如何配置JMeter生成非持久HTTP请求及端口耗尽时的行为咨询

Great question! Let’s break this down into two clear parts: how to configure JMeter for non-persistent connections with the Connection: close header, and what happens when you hit port exhaustion due to overwhelming TIME_WAIT states.

To force JMeter to use a brand-new TCP connection for every request and send the Connection: close header, follow these straightforward steps:

• Add an HTTP Header Manager: Drop this element into your test plan (or per thread group if you only want this behavior for specific requests). Add a header entry where the name is Connection and the value is close. This explicitly tells the server to terminate the TCP connection immediately after sending its response.
• Disable Keep-Alive in HTTP Requests: Open your HTTP Request element, navigate to the Advanced tab, and uncheck the Use keepalive box. This ensures JMeter doesn’t attempt to reuse connections for subsequent requests, perfectly aligning with the Connection: close header’s intent.
• Optional: Globally Disable Connection Reuse: If you want to enforce this behavior across all requests (instead of configuring each element individually), tweak JMeter’s HttpClient settings:
  • Add a JVM argument when starting JMeter: -Dhttpclient4.reuseconnections=false
  • Or set the same property in your user.properties or jmeter.properties file: httpclient4.reuseconnections=false

When you fire requests fast enough that most of your client’s ephemeral ports get stuck in the TIME_WAIT state (a normal TCP protocol behavior that keeps ports reserved for ~60 seconds by default to handle delayed packets), JMeter will hit hard limits with these noticeable effects:

• Connection Failures: You’ll start seeing java.net.BindException: Address already in use: connect errors in the JMeter logs and test results. This means the JVM can’t grab a new port to initiate a fresh TCP connection.
• Mass Request Failures: Throughput will plummet instantly, and a large percentage of requests will fail. Threads may hang or throw unhandled exceptions as they wait for a port that never becomes available.
• OS-Level Performance Hits: Even before full port exhaustion, your operating system will struggle to manage hundreds or thousands of TIME_WAIT sockets, leading to increased latency for all network operations on the client machine.
• Slow Recovery: Once port exhaustion occurs, JMeter can’t resume normal testing until enough TIME_WAIT ports expire (controlled by your OS’s tcp_fin_timeout setting, usually 60 seconds). You’ll either need to stop the test and wait for ports to free up, or adjust OS TCP settings to reduce the TIME_WAIT duration (note: this can have stability trade-offs for other network traffic on the machine).

内容的提问来源于stack exchange，提问作者flow2k