Skip to content

[ISSUE #10441] Reduce per-RPC allocation in metrics by caching static AttributeKey instances#10443

Open
wang-jiahua wants to merge 1 commit into
apache:developfrom
wang-jiahua:perf/static-attributekey-metrics-caching
Open

[ISSUE #10441] Reduce per-RPC allocation in metrics by caching static AttributeKey instances#10443
wang-jiahua wants to merge 1 commit into
apache:developfrom
wang-jiahua:perf/static-attributekey-metrics-caching

Conversation

@wang-jiahua

@wang-jiahua wang-jiahua commented Jun 8, 2026

Copy link
Copy Markdown
Contributor

Summary

Reduce per-RPC allocation by replacing reflective AttributeKey.stringKey(name) / AttributeKey.booleanKey(name) calls with pre-built static AttributeKey<T> singletons.

Problem

JFR heap dump showed 38,182 InternalAttributeKeyImpl instances for only 6 distinct key names. Every AttributesBuilder.put(String, T) call internally creates a fresh AttributeKey — a stateless immutable object that should be a singleton.

Fix

Define static final AttributeKey<T> LABEL_*_KEY constants in BrokerMetricsConstant and RemotingMetricsConstant, then migrate all put(String, T) callsites to put(AttributeKey<T>, T).

Files

File Change
BrokerMetricsConstant.java +19: 14 typed AttributeKey<> constants
BrokerMetricsManager.java Migrate put calls to typed keys
PopMetricsManager.java Migrate put calls to typed keys
RemotingMetricsConstant.java +10: 5 typed AttributeKey<> constants
RemotingMetricsManager.java Migrate put calls + .equals() comparison

Split Note

Three misc optimizations previously bundled here (NopStatusListener, TopicMessageType.metricsValue cache, RemotingHelper eager eval fix) have been split into PR #10491.

Copilot AI review requested due to automatic review settings June 8, 2026 09:21

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR focuses on reducing allocation/CPU overhead on metrics and remoting hot paths by introducing AttributeKey singletons and adding small caches for frequently repeated attribute/metric lookups.

Changes:

  • Added fast-path caching for remoting request/response code distribution counting.
  • Introduced typed OpenTelemetry AttributeKey constants and attribute caching/build helpers in remoting and broker metrics.
  • Reduced repeated string allocations (e.g., cached lowercase metrics value, avoided Map.getOrDefault boxing paths).

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 6 comments.

Show a summary per file
File Description
remoting/src/main/java/org/apache/rocketmq/remoting/netty/RemotingCodeDistributionHandler.java Adds “last code” fast path to reduce map lookups for hot request/response codes.
remoting/src/main/java/org/apache/rocketmq/remoting/metrics/RemotingMetricsManager.java Adds attribute caching and typed label keys to reduce OpenTelemetry attribute overhead.
remoting/src/main/java/org/apache/rocketmq/remoting/metrics/RemotingMetricsConstant.java Introduces typed AttributeKey singletons for labels.
remoting/src/main/java/org/apache/rocketmq/remoting/common/RemotingHelper.java Minor optimization to avoid getOrDefault overhead for code descriptions.
common/src/main/java/org/apache/rocketmq/common/attribute/TopicMessageType.java Caches lowercase metrics value to avoid repeated conversions.
broker/src/main/resources/rmq.broker.logback.xml Suppresses Logback status output via NopStatusListener.
broker/src/main/java/org/apache/rocketmq/broker/metrics/PopMetricsManager.java Switches to typed label keys for attributes.
broker/src/main/java/org/apache/rocketmq/broker/metrics/BrokerMetricsManager.java Adds topic attributes cache and switches to typed label keys.
broker/src/main/java/org/apache/rocketmq/broker/metrics/BrokerMetricsConstant.java Introduces typed AttributeKey singletons for broker metrics labels.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread broker/src/main/java/org/apache/rocketmq/broker/metrics/BrokerMetricsManager.java Outdated
Comment thread broker/src/main/resources/rmq.broker.logback.xml
@wang-jiahua wang-jiahua force-pushed the perf/static-attributekey-metrics-caching branch 2 times, most recently from 607c58a to 52c49e8 Compare June 8, 2026 10:02
@codecov-commenter

codecov-commenter commented Jun 8, 2026

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 31.37255% with 35 lines in your changes missing coverage. Please review.
✅ Project coverage is 48.04%. Comparing base (82a6a78) to head (f8e69a4).
⚠️ Report is 19 commits behind head on develop.

