[HDDS-6520] Refactor RDBTable#getRangeKVs()

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Not A Problem
• Labels: pull-request-available
• Assignee: Kaijie Chen
• Reporter: Kaijie Chen
• Created: 28/Mar/22 04:06
• Updated: 20/Oct/22 03:44
• Resolved: 20/Oct/22 03:44

Description

Refactor RDBTable#getRangeKVs(), cleanup code and make getting prevKey more efficient.

Question

Why RDBTable#getRangeKVs() should return empty list if startKey is not found?
It makes getting all key-value pairs awkward.

Current

For example, if the keys in table are [10, 20, 30, 40, 50, 60, 70, 80], and we use count = 3 in each iteration.

1. startKey = None, result = [10, 20, 30]
2. startKey = 30, result = [30, 40, 50] (can't use 31 as strartKey or we will get empty result)
3. startKey = 50, result = [50, 60, 70] 
4. startKey = 70, result = [70, 80]

And the total results is [10, 20, 30, 30, 40, 50, 50, 60, 70, 70, 80], note the startKey in each iteration is duplicated.

Approach 1

If we can seek for the next key if  startKey does not exist, we can use 31 in the 2nd iteration and get [40, 50, 60].

1. startKey = None, result = [10, 20, 30]
2. startKey = 31, result = [40, 50, 60]
3. startKey = 61, result = [70, 80]

Verified CI can pass if we remove the startKey exist check.
https://github.com/apache/ozone/pull/3339#issuecomment-1118150272

Approach 2

Or if we can change startKey to prevKey, the iterations will look more straightforward.

1. prevKey = None, result = [10, 20, 30]
2. prevKey = 30, result = [40, 50, 60]
3. prevKey = 60, result = [70, 80]

However, this approach may require additional changes in the the codebase.

Issue Links

• is fixed by: HDDS-6428 Add prefix iterator support to RDBTable.
• links to: GitHub Pull Request #3339