I found a solution for my problem. I think this was a kubernetes-specific thing. After some research I found out that the worker-process get the status Zombie after sending the SIGTERM and the parent-process was not able to clean up the Zombie. So after adding tini as init-process (on dockerfile and deployment.yaml) and make sure it gets PID1 now the zombies are cleaned up and the worker-process is terminating when I send the SIGTERM from the admin-process.

I also tried to improve my signal-handler to make it safe again (thanks for your information in the comments). Now here the updated signal-handler and some parts of the rest of the program:

volatile atomic_int execute_loop = 1; 

void handle_sigterm() {
    error_msg(sip_man_log, "(MAIN) INFO: Signal reached.");
    execute_loop = 0; 
    error_msg(sip_man_log, "(MAIN) INFO: Closing all sockets.");
    close(sockfd_ext);
    close(connfd); 
    close(sockfd); 
}

int main()
{
    struct sockaddr_in  sockaddr, connaddr, sockaddr_ext;
    unsigned int        connaddr_len;
    char                buffer[8192];
    int                 rv, rv_ext;

    signal(SIGTERM, handle_sigterm); 
    
    sockfd = socket(AF_INET, SOCK_STREAM, 0);
    [...]

    while(execute_loop)
    {
        connaddr_len = sizeof(connaddr);
        connfd = accept(sockfd, (struct sockaddr*)&connaddr, &connaddr_len);
    [...]
    }
    free_manipulation_table(int_modification_table);
    free_manipulation_table(ext_modification_table); 
    free_manipulation_table(mir_modification_table);
    error_msg(sip_man_log, "(MAIN) INFO: Terminating Server by SIGTERM"); 

    return 0;
}

It's very shortened and I skipped most of the error-handling. Thanks for your help!

Dennis