[leo_storage] Possibility to become imbalance of a total mq's msgs during a rebalance (detach-node)

[yosukehara]

Description

In the current version, v1.2.21, after executing detach and rebalance command, LeoStorage assigns consumption messages of a rebalance into a primary of a vnode(RING). According to that, there is a possibility to become imbalance of a total messages between storage-nodes.

Solution

LeoStorage adjusts a total number of assigned storage-nodes so that a storage-node is able to have same consumption messages with other nodes into a LeoFS' cluster.

[windkit]

[yosukehara]

This issue seems to be improved with a yesterday update. We're going to test the detach-feature for more several times.

[Result]

• 1M objects rebalance (detach)
• Duration: approx 50 min

[yosukehara]

To be honest, I've found that there is one issue which is imbalance remote access during the data-rebalance like as below:

In the next action, I need to fix imbalance remote access of the data-rebalance.

[yosukehara]

We've checked the data-rebalance performance after executing a detach node from 8 nodes to 7 nodes as below:

[Result]

• 10M objects rebalance (detach)
• Duration: approx 11 hours

As I mentioned at the former comment, I'm going to improve the remote access processing as the next action.

[yosukehara]

We've checked the data-rebalance performance again, which is the same situation of the previous test. The data-rebalance performance has been dramatically increased, the duration time from 11 hours to 8.5 hours.

[yosukehara]

Today, we've recognized this issue was fixed.