Hello, everyone!
Today, I'd like to share a case with you.
Problem Description
OSN 9800 Memory Leak Causes the MEM_OVER Alarm to Be Reported
Problem Analysis
Step 1: According to the alarm information, the active SCC board in slot 78 reported the MEM_OVER alarm. The log shows that the memory usage of the map part of the L0L1 process exceeds 716M, which is abnormal. It is suspected that memory leakage occurs in the L0L1 process.
Step 2: Analyze and confirm that Mmap memory statistics directly invoke the mmap() interface or invoke the cloud () interface to apply for a memory larger than 128k.
Step 3: Observe the map memory growth rule and map the virtual memory mapping table of the L0L1 process. It is found that the memory usage keeps increasing by 1 M.
Step 4: The mirroring environment of the same version is set up in the lab. The memory usage of the mmap keeps increasing, which is the same as that on the live network.
Step 5: When the system memory is increasing, use a tool to capture the call stack that is larger than 128k. Analyze the virtual memory address segment added by the map. It is found that the 1M memory block is applied to the opendir()/fopen() interface.
Step 6: Analyze the call stack code and memory management mechanism. It is inferred that the small memory is applied by the () interface, but the memory is not released.
Step 7: Install the log software in the lab. It is found that the size of the memory that is not released is 24(0x18) bytes.
Step 8: According to the code analysis, when the socket option is set to a specified parameter, the memory that is applied for 24 bytes is not released. This is a software defect.
Step 9: After the software is installed and restored in the lab, the memory usage is not continuously increased for 24 hours. It is confirmed that the memory leak occurs on the main control board due to the Sock_xxxx() interface.
Root Cause
A memory leak occurs on the Sock_xxxx() interface. As a result, the system memory usage increases and the MEM_OVER alarm is reported.
Solution Description
The problem has been resolved in V1R19C10SPC100 and is planned to be upgraded on the BHARTI network.
Welcome to leave a message below.
We study together.
Thank you!
