Primary Shards MUCH larger than Replicas

For larger indices, I'm seeing a pattern where my primary shards are significantly larger than replicas. For example, using the _cat/shards API, I'm getting:

524183765326406aba73345d470e727c-2018-07-30                    0  p STARTED  9917382   19.9gb prod-es-data-hot-03e9547ee64f9d19c
524183765326406aba73345d470e727c-2018-07-30                    0  r STARTED  9917269    6.6gb prod-es-data-hot-022b8be4866a004c4
524183765326406aba73345d470e727c-2018-07-30                    1  p STARTED  9906022   20.5gb prod-es-data-hot-07491395f0bb84af5
524183765326406aba73345d470e727c-2018-07-30                    1  r STARTED  9905929    6.5gb prod-es-data-hot-08454b4b222c130ef
524183765326406aba73345d470e727c-2018-07-30                    2  p STARTED  9911138   18.5gb prod-es-data-hot-0226703b13afd4fe1
524183765326406aba73345d470e727c-2018-07-30                    2  r STARTED  9910932    6.5gb  prod-es-data-hot-0c78098bc764541e3
524183765326406aba73345d470e727c-2018-07-30                    3  p STARTED  9908133   20.2gb  prod-es-data-hot-07f6b364df3787667
524183765326406aba73345d470e727c-2018-07-30                    3  r STARTED  9907897    6.6gb prod-es-data-hot-0e9476b7710e99538
524183765326406aba73345d470e727c-2018-07-30                    4  p STARTED  9907219   19.5gb prod-es-data-hot-08fb8c2ded8c64e3c
524183765326406aba73345d470e727c-2018-07-30                    4  r STARTED  9907181    6.6gb prod-es-data-hot-0d8aff97c8ff29904
524183765326406aba73345d470e727c-2018-07-30                    5  p STARTED  9905674   20.1gb  prod-es-data-hot-0812c487071e9446f
524183765326406aba73345d470e727c-2018-07-30                    5  r STARTED  9905482    6.6gb prod-es-data-hot-055876f71221351ec

Given prior usage, I expect each shard to be roughly 6-8 GB, which my replicas are. However, my primaries are three times this size! Is this expected? Is there anything I can do, short of setting the index to RO and doing a force-merge? Having wildly differently sized shards is complicating my efforts to balance disk and CPU usage across my cluster.

I'm currently using ES 6.2.4.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.