Applicable scenario

• Product: ZCF
• Component: Cloud
• Version: 5.5.x
• Deployment topology: dual management nodes
• Trigger: A VM loses its GPU after DRS migration and requires replacement of management-node JAR packages.

Symptoms

• A VM loses its GPU after DRS migration.
• The remediation requires replacing both mevoco-5.5.x.jar and drs-5.5.x.jar.
• The change does not affect continued VM operation but requires a management-platform service restart.

Determination method

1. Confirm that the GPU state is abnormal after DRS migration.
2. After replacement, check that related services and DRS behavior return to normal.
3. Confirm management-platform services recover before re-enabling HA.

Cause

The issue requires synchronized replacement of the mevoco and drs core JAR packages, indicating that GPU loss involves both migration orchestration and related network/resource handling logic.

Resolution

1. Preparation

• Risk level: high.
• Disable platform HA.
• Back up the database.
• Stop services, back up, replace, and restart on both management nodes.

2. Procedure
3. Disable platform HA.

• Expected result: The platform enters a maintainable state.

2. Back up the database:

   zstack-ctl dump_mysql --file-name 4.x-mn1
   zstack-ctl dump_mysql --file-name 4.x-mn2

• Expected result: A database rollback baseline is available.

3. Stop management services on both management nodes:

   zsha2 stop-node

• Expected result: JAR packages can be replaced safely.

4. Back up the original JAR packages:

   cp /usr/local/zstack/apache-tomcat/webapps/zstack/WEB-INF/lib/mevoco-5.5.0.jar /root/mevoco-5.5.0.jar.bak
   cp /usr/local/zstack/apache-tomcat/webapps/zstack/WEB-INF/lib/drs-5.5.0.jar /root/drs-5.5.0.jar.bak

• Expected result: Rollback versions are retained.

5. Replace the JAR packages:

   cp mevoco-5.5.0.jar /usr/local/zstack/apache-tomcat/webapps/zstack/WEB-INF/lib/mevoco-5.5.0.jar
   cp drs-5.5.0.jar /usr/local/zstack/apache-tomcat/webapps/zstack/WEB-INF/lib/drs-5.5.0.jar

• Expected result: The fix packages are overlaid.

6. Correct file ownership:

   chown zstack:zstack /usr/local/zstack/apache-tomcat/webapps/zstack/WEB-INF/lib/drs-5.5.0.jar
   chown zstack:zstack /usr/local/zstack/apache-tomcat/webapps/zstack/WEB-INF/lib/mevoco-5.5.0.jar

• Expected result: Ownership is correct.

7. Restart management services:

   zsha2 start-node

• Expected result: The new JAR packages take effect and the platform resumes operation.

8. Check services, then re-enable platform HA.

• Expected result: The DRS GPU-loss issue is resolved.

Verification

• Both management services recover normally.
• GPUs are no longer lost after DRS migration.
• Services run normally.
• The platform has no exceptions after HA is restored.

Risks and rollback

• Risk: Replacing two core JAR packages together can affect management-platform behavior if versions do not match.
• Risk level: high.
• Rollback:

1. Disable HA and back up the database.
2. Restore from /root/mevoco-5.5.x.jar.bak and /root/drs-5.5.x.jar.bak.
3. Restart management-platform services and reconnect physical hosts.
4. Re-enable HA.