After upgrading from version 5.2.3 to 7.5.0, the disk write volume has significantly decreased, various latencies have greatly reduced, and the overall QPS has improved. The only downside is that the IOPS and disk utilization have increased (HDD).
It can be understood as one region per RocksDB. The advantage of this is that compaction between regions will not affect each other, and each will handle its own. Previously, mixing them together would cause mutual interference, leading to read and write amplification.
The Partitioned Raft KV feature has only been supported since v6.6.0 and is still an experimental feature in v7.5.0, making it inconvenient to enable in production environments. Additionally, it can only be set during the initial deployment of a new cluster.
I’ve heard that this feature has not been advanced due to prioritization issues and there is a possibility that it might be canceled in the future.
Boss, your IOPS increase should be a normal phenomenon since the business QPS has gone up.
From v5 to v7.5, did you make any adaptations or compatibility adjustments on the business side? Do you have any other experiences or pitfalls to share?
Yes, it’s normal. For less important databases, they are usually upgraded directly after data entry and a small number of queries, and there are no major issues.
There are likely other higher priority tasks being worked on, such as further enhancing stability and performance. Additionally, after a comprehensive evaluation, the risk-reward ratio is not as high as expected. The features of PRKV are not currently a priority for advancement, but it can still be used in the testing environment.