[v3,1/2] controllers/cpuacct: skip cpuacct_100_100 on small memory systems

Running cpuacct_100_100 on a system with less than ~4 GB of RAM fails:

    cpuacct 1 TINFO: timeout per run is 0h 5m 0s
    cpuacct 1 TINFO: cpuacct: /sys/fs/cgroup/cpu,cpuacct
    cpuacct 1 TINFO: Creating 100 subgroups each with 100 processes
    testcases/bin/cpuacct.sh: 0: Cannot fork

In dmesg:

    LTP: starting cpuacct_100_100 (cpuacct.sh 100 100)
    cgroup: fork rejected by pids controller in /user.slice/user-1000.slice/session-1.scope

The reason is cgroups pid limit set by systemd user.slice.  For example
on 2 GB RAM machine it is set as:
    /sys/fs/cgroup/pids/user.slice/user-0.slice/pids.max:5207

Add a check for both free memory and maximum number of pids (when using
systemd) to skip the test.  Systems not using systemd might still fail.

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>

---

Changes since v2:
1. Correct realuid->real_uid typo.
2. Correct handling dash shell tests for return value.
3. Correct handling numbers - infinite.

Changes since v1:
1. Add checking for pids.
2. Accurately check the memory constraints.
---
 .../kernel/controllers/cpuacct/cpuacct.sh     | 60 +++++++++++++++++++
 1 file changed, 60 insertions(+)

Hi!
> +check_free_memory()
> +{
> +	local memneeded
> +	local memfree=`awk '/MemAvailable/ {print $2}' /proc/meminfo`
> +
> +	if [ $? -ne 0 ]; then
> +		local memcached
> +
> +		memfree=`awk '/MemFree/ {print $2}' /proc/meminfo`
> +		test $? -eq 0 || return 0
> +
> +		memcached=`awk '/MemCached/ {print $2}' /proc/meminfo`
> +		test $? -eq 0 || return 0
> +
> +		memfree=$((memfree + memcached))
> +	fi
> +
> +	# On x86_64, each 100 of processes were using ~16 MB of memory,
> +	# so try to estimate the needed free memory based on this.
> +	memneeded=$((max * nbprocess * 16384 / 100))
> +
> +	if [ $memfree -lt $memneeded ]; then
> +		tst_brk TCONF "not enough of free memory on this system (approximate need $memneeded kB, free $memfree kB)"
> +	fi
> +	tst_res TINFO "memory requirements fulfilled (approximate need $memneeded kB, free $memfree kB