DongNyoung Lee created ZOOKEEPER-5072:
-----------------------------------------
Summary: Add opt-in warning for slow ZooKeeper Java client
callbacks
Key: ZOOKEEPER-5072
URL: https://issues.apache.org/jira/browse/ZOOKEEPER-5072
Project: ZooKeeper
Issue Type: Improvement
Components: java client
Affects Versions: 3.10.0
Environment: ZooKeeper Java client
Reporter: DongNyoung Lee
h2. Background
ZooKeeper Java client callbacks are dispatched by a single {{{}EventThread{}}}.
A slow{{{} Watcher{}}}, {{{}AsyncCallback{}}}, or {{LocalCallback}} can
therefore delay subsequent watcher events, asynchronous callbacks, and
connection or session state events.
Operators currently have limited visibility into whether client-side callback
execution is contributing to observed delays.
h2. Proposal
Add an opt-in, threshold-based {{WARN}} log for slow ZooKeeper client callback
execution.
Proposed configuration:
{{zookeeper.client.slowCallbackThresholdMs}}
* Default value: {{0}} — disabled
* Values less than or equal to {{0}} disable callback duration measurement and
warning logs.
* When enabled, the client logs a {{WARN}} when callback execution time
exceeds the configured threshold.
* Callback duration should be measured using a monotonic clock.
The diagnostic logging must preserve the existing {{EventThread}} dispatch
model. It must not change callback ordering, callback threading, queue
structures, or exception-handling behavior.
h2. Non-goals
* Changing the callback executor
* Changing callback dispatch behavior
* Changing callback ordering
* Measuring queued dispatch delay
* Modifying queue structures
* Adding log rate limiting in the initial implementation
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