Files with missing lines Patch % Lines
.../rocketmq/broker/metrics/BrokerMetricsManager.java 0.00% 25 Missing ⚠️
...etmq/remoting/metrics/RemotingMetricsConstant.java 0.00% 5 Missing ⚠️
...che/rocketmq/broker/metrics/PopMetricsManager.java 33.33% 4 Missing ⚠️
...ketmq/remoting/metrics/RemotingMetricsManager.java 0.00% 1 Missing ⚠️
Additional details and impacted files
@@              Coverage Diff              @@
##             develop   #10443      +/-   ##
=============================================
- Coverage      48.06%   48.04%   -0.03%     
- Complexity     13313    13321       +8     
=============================================
  Files           1377     1377              
  Lines         100632   100726      +94     
  Branches       12995    13010      +15     
=============================================
+ Hits           48369    48392      +23     
- Misses         46344    46383      +39     
- Partials        5919     5951      +32     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@oss-sentinel-ai oss-sentinel-ai left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review by github-manager-bot

Summary

This PR caches OpenTelemetry AttributeKey instances as static finals to avoid per-RPC allocation, optimizes TopicMessageType lookup, adds a volatile inline cache for RemotingCodeDistributionHandler, and adds Logback NopStatusListener to suppress startup log noise.

Findings

  • [Good] BrokerMetricsConstant.java:69-86 — Pre-built typed AttributeKey singletons eliminate per-call allocation on the metrics hot path.
  • [Good] BrokerMetricsManager.java:105-120 — Import changes correctly reference the new cached keys.
  • [Good] RemotingCodeDistributionHandler.java:35-68 — Volatile inline cache pattern correctly avoids ConcurrentHashMap.computeIfAbsent() on repeated request codes.
  • [Good] TopicMessageType.java — Enum-based lookup is more efficient than string comparison.
  • [Good] rmq.broker.logback.xml — NopStatusListener suppresses Logback startup noise in production.

Suggestions

  1. The volatile inline cache pattern in RemotingCodeDistributionHandler is effective for sequential workloads. Consider documenting that this optimization assumes temporal locality in request codes.
  2. Consider adding a microbenchmark to quantify the allocation reduction.

Verdict

Approved — Solid performance optimization with clean implementation.


Automated review by github-manager-bot

@wang-jiahua wang-jiahua force-pushed the perf/static-attributekey-metrics-caching branch from 8eeec77 to f8e69a4 Compare June 12, 2026 08:45
wang-jiahua pushed a commit to wang-jiahua/rocketmq that referenced this pull request Jun 12, 2026
… metrics hot path

Builds on top of PR apache#10443 (which introduces the typed AttributeKey constants).

After AttributeKey instances are de-duplicated, the next allocation hot
spot in the metrics path is the Attributes object itself. For repeated
parameter combinations the broker keeps rebuilding identical immutable
Attributes (each ~200-300 byte). This commit introduces a two-level
cache (volatile inline cache + ConcurrentHashMap fallback) at three
points:

- BrokerMetricsManager.getOrBuildTopicAttributes(topic, msgType, isSystem):
  4-field volatile inline cache + CHM keyed by 'topic|msgType|isSystem'.

- RemotingMetricsManager.getOrBuildAttributes(reqCode, respCode, isLongPolling, result):
  long-packed cache key (35 bits across 4 args, no String allocation),
  volatile inline cache + CHM. Falls back to direct build for unknown
  result values.

- RemotingCodeDistributionHandler: replace two independent volatile fields
  (lastCode, lastAdder) with an immutable CachedCounter holder published
  through a single volatile reference. A single volatile read per inc()
  returns a self-consistent (code, adder) snapshot, eliminating a
  torn-read between the two fields under @sharable concurrent access
  from multiple Netty EventLoop threads.

@RockteMQ-AI RockteMQ-AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review by github-manager-bot

Summary

Pre-builds AttributeKey singletons as static constants in BrokerMetricsConstant and updates all call sites to use them, eliminating per-RPC allocation of InternalAttributeKeyImpl objects.

Findings

  • [Critical] BrokerMetricsConstant.java:69-83 — All 15 AttributeKey singletons are correctly typed (stringKey for String labels, booleanKey for boolean labels). Each call to AttributeKey.stringKey("name") creates a new object; caching eliminates this per-RPC allocation.
  • [Critical] BrokerMetricsManager.java, PopMetricsManager.java, RemotingMetricsManager.java — All call sites consistently updated to use the cached keys. No remaining AttributeKey.stringKey() calls on hot paths.
  • [Info] RemotingMetricsManager.java:125-135 — The newAttributes() helper now uses cached keys. This is called on every RPC completion, so the savings are proportional to RPC throughput.

Suggestions

No concerns. This is a straightforward, low-risk optimization with clear benefits on high-throughput brokers.

Verdict: LGTM. Clean optimization with consistent application across all metrics call sites.


Automated review by github-manager-bot

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants