Uploaded image for project: 'Core Server'
  1. Core Server
  2. SERVER-90955

Equal Distribution at Zone-level in Data Size Balancing

    • Type: Icon: Improvement Improvement
    • Resolution: Won't Do
    • Priority: Icon: Major - P3 Major - P3
    • None
    • Affects Version/s: 6.0.15, 7.0.9, 8.1.0-rc0, 7.3.2, 8.0.0-rc6
    • Component/s: Sharding
    • None
    • Cluster Scalability

      In MongoDB 6.0 the balancing algorithm changed to balance by data size at a collection level per shard. Meaning that a 1TB collection in a three shard cluster would be equally distributed when each shard had 333GB of the collection's data stored on it. 

      This change in the balancing algorithm had an unintended consequence of changing how data with zones were managed.  

      In 6.0 the balancer works towards a solution of having an equal data distribution per shard per collection leading to a distribution that is unequal per shard per zone if a shard is tagged to multiple zone ranges. 

       

      In 5.0 when the balancer was determining the equal distribution of data by the number of chunks the balancer would keep each zone range equally balanced across multiple shards on a per zone level.

       

      We should investigate the feasibility of implementing the 5.0 behavior of keeping an equal distribution of data across zone ranges that span multiple shards 

        1. Screenshot 2024-05-22 at 1.12.26 PM.png
          155 kB
          Matt Panton
        2. Screenshot 2024-05-22 at 1.12.33 PM.png
          144 kB
          Matt Panton

            Assignee:
            Unassigned Unassigned
            Reporter:
            matt.panton@mongodb.com Matt Panton
            Votes:
            0 Vote for this issue
            Watchers:
            10 Start watching this issue

              Created:
              Updated:
              Resolved: