Upgrade Guide - Scylla Manager 1.0.x to 1.1.x¶

Backup the data¶

Before any major procedure, like an upgrade, it is recommended to backup all the data to an external device. It is recommended to backup the scylla_manager keyspace from Scylla Manager’s backend, following this backup procedure.

Download and install the new release¶

Before upgrading, check what version you are running now using rpm -qa | grep scylla-manager. You should use the same version in case you want to rollback the upgrade.

To upgrade:

1. Update the Update the Scylla Manager repo: to 1.1.x

2. Run:

sudo yum update scylla-manager -y

Validate¶

1. Check Scylla Manager status with systemctl status scylla-manager.service. Confirm the service is active (running).

2. Confirm that the upgrade changed the Client and Server version. Run sctool version and make sure both are 1.1.x version

3. Confirm that following the update, that your managed clusters are still present. Run sctool cluster list

4. Confirm that following the upgrade, there are no repairs in a stopped state. Run sctool task list to list the repair tasks in progress. If any are in a stopped state, run sctool repair unit schedule <repair-unit-id> --start-date=now to resume.

Update scylla-manager.yaml (optional)¶

As part of the upgrade procudre, paramaters were added to /etc/scylla-manager/scylla-manager.yaml. See the new values below. There is no need to update the scylla-manager.yaml file as part of the upgrade.

# Repair service configuration.
repair:
  # Granularity of repair. Repair works on segments, segment is a continuous
  # token range.
  #
  # Set the maximal number of tokens in a segment (zero is no limit).
  segment_size_limit: 0

  # Set number of segments to be repaired in one Scylla command.
  segments_per_repair: 1

  # Error tolerance.
  #
  # Set how many segments may fail to repair. Note that the manager would retry
  # to repair the failed segments. If the limit is exceeded, however, repair
  # will stop and the next repair will start from the beginning.
  segment_error_limit: 100

  # Fail-fast, set to true if you want repair to stop on first error. Unlike
  # segment_error_limit this allows for resuming the stopped repair.
  stop_on_error: false

  # Set wait time if Scylla failed to execute a repair command. Note that if
  # stop_on_error is true this has no effect.
  error_backoff: 10s

  # Set how often to poll Scylla node for command status.
  poll_interval: 200ms

  # Set time offset between the automated scheduler run and the scheduled
  # repairs. If scheduler runs at midnight the repairs would start at
  # midnight + this value. This gives you the opportunity to audit and modify
  # the scheduled repairs.
  auto_schedule_delay: 2h

  # Set maximal time after which a restarted repair is forced to start from the
  # beginning.
  max_run_age: 36h

  # Distribution of data among cores (shards) within a node.
  # Copy value from Scylla configuration file.
  murmur3_partitioner_ignore_msb_bits: 12

Downgrade to previous release¶

1. Stop Scylla Manager

sudo systemctl stop scylla-manager

2. Drop scylla_manager keyspace from the remote datastore

cqlsh -e "DROP KEYSPACE scylla_manager"

3. Remove Scylla Manager repo

sudo rm -rf /etc/yum.repos.d/scylla-manager.repo
sudo yum clean all

4. Update the Scylla Manager repo to 1.0.x

5. Install previous version

sudo yum downgrade scylla-manager scylla-manager-server scylla-manager-client -y

Start Scylla Manager¶

sudo systemctl start scylla-manager

Validate Scylla Manager Version¶

Validate Scylla Manager version:

sctool version

The version should match with the results you had previously.