Using the pg-upmap

Starting in Luminous v12.2.z there is a new pg-upmap exception table in the OSDMap that allows the cluster to explicitly map specific PGs to specific OSDs. This allows the cluster to fine-tune the data distribution to, in most cases, perfectly distributed PGs across OSDs.

The key caveat to this new mechanism is that it requires that all clients understand the new pg-upmap structure in the OSDMap.

Enabling

New clusters will have this module on by default. The cluster must only have luminous (and newer) clients. You can the turn the balancer off with:

ceph balancer off

To allow use of the feature on existing clusters, you must tell the cluster that it only needs to support luminous (and newer) clients with:

ceph osd set-require-min-compat-client luminous

This command will fail if any pre-luminous clients or daemons are connected to the monitors. You can see what client versions are in use with:

ceph features

Balancer module

The balancer module for ceph-mgr will automatically balance the number of PGs per OSD. See Balancer

Offline optimization

Upmap entries are updated with an offline optimizer built into osdmaptool.

  1. Grab the latest copy of your osdmap:

    ceph osd getmap -o om
    
  2. Run the optimizer:

    osdmaptool om --upmap out.txt [--upmap-pool <pool>]
             [--upmap-max <max-optimizations>] [--upmap-deviation <max-deviation>]
             [--upmap-active]
    

    It is highly recommended that optimization be done for each pool individually, or for sets of similarly-utilized pools. You can specify the --upmap-pool option multiple times. "Similar pools" means pools that are mapped to the same devices and store the same kind of data (e.g., RBD image pools, yes; RGW index pool and RGW data pool, no).

    The max-optimizations value is the maximum number of upmap entries to identify in the run. The default is 10 like the ceph-mgr balancer module, but you should use a larger number if you are doing offline optimization. If it cannot find any additional changes to make it will stop early (i.e., when the pool distribution is perfect).

    The max-deviation value defaults to 5. If an OSD PG count varies from the computed target number by less than or equal to this amount it will be considered perfect.

    The --upmap-active option simulates the behavior of the active balancer in upmap mode. It keeps cycling until the OSDs are balanced and reports how many rounds and how long each round is taking. The elapsed time for rounds indicates the CPU load ceph-mgr will be consuming when it tries to compute the next optimization plan.

  3. Apply the changes:

    source out.txt
    

    The proposed changes are written to the output file out.txt in the example above. These are normal ceph CLI commands that can be run to apply the changes to the cluster.

The above steps can be repeated as many times as necessary to achieve a perfect distribution of PGs for each set of pools.

You can see some (gory) details about what the tool is doing by passing --debug-osd 10 and even more with --debug-crush 10 to osdmaptool.