minio/docs/distributed/DECOMMISSION.md

Decommissioning

Decommissiong is a mechanism in MinIO to drain older pools (usually with old hardware) and migrate the content from such pools to a newer pools (usually better hardware). Decommissioning spreads the data across all pools - for example if you decommission pool1, all the data from pool1 shall be spread across pool2 and pool3 respectively.

Features

• A pool in decommission still allows READ access to all its contents, newer WRITEs will be automatically scheduled to only new pools.
• All versioned buckets maintain the same order for "versions" for each objects after being decommissioned to the newer pools.

How to decommission a pool?

λ mc admin decommission start alias/ http://minio{1...2}/data{1...4}

Status decommissioning a pool

Decommissioning without args lists all pools

λ mc admin decommission status alias/
┌─────┬─────────────────────────────────┬──────────────────────────────────┬────────┐
│ ID  │ Pools                           │ Capacity                         │ Status │
│ 1st │ http://minio{1...2}/data{1...4} │ 439 GiB (used) / 561 GiB (total) │ Active │
│ 2nd │ http://minio{3...4}/data{1...4} │ 329 GiB (used) / 421 GiB (total) │ Active │
└─────┴─────────────────────────────────┴──────────────────────────────────┴────────┘

Decommissioning status

λ mc admin decommission status alias/ http://minio{1...2}/data{1...4}
Progress: ===================> [1GiB/sec] [15%] [4TiB/50TiB]
Time Remaining: 4 hours (started 3 hours ago)

A pool not under decommissioning will throw an error

λ mc admin decommission status alias/ http://minio{1...2}/data{1...4}
ERROR: This pool is not scheduled for decommissioning currently.

Canceling a decommission?

Stop an on-going decommission in progress, mainly used in situations when the load may be too high and you may want to schedule the decommission at a later point in time.

mc admin decommission cancel without an argument, lists out any on-going decommission in progress.

λ mc admin decommission cancel alias/
┌─────┬─────────────────────────────────┬──────────────────────────────────┬──────────┐
│ ID  │ Pools                           │ Capacity                         │ Status   │
│ 1st │ http://minio{1...2}/data{1...4} │ 439 GiB (used) / 561 GiB (total) │ Draining │
└─────┴─────────────────────────────────┴──────────────────────────────────┴──────────┘

NOTE: Canceled decommission will not make the pool active again, since we might have potentially partial namespace on the other pools, to avoid this scenario be absolutely sure to make decommissioning a planned well thought activity. This is not to be run on a daily basis.

λ mc admin decommission cancel alias/ http://minio{1...2}/data{1...4}
┌─────┬─────────────────────────────────┬──────────────────────────────────┬────────────────────┐
│ ID  │ Pools                           │ Capacity                         │ Status             │
│ 1st │ http://minio{1...2}/data{1...4} │ 439 GiB (used) / 561 GiB (total) │ Draining(Canceled) │
└─────┴─────────────────────────────────┴──────────────────────────────────┴────────────────────┘

If for some reason decommission fails in between, the status will indicate decommission as failed instead.

λ mc admin decommission status alias/
┌─────┬─────────────────────────────────┬──────────────────────────────────┬──────────────────┐
│ ID  │ Pools                           │ Capacity                         │ Status           │
│ 1st │ http://minio{1...2}/data{1...4} │ 439 GiB (used) / 561 GiB (total) │ Draining(Failed) │
│ 2nd │ http://minio{3...4}/data{1...4} │ 329 GiB (used) / 421 GiB (total) │ Active           │
└─────┴─────────────────────────────────┴──────────────────────────────────┴──────────────────┘

Restart a canceled or failed decommission?

λ mc admin decommission start alias/ http://minio{1...2}/data{1...4}

When decommission is 'Complete'?

Once decommission is complete, it will be indicated with Complete status. Complete means that now you can now safely remove the first pool argument from the MinIO command line.

λ mc admin decommission status alias/
┌─────┬─────────────────────────────────┬──────────────────────────────────┬──────────┐
│ ID  │ Pools                           │ Capacity                         │ Status   │
│ 1st │ http://minio{1...2}/data{1...4} │ 439 GiB (used) / 561 GiB (total) │ Complete │
│ 2nd │ http://minio{3...4}/data{1...4} │ 329 GiB (used) / 421 GiB (total) │ Active   │
└─────┴─────────────────────────────────┴──────────────────────────────────┴──────────┘

• On baremetal setups if you have MINIO_VOLUMES="http://minio{1...2}/data{1...4} http://minio{3...4}/data{1...4}" you can remove the first argument http://minio{1...2}/data{1...4} to update your MINIO_VOLUMES setting and using systemctl restart minio on all the servers in the setup in parallel.

• On Kubernetes setups MinIO the statefulset specification needs to be modified by changing the command line input for the MinIO container. Once the relevant changes are done proceed to execute kubectl apply -f statefulset.yaml.

• On Operator based MinIO deployments you need to modify the tenant.yaml specification and modify the pools: section from two entries to a single entry, once relevant changes are done proceed to execute kubectl apply -f tenant.yaml.

Without a 'Complete' status any 'Active' or 'Draining' pool(s) are not allowed to be removed once configured.

NOTE

• Empty delete marker's i.e objects with no other successor versions are not transitioned to the new pool, to avoid any empty metadata being recreated on the newer pool. We do not think this is needed, please open a GitHub issue if you think otherwise.

TODO

• Richer progress UI is not present at the moment, this will be addressed in subsequent releases. Currently however a RATE of data transfer and usage increase is displayed via mc.
• Transitioned Hot Tier's as pooled setups are not currently supported, attempting to decommission buckets with ILM Transition will be rejected by the server. This will be supported in future releases.
• Embedded Console UI does not support Decommissioning through the UI yet. This will be supported in future releases.