HDFS-17935. [ARR] Separate router async handlers for active and observer#8548
HDFS-17935. [ARR] Separate router async handlers for active and observer#8548kokonguyen191 wants to merge 4 commits into
Conversation
|
💔 -1 overall
This message was automatically generated. |
|
Unrelated javadocs failure, though it has been popping up for a few pull requests now. Maybe it's time to fix it |
|
💔 -1 overall
This message was automatically generated. |
|
Hi @ZanderXu @KeeProMise @hfutatzhanghb, please take a look when you're free, thanks! |
ZanderXu
left a comment
There was a problem hiding this comment.
Thanks @kokonguyen191 for your contribution.
This is a good idea. It can decouple the impact between the Active NameNode and Observer NameNodes on the RBF side, especially in scenarios where there are many Observer NameNodes.
However, it would be better to add a configuration to control whether this feature is enabled, because in most cases it is acceptable to use the same executor without differentiating between them.
|
💔 -1 overall
This message was automatically generated. |
Description of PR
Most router traffic are reads, and currently all traffic to a namespace is handled by one same executor, so high observer load can increase competition for active ops. It would be beneficial to split the handler pools into 2 separate sets, one for active, one for observer.
Added a simple unit test (LLM-gen for skeleton then rewrote) to count how many calls are proxied through which executor. Contains content generated by Codex/ChatGPT-5.5.
How was this patch tested?
UT. Have been running on production environment for months.
AI Tooling
If an AI tool was used:
where is the name of the AI tool used.
https://www.apache.org/legal/generative-tooling.html