Topics to cover:

VSZ and RSS

VSZ: Virtual memory SiZe - malloc-ing will increase the VSZ.
RSS: Resident Set Size - will not be increased by malloc-ing, but will increase only when actually writing to that pre-allocated memory, by the ammount written. So for example you can allocate 10 GiB of RAM and only write 2 GiB to it. That will mean VSZ += 10 GiB and RSS += 2 GiB.

RSS is always smaller than (<) VSZ.

Remember that VSZ and RSS values are not very easy to explain. So, for example while a simple hello world C++ program always starts with the same VSZ = 5924 kB, the RSS is approximately equal to 1872 kB, but it varies on each run. Looking at /proc/PID/smap you can see for example that the [stack] is not always constant. See also: https://stackoverflow.com/questions/67904845/rss-stack-memory-allocation-not-consistent .

Here’s an example:

#include <unistd.h> // getpid()
#include <iostream>
#include <string>
using namespace ::std;

int main()
{
  cout << "PID=" << getpid() << endl;
  string a = "";
  // At this point ps and top should report that this process is using:
  // VSZ: 5924 kB and RSS ~= 1872 kB (RSS will vary from one run to another)
  getline(cin, a);


  size_t no_bytes = 50*1024*1024; // 50 MB
  cout << "Allocating " << no_bytes << " bytes ..." << endl;
  char * p = (char*)malloc(no_bytes);
  if( p == 0 )
  {
    cout << "Allocation failed" << endl;
    return 1;
  }
  cout << "Allocation succeeded" << endl;
  // This means that VSZ is now: 57128 kB
  getline(cin, a);


  cout << "Filling memory - " << no_bytes << " bytes ..." << endl;
  for(char * c = (char*)p; c < p+no_bytes/1; c++) // only fill partially 1/2 ?
  {
    // cout << "c=" << (int*)c << ". *c set to 0" << endl;
    *c = *(char*)(&c);
  }
  cout << "Filling succeeded" << endl;
  // This means that RSS is now ~ 54468 kB
  // RSS += 52712 or 52628 ... this also varies
  getline(cin, a);


  return 0;
}

OOM killer

The OOM (Out Of Memory) killer, is invoked by the kernel when the memory is full(or almost full?). This will kill a process in order to free memory on the system. It will try to kill the process which uses most memory, thus to free as much memory as possible. But it will also try to kill the most unimportant process. For more details see the oom_score of processes. See also: https://unix.stackexchange.com/a/153586 . Note that OOM Killer can also be invoked per control-group. Thus having an OOM Killer kill your process only means that a certain memory limit was reached, not necessary that the host ran out of RAM - maybe you just had a container memory limit set to your process.

Remember that the default linux page size is 4KB = 4 * 1000 bytes. And in the kernel logs, when the OOM killer is invoked, the vm and rss memories are expressed in pages of (4000 bytes each).

core dump

Using a kernel parameter config variable, you can configure it where to dump the core files. Core files contain memory used by the application, and can be used together with the tool gdb and your binary file(that produced the core), in order to debug the problem and see call-stack, variable values and so on.

ram used vs buff+cache vs free

Disk caching in Linux uses your free RAM, creating the feeling that you are low on ram. See : https://www.linuxatemyram.com/ . This cache is supposed to be freed when an application needs more ram. But I still feel that in some cases of high demand, it’s not freed fast enough.


---

Here's a small snippet that you can use to run C/C++ code "as a script", just copy and paste it to run it:
```bash
# ============================================================================ #
(
tmp_bin=$(mktemp) &&
mv ${tmp_bin} ${tmp_bin}.bin &&
tmp_bin="${tmp_bin}.bin" &&

tmp_src=$(mktemp) &&
mv ${tmp_src} ${tmp_src}.cpp &&
tmp_src="${tmp_src}.cpp" &&

(cat - <<'EOF'
// --------------------------- CPP source code begin ------------------------ //
#include <iostream>
using namespace ::std;

int main()
{
  cout << "Hello world" << endl;
  return 0;
}
// --------------------------- CPP source code end -------------------------- //
EOF
) > ${tmp_src} &&

g++ ${tmp_src} -O3 -o ${tmp_bin} &&
rm -f ${tmp_src} &&
${tmp_bin};
rm -f ${tmp_bin}
) # Take the following empty line when you copy-paste

# ============================================================================ #

Or this one for C:

# ============================================================================ #
(
tmp_bin=$(mktemp) &&
mv ${tmp_bin} ${tmp_bin}.bin &&
tmp_bin="${tmp_bin}.bin" &&

tmp_src=$(mktemp) &&
mv ${tmp_src} ${tmp_src}.cpp &&
tmp_src="${tmp_src}.cpp" &&

(cat - <<'EOF'
// ---------------------------- C source code begin ------------------------- //
#include <stdio.h> // printf()

int main()
{
  printf("Hello C !\n");
  return 0;
}
// ---------------------------- C source code end --------------------------- //
EOF
) > ${tmp_src} &&

gcc ${tmp_src} -o ${tmp_bin} &&
rm -f ${tmp_src} &&
${tmp_bin};
rm -f ${tmp_bin}
) # Take the following empty line when you copy-paste

# ============================================================================ #