zhangtongr created ZOOKEEPER-4914:
-------------------------------------
Summary: [QUESTION] Strategies for monitoring and preventing
"unreasonable length" errors in session closure
Key: ZOOKEEPER-4914
URL: https://issues.apache.org/jira/browse/ZOOKEEPER-4914
Project: ZooKeeper
Issue Type: Bug
Components: server
Affects Versions: 3.6.3
Reporter: zhangtongr
Description:
===========
I'm encountering potential risks with "unreasonable length" errors during
session closure, particularly when sessions have numerous ephemeral nodes. I'd
like to discuss possible monitoring and prevention strategies.
Current Situation:
-----------------
1. When sessions with many ephemeral nodes are closed, all node paths are
collected into a single transaction
2. If the combined size exceeds jute.maxbuffer, it results in "unreasonable
length" errors
3. Currently lacking effective ways to predict or prevent this issue
Questions:
---------
1. Monitoring Strategy:
* What metrics should we monitor to predict potential issues?
* Are there existing metrics for tracking session transaction sizes?
* How can we monitor the growth of ephemeral nodes per session?
2. Prevention Approaches:
* What are the recommended approaches to prevent this issue?
* Is there a way to estimate transaction size before session closure?
* Are there best practices for managing large numbers of ephemeral nodes?
3. Configuration Guidelines:
* What's the recommended jute.maxbuffer setting for different scenarios?
* Are there other relevant configuration parameters?
---------------
Would appreciate insights on:
1. Additional metrics to monitor
2. Early warning indicators
3. Prevention strategies
4. Best practices for large-scale deployments
Thank you for any guidance or suggestions.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
