[CERTTF-452] Modify submit action behaviour for reservation jobs (#400

) * feat: introduce action step for reservation jobs * feat: add workflow groups * feat: include `JOB_ID` in group messages and notice * feat: add notice with exit status * feat: retrieve and return IP of reserved machine * docs: extend README to describe polling behaviour for reservations
canonical · Nov 12, 2024 · bb8982a · bb8982a
1 parent a77c0ee
commit bb8982a
Show file tree

Hide file tree

Showing 2 changed files with 90 additions and 9 deletions.
diff --git a/.github/actions/submit/action.yaml b/.github/actions/submit/action.yaml
@@ -26,6 +26,9 @@ outputs:
   id:
     description: 'The ID of the submitted job'
     value: ${{ steps.submit.outputs.id }}
+  device-ip:
+    description: 'The IP of the reserved device (if applicable)'
+    value: ${{ steps.track-reservation.outputs.ip }}
 runs:
   using: composite
   steps:
@@ -34,6 +37,7 @@ runs:
       env:
         SERVER: https://${{ inputs.server }}
       run: |
+        echo "::group::Test connection to Testflinger server"
         STATUS=$(curl --stderr error.log -Ivw "%{http_code}\n" -o /dev/null $SERVER/jobs)
         ERROR=$?
         if [ ! $ERROR -eq 0 ]; then
@@ -47,11 +51,13 @@ runs:
         else
           echo "Successful server ping at $SERVER, status: $STATUS"
         fi
+        echo "::endgroup::"
 
     - name: Create job file, if required
       id: create-job-file
       shell: bash
       run: |
+        echo "::group::Create job file (if required)"
         if [ -n "${{ inputs.job-path }}" ]; then
         # the `$JOB` environment variable points to the job path provided as input
         echo "JOB=${{ inputs.job-path }}" >> $GITHUB_ENV
@@ -69,14 +75,34 @@ runs:
         # the `$JOB` environment variable points to the newly created job file
         echo "JOB=$FILE" >> $GITHUB_ENV
         fi
+        echo "::endgroup::"
 
     - name: Display contents of job file (for verification)
       shell: bash
-      run: cat "$JOB"
+      run: |
+        echo "::group::Display job file"
+        cat "$JOB"
+        echo "::endgroup::"
+
+    - name: Determine if this is a reservation job
+      id: check-reservation
+      shell: bash
+      run: |
+        echo "::group::Determine if this is a reservation job"
+        if grep -q "^reserve_data:" $JOB; then
+          RESERVATION=true
+        else
+          RESERVATION=false
+        fi
+        echo "reservation=$RESERVATION" | tee -a $GITHUB_OUTPUT
+        echo "::endgroup::"
 
     - name: Install prerequisites
       shell: bash
-      run: sudo snap install testflinger-cli jq
+      run: |
+        echo "::group::Install prerequisites"
+        sudo snap install testflinger-cli jq
+        echo "::endgroup::"
 
     - name: Submit job to the Testflinger server
       id: submit
@@ -86,26 +112,30 @@ runs:
         SERVER: https://${{ inputs.server }}
         RELATIVE_TO: ${{ inputs.attachments-relative-to }}
       run: |
+        echo "::group::Submit job to the Testflinger server"
         if [ -n "$RELATIVE_TO" ]; then
           RELATIVE="--attachments-relative-to $RELATIVE_TO"
         fi
         JOB_ID=$(testflinger --server $SERVER submit --quiet "$RELATIVE" "$JOB")
-        echo "job id: $JOB_ID"
         echo "id=$JOB_ID" >> $GITHUB_OUTPUT
+        echo "::endgroup::"
+        echo "::notice::Submitted job $JOB_ID"
 
