Increased CPU utilization after upgrading from Node.js v20.20.0 to v22.x

Version

v22.22.0 / v22.22.2 / 24.15.0

Platform

Oracle Linux 8 latest

Subsystem

No response

What steps will reproduce the bug?

1. Run workload on Node.js v20.20.0 → CPU ~68–70%
2. Upgrade to Node.js v22.22.x or 24.15.0
3. Run identical workload (same TPS, config)
4. Observe CPU increase (~86–88%)
5. Revert to v20 → CPU returns to baseline

How often does it reproduce? Is there a required condition?

consistently 100% reproducable

What is the expected behavior? Why is that the expected behavior?

no bigger than 5% CPU degradation

What do you see instead?

20% CPU degradation

Additional information

We are observing a significant increase in CPU utilization after upgrading our runtime from Node.js v20.20.0 → v22.x (22.22.x), with no application-level changes.

A/B testing strongly indicates the regression is tied to the Node.js runtime upgrade rather than application logic.

Impact

• ~20% relative CPU increase observed
• Consistent across multiple test runs
• No functional regression, but clear performance degradation

---

Environment

• Node.js versions:
  • v20.20.0 (baseline)
  • v22.22.0 / v22.22.2
• Workload:
  • High-throughput service (~17–18K TPS)
  • Distributed/multi-node setup
• Runtime flags:
  • --max-semi-space-size=256
  • --max-old-space-size=3072

---

Observed Behavior

Scenario	Node Version	CPU Utilization
Baseline	v20.20.0	~68–70%
Upgraded runtime	v22.22.x	~86–88%
v22.x + --max-semi-space-size=16	v22.22.x	~88%
Reverted runtime	v20.20.0	~68%

---

Key Findings

1. Strong correlation with Node.js version

   • Reverting v22 → v20 reduces CPU by ~12–20%
   • No application changes between runs

2. Memory tuning did not help

   • Adjusting --max-semi-space-size showed no improvement
   • Lower values increased GC frequency without reducing CPU

3. Latest v22 versions do not fix the issue

   • Testing with v22.22.2 shows same behavior

4. Maglev likely not the cause

   • Disabled by default in Node ≥22.9.0
   • Issue persists on versions where Maglev is disabled

---

What Was Ruled Out

• Application-level regression
• Load/TPS variation
• GC tuning via semi-space adjustments
• Early v22-only issues

---

Suspected Cause

Likely related to runtime-level changes introduced in Node.js v22, such as:

• V8 upgrade (v11.x → v12.x)
• JIT execution changes
• GC or memory management differences
• libuv / stream internal changes

---

Reproduction Steps (High-level)

1. Run workload on Node.js v20.20.0 → CPU ~68–70%
2. Upgrade to Node.js v22.22.x
3. Run identical workload (same TPS, config)
4. Observe CPU increase (~86–88%)
5. Revert to v20 → CPU returns to baseline

---

Additional Notes

• No explicit Node.js release notes mention this regression
• Issue is reproducible under high throughput
• Not observed in lower TPS environments (≤5–7K TPS)

##Questions##

• Are there known CPU regressions or behavior changes in Node.js v22 / V8 12.x?
• Are there recommended runtime flags or GC tuning options for high-throughput workloads in v22?
• Any guidance on mitigating CPU increase when upgrading from v20 → v22?
• Will an upgrade to the latest available version of Node.js v26 fix this issue?

首先，在 Oracle Linux 8 上使用列出的内存标志，在 Node.js v20.20.0 和 v22.22.x 中复现已报告的高吞吐量工作负载，同时保持 TPS 和配置一致。比较 CPU 利用率，并记录一个最小且可重复的工作负载，或隔离出负责该运行时变更的原因；issue 中未指定源文件或测试。