Cray System Management Documentation > Upgrade CSM > CSM 1.3 to 1.4 Upgrade Process

CSM 1.3 to 1.4 Upgrade Process

Introduction

This document guides an administrator through the upgrade of Cray Systems Management from version 1.3 to version 1.4. This procedure works for all patch versions of the source and target CSM releases.

This procedure is also the correct one to follow in the unusual situation of upgrading from a pre-release version of CSM 1.4.0 to a newer version of CSM 1.4.

When upgrading a system, follow this top-level file from top to bottom. The content on this top-level page is meant to be terse. For additional reference material on the upgrade processes and scripts mentioned explicitly on this page, see resource material.

Important notes

Service request adjustments are needed for small systems

For systems with only three worker nodes (typically Testing and Development Systems (TDS)), prior to proceeding with this upgrade, CPU limits MUST be lowered on several services in order for this upgrade to succeed. This step is executed automatically as part of Stage 0.3 of the upgrade. See TDS Lower CPU Requests for more information.
Independently, for three-worker systems the customizations.yaml file is edited automatically during the upgrade, prior to deploying new CSM services. These settings are contained in /usr/share/doc/csm/upgrade/scripts/upgrade/tds_cpu_requests.yaml. This file can be modified (prior to proceeding with this upgrade), if other settings are desired in the customizations.yaml file for this system.

For more information about modifying customizations.yaml and tuning for specific systems, see Post-Install Customizations.

Upgrade stages

NOTE: Storage nodes are not upgraded as part of the CSM 1.3 to CSM 1.4 upgrade.

Important: Take note of the below content for troubleshooting purposes, in the event that issues are encountered during the upgrade process.

General upgrade troubleshooting

If the execution of the upgrade procedure fails, it is safe to rerun the failed script. If a rerun still fails, wait for 10 seconds and then run it again. If the issue persists, then refer to the below troubleshooting procedures.
General Kubernetes troubleshooting

For general Kubernetes commands for troubleshooting, see Kubernetes Troubleshooting Information.
PXE boot troubleshooting

If execution of the upgrade procedures results in NCNs that have errors booting, then refer to the troubleshooting procedures in the PXE Booting Runbook.
NTP troubleshooting

During upgrades, clock skew may occur when rebooting nodes. If one node is rebooted and its clock differs significantly from those that have not been rebooted, it can cause contention among the other nodes. Waiting for chronyd to slowly adjust the clocks can resolve intermittent clock skew issues. This can take up to 15 minutes or longer. If it does not resolve on its own, then follow the Configure NTP on NCNs procedure to troubleshoot it further.
Bare-metal Etcd recovery

During the upgrade process of the master nodes, if it is found that the bare-metal Etcd cluster (that houses values for the Kubernetes cluster) has a failure, it may be necessary to restore that cluster from backup. See Restore Bare-Metal etcd Clusters from an S3 Snapshot for that procedure.
Bare-metal Etcd certificate

After upgrading, apiserver-etcd-client certificate may need to been renewed. See Kubernetes and Bare Metal EtcD Certificate Renewal for procedures to check and renew this certificate.
Back-ups for etcd-operator Clusters

After upgrading, if health checks indicate that Etcd pods are not in a healthy/running state, recovery procedures may be needed. See Backups for etcd-operator Clusters for these procedures.
Recovering from Postgres database issues

After upgrading, if health checks indicate the Postgres pods are not in a healthy/running state, recovery procedures may be needed. See Troubleshoot Postgres Database for troubleshooting and recovery procedures.
Back-ups for Postgres databases

After upgrading, if any *postgresql-db-backup cronjob pods are in error, see NCN Resource Checks. If the most recent *postgresql-db-backup cronjob pod is in error and the pod log indicates a failure due to pg_dumpall: error: pg_dump failed on database ..., contact support to further investigate and resolve.
Troubleshooting Spire pods not starting on NCNs

See Troubleshoot Spire Failing to Start on NCNs.
Troubleshoot SLS not working

See SLS Not Working During Node Rebuild.
Troubleshoot Check for duplicate and DNS entries for NCN and UANs test failure

See Check for duplicate and DNS entries for NCN and UANs test failure

Rerun a step

When running master node and storage node upgrade scripts, each script records what has been done successfully on a node. This is recorded in the /etc/cray/upgrade/csm/{CSM_VERSION}/{NAME_OF_NODE}/state file. If a rerun is required, the recorded steps to be re-run must be removed from this file.

(ncn#) Here is an example of state file of ncn-m001:

cat /etc/cray/upgrade/csm/csm-{CSM_VERSION}/ncn-m001/state

Example output:

[2021-07-22 20:05:27] UNTAR_CSM_TARBALL_FILE
[2021-07-22 20:05:30] INSTALL_CSI
[2021-07-22 20:05:30] INSTALL_WAR_DOC
[2021-07-22 20:13:15] SETUP_NEXUS
[2021-07-22 20:13:16] UPGRADE_BSS <=== Remove this line if you want to rerun this step
[2021-07-22 20:16:30] CHECK_CLOUD_INIT_PREREQ
[2021-07-22 20:19:17] APPLY_POD_PRIORITY
[2021-07-22 20:19:38] UPDATE_BSS_CLOUD_INIT_RECORDS
[2021-07-22 20:19:38] UPDATE_CRAY_DHCP_KEA_TRAFFIC_POLICY
[2021-07-22 20:21:03] UPLOAD_NEW_NCN_IMAGE
[2021-07-22 20:21:03] EXPORT_GLOBAL_ENV
[2021-07-22 20:50:36] PREFLIGHT_CHECK
[2021-07-22 20:50:38] UNINSTALL_CONMAN
[2021-07-22 20:58:39] INSTALL_NEW_CONSOLE

See the inline comment above on how to rerun a single step.
In order to rerun the whole upgrade of a node, delete its state file.

Skip a step after running it manually

When running master node and storage node upgrade scripts, each script records what has been done successfully on a node. This is recorded in the /etc/cray/upgrade/csm/{CSM_VERSION}/{NAME_OF_NODE}/state file. If a step fails in the upgrade script and then is successfully run manually, this step needs to be added to the state file so it will be skipped by the upgrade procedure.