state

tests/RobotFramework/tests/cumulocity/self-update/apt_package_pinning

@@ -0,0 +1,31 @@
+Package: tedge-full
+Pin: version %%VERSION%%
+Pin-Priority: 1001
+
+Package: tedge
+Pin: version %%VERSION%%
+Pin-Priority: 1001
+
+Package: tedge-mapper
+Pin: version %%VERSION%%
+Pin-Priority: 1001
+
+Package: tedge-agent
+Pin: version %%VERSION%%
+Pin-Priority: 1001
+
+Package: c8y-firmware-plugin
+Pin: version %%VERSION%%
+Pin-Priority: 1001
+
+Package: c8y-remote-access-plugin
+Pin: version %%VERSION%%
+Pin-Priority: 1001
+
+Package: tedge-apt-plugin
+Pin: version %%VERSION%%
+Pin-Priority: 1001
+
+Package: tedge-watchdog
+Pin: version %%VERSION%%
+Pin-Priority: 1001

tests/RobotFramework/tests/cumulocity/self-update/software_update.toml

@@ -0,0 +1,38 @@
+operation = "software_update"
+
+[init]
+action = "proceed"
+on_success = "executing"
+
+[executing]
+action = "proceed"
+on_success = "prepare"
+
+[prepare]
+script = "sudo /etc/tedge/sm-plugins/apt prepare"
+on_success = "self-update"
+
+[self-update]
+# Note: update tedge as it is used by the tedge-agent (as tedge is a multi-call binary)
+# Note: installing the new tedge package will not restart the tedge-agent service
+script = "sudo /etc/tedge/sm-plugins/apt install tedge"
+on_success = "restart-agent"
+
+[restart-agent]
+background_script = "sudo systemctl restart tedge-agent"
+on_exec = "await-restart-agent"
+
+[await-restart-agent]
+script = "sleep 10"
+on_success = "finalize"
+on_kill = "finalize"
+
+[finalize]
+script = "sh -c 'echo Executing state: ${.payload.status}'"
+on_success = "successful"
+
+[successful]
+action = "cleanup"
+
+[failed]
+action = "cleanup"

tests/RobotFramework/tests/cumulocity/self-update/tedge_self_update.robot

@@ -144,6 +144,49 @@ Update tedge version from base to current using Cumulocity
     ${pid_after}=    Service Should Be Running    tedge-agent
     Should Not Be Equal    ${pid_before}    ${pid_after}
 
+Update tedge Using a Custom Software Update Workflow
+    [Documentation]    thin-edge.io needs to support updating across a non-versioned workflow
+    ...    which occurs when users are updating from tedge <= 1.3.1 to tedge >= 1.4.0
+    ...    Once the new version of tedge-agent starts, it needs to recognize workflows
+    ...    that don't have a version (as the workflow version feature did not exist prior to tedge 1.4.0)
+    ${PREV_VERSION}=    Set Variable    1.3.1
+
+    # Install base version (using apt package pinning, then remove it)
+    Pin thin-edge.io APT Packages    ${PREV_VERSION}
+    Execute Command    cmd=curl -fsSL thin-edge.io/install.sh | sh -s
+    Unpin thin-edge.io APT Packages
+
+    # Register device (using already installed version)
+    Execute Command
+    ...    cmd=test -f ./bootstrap.sh && env DEVICE_ID=${DEVICE_SN} ./bootstrap.sh --no-install --no-secure || true
+    Device Should Exist    ${DEVICE_SN}
+
+    Device Should Have Installed Software
+    ...    tedge-agent,${PREV_VERSION}
+
+    ${agent_version}=    Execute Command    tedge-agent --version    strip=${True}
+    Should Be Equal As Strings    ${agent_version}    tedge-agent ${PREV_VERSION}
+
+    # Allow tedge user to restart services (used in the workflow)
+    Execute Command
+    ...    cmd=echo "tedge ALL = (ALL) NOPASSWD: /usr/bin/systemctl restart *" | sudo tee /etc/sudoers.d/tedge_admin
+    # Copy Workflow
+    Transfer To Device    ${CURDIR}/software_update.toml    /etc/tedge/operations/
+    # Reload the workflows (as tedge <= 1.3.1 did not support reloading workflows at runtime)
+    Restart Service    tedge-agent
+
+    # Configure local repository containing the new version
+    Create Local Repository
+
+    # Note: this just trigger the operation, the contents does not
+    # actually matter as the workflow uses some hardcoded values
+    ${operation}=    Install Software
+    ...    tedge-agent,${NEW_VERSION}
+    Operation Should Be SUCCESSFUL    ${operation}    timeout=120
+
+    ${agent_version}=    Execute Command    tedge-agent --version    strip=${True}
+    Should Be Equal As Strings    ${agent_version}    tedge-agent ${NEW_VERSION}
+
 
 *** Keywords ***
 Custom Setup
@@ -172,3 +215,13 @@ Create Local Repository
     Execute Command    cd /opt/repository/local && dpkg-scanpackages -m . > Packages
     Execute Command
     ...    cmd=echo 'deb [trusted=yes] file:/opt/repository/local /' > /etc/apt/sources.list.d/tedge-local.list
+
+Pin thin-edge.io APT Packages
+    [Arguments]    ${VERSION}
+    Transfer To Device    ${CURDIR}/apt_package_pinning    /etc/apt/preferences.d/tedge
+    Execute Command    cmd=sed -i 's/%%VERSION%%/${VERSION}/g' /etc/apt/preferences.d/tedge
+
+Unpin thin-edge.io APT Packages
+    Execute Command    cmd=rm -f /etc/apt/preferences.d/tedge
+    # Remove thin-edge.io public repositories to avoid affecting the selected version
+    Execute Command    cmd=rm -f /etc/apt/sources.list.d/thinedge-*.list