Hide Forgot
Description of problem: We have seen issues where an improper number of rgw bucket shards are configured leading to performance issues and issues during recovery. Current documentation only highlights possible performance issues when large numbers of objects are placed into buckets. But does not detail recommendations when deciding on number of shards to configure, or how to determine proper number of shards. Current doc detailing bucket sharding: https://access.redhat.com/documentation/en/red-hat-ceph-storage/1.3/single/object-gateway-guide-for-red-hat-enterprise-linux/#bucket_sharding Version-Release number of selected component (if applicable): 1.3.x
Hello Bara, So in 1.3 RGW used Federated Gateways for replicating data across multiple clusters in which regions are used and in 2.0 it has changed to Multi-site in which regions are converted to zonegroups. So for [1] in the 1.3 docs it should be labeled "federated configuration". In 2.0 it should be labeled "multi-site configuration". The steps would not be the same as the region.json which is updated with the bucket_index_max_shards setting in the configuration for that region would be need to be updated in the zonegroup.json in multi-site configurations. For [2] the "rgw_override_bucket_index_max_shards" is a global setting which enables bucket index sharding on all new buckets created after a value is given to this setting. The "bucket_index_max_shards" is set to achieve a consistent shard counts for zones in a region or zonegroup for failover. @Yehuda do you have any additional comments or concerns?
Thanks Bara, for providing backup links. Doc is updated with formula to calculate shards and max shards value. Moving this bug to verified state.