Go memory diagnostics

1. Log on to SysOM. In the top-left corner of the page, select the region where the target instance is located.

2. In the left-side navigation pane, click system diagnostics.

3. Set Diagnosis Mode to node diagnostics. In the Diagnostic Type list, select memory diagnostics. In the list of diagnostic items, select Go memory diagnostics.

4. Configure the diagnostic parameters:

   Parameter	Description	Required
   Instance ID	The ID of the target ECS instance.	Yes
   PID	The PID of the target Go process.	Yes

5. After configuring the parameters, click Run Diagnosis. The system automatically runs the diagnosis and generates a report.

6. In the Diagnostic Records section, click View Report in the Actions column to open the diagnostic report.

Basic information

The Basic information section displays basic information about the diagnosis, including:

Conclusion

This section summarizes the diagnostic findings and identifies the likely causes of memory usage issues. Possible conclusions include:

• High Go heap memory usage: The inUseSpan proportion of the Go process is abnormally high, which may indicate a memory leak or excessive object allocation.

• Delayed Go memory release or fragmentation: Go memory is not released promptly, resulting in an abnormal proportion of deadSpan or manualSpan.

• High non-Go heap usage: The proportion of non-anonymous page memory (nonAnonPageMem) is excessively high, indicating a large number of file mappings or CGO calls.

• System status is normal, no intervention required: The memory usage of the Go process is normal with no significant anomalies.

Suggestions

This section provides suggestions based on the diagnostic results. The following are suggestions for common scenarios:

High Go heap memory usage

• Check for unreleased large objects or goroutine leaks.

• Use pprof for in-depth analysis to locate memory allocation hotspots.

• Use sync.Pool to reuse objects.

• Check for unnecessary closure captures that prevent memory from being released.

Delayed Go memory release or fragmentation

• Adjust the GODEBUG parameter, for example, by setting madvdontneed=1 to return memory to the OS more quickly.

• Review code that frequently allocates and frees small objects.

• Use sync.Pool to reduce memory fragmentation.

• Periodically trigger runtime.GC() to accelerate garbage collection.

High non-Go heap usage

• Check for unreleased memory from large mmap file mappings.

• Analyze memory allocation from CGO calls.

• Check for unclosed file descriptors.

• Optimize file reading methods to avoid loading entire files into memory.

Details

The Details section provides a detailed breakdown of memory usage, covering the following areas:

How do I get the PID of a Go process?

You can get the PID of a Go process using the following commands:

# Method 1: Find by process name
ps aux | grep <go_program_name>

# Method 2: Find by port (if the Go program listens on a specific port)
lsof -i :<port_number>

# Method 3: If the program prints its own PID
# Add the following code at program startup: fmt.Printf("PID: %d\n", os.Getpid())

What is the difference between Go memory diagnostics and OOM diagnostics?

• OOM diagnostics: Analyzes Out of Memory (OOM) events to determine their causes, such as insufficient system memory or exceeding cgroup memory limits.

• Go memory diagnostics: Analyzes memory usage in a running Go process. You can run it on a specified Go process at any time without waiting for an OOM event.

Which Go versions does Go memory diagnostics support?

Go memory diagnostics supports Go 1.18 and later. For best results, we recommend using the latest stable version of Go.

Will the diagnosis affect the running Go process?

The diagnostic process reads information from the /proc file system and Go runtime debugging data, which has a minimal performance impact on the Go process.

What should I do if the diagnosis fails?

Verify the following:

• Make sure that SysOM is correctly installed on the target instance.

• Verify that the PID corresponds to an existing Go process.

• Ensure that the current user has sufficient permissions to access the target instance.

• Check the error messages in the diagnostic records and troubleshoot based on the information provided.