-    - name: Track the status of the job and mirror its exit status
-      if: inputs.poll == 'true' && inputs.dry-run != 'true'
+    - name: Track the status of the job (non-reservation)
+      if: inputs.poll == 'true' && inputs.dry-run != 'true' && steps.check-reservation.outputs.reservation != 'true'
       shell: bash
       env:
         SERVER: https://${{ inputs.server }}
         JOB_ID: ${{ steps.submit.outputs.id }}
       run: |
+        echo "::group::Track the status of job ${{ env.JOB_ID }}"
         # poll
         PYTHONUNBUFFERED=1 testflinger --server $SERVER poll $JOB_ID
-        # retrieve results:
+        echo "::endgroup::"
+        echo "::group::Retrieve job results, determine exit status"
         # the exit status is the maximum status of the individual test phases
         # (excluding the setup and cleanup phases)
-        STATUS=$(\
+        EXIT_STATUS=$(\
           testflinger --server $SERVER results $JOB_ID  \
           | jq 'to_entries
               | map(
@@ -117,8 +147,48 @@ runs:
                 )
               | max'
         )
-        echo "Test exit status: $STATUS"
-        exit $STATUS
+        echo "::endgroup::"
+        echo "::notice::Test exit status: $EXIT_STATUS"
+        exit $EXIT_STATUS
+
+    - name: Track the status of the job (reservation)
+      id: track-reservation
+      if: inputs.poll == 'true' && inputs.dry-run != 'true' && steps.check-reservation.outputs.reservation == 'true'
+      shell: bash
+      env:
+        SERVER: https://${{ inputs.server }}
+        JOB_ID: ${{ steps.submit.outputs.id }}
+        TERMINATION: "TESTFLINGER SYSTEM RESERVED"
+      run: |
+        echo "::group::Track the status of reservation job ${{ env.JOB_ID }}"
+        # specify file for capturing output and clear it
+        CAPTURED_OUTPUT=output
+        > "$CAPTURED_OUTPUT"
+        # perform one-shot polling until the reserve phase has been reached
+        # and the captured output contains the TERMINATION string
+        while true; do
+          PYTHONUNBUFFERED=1 testflinger --server $SERVER poll --oneshot $JOB_ID | tee -a "$CAPTURED_OUTPUT"
+          JOB_STATUS="$(testflinger --server $SERVER status $JOB_ID)"
+          if [ "$JOB_STATUS" = "reserve" ]; then
+            if grep -q "$TERMINATION" "$CAPTURED_OUTPUT"; then
+              EXIT_STATUS=0
+              break
+            fi
+          elif [[ "$JOB_STATUS" =~ ^(complete|cancelled)$ ]]; then
+            EXIT_STATUS=1
+            break
+          fi
+          sleep 10
+        done
+        echo "::endgroup::"
+        echo "::group::Retrieve IP of reserved machine"
+        DEVICE_IP=$(grep -m 1 "^Now try logging into the machine" "$CAPTURED_OUTPUT" | sed -n "s/.*@\([0-9.]*\).*/\1/p")
+        [ "$?" -eq 0 ] && DEVICE_IP_FLAG=true
+        [ -n "$DEVICE_IP_FLAG" ] && echo "ip=$DEVICE_IP" >> $GITHUB_OUTPUT
+        echo "::endgroup::"
+        echo "::notice::Reserve exit status: $EXIT_STATUS"
+        [ -n "$DEVICE_IP_FLAG" ] && echo "::notice::Device IP: $DEVICE_IP"
+        exit $EXIT_STATUS
 
     - name: Cancel the job if the action is cancelled
       if: ${{ cancelled() }}
@@ -127,4 +197,6 @@ runs:
         SERVER: https://${{ inputs.server }}
         JOB_ID: ${{ steps.submit.outputs.id }}
       run: |
+        echo "::group::Cancel job"
         testflinger --server $SERVER cancel $JOB_ID
+        echo "::endgroup::"
diff --git a/README.md b/README.md
@@ -74,3 +74,12 @@ in any of the subsequent steps of the workflow:
         testflinger results ${{ steps.submit-job.outputs.id }}"
 ```
 In this example, `submit-job` is the step where the `submit` action is used.
+
+If the submitted job is a reservation job (i.e. includes `reserve_data`) then
+setting the `poll` input argument to `true` results in modified behaviour: the
+job is polled only until the reservation phase is complete, instead of waiting
+for the entire job to complete (which happens when the reservation timeout
+expires or the job is cancelled). There will be no output to record after the
+reservation so there is little point in polling and idly occupying the runner.
+However, please do remember to manually cancel the job after you are done with
+the reserved device.