[KAFKA-2535] When removing a partition from an out of sync broker, it should not need to join the ISR first

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 0.8.2.0
• Component/s: replication
• Assignee: Neha Narkhede
• Reporter: Sean Fellows
• Created: 11/Sep/15 17:43
• Updated: 11/Sep/15 17:43

Description

I'm not 100% sure this improvement would be legitimate, but I will describe the issue I have been seeing and hopefully folks can advise.

To reproduce:
1. Start with a broker that is out of sync for a particular partition.
2. Issue a reassignment that moves that partition off of that broker.

Observed behavior:
The broker continues replicating until it joins the ISR.

Desired behavior:
It seems to me that this broker should stop replication immediately. If this broker is able to replicate from somewhere, it means the ISR is not empty, so it should be safe to delete this partition's data from this broker.

The reason I am interested in this is that I would like to be able to reduce disk usage on a broker by removing partitions from it. But if it has fallen behind on replication it continues to need more disk space until it has caught up. But it is just going to delete this data anyway, so it seems like this optimization should be possible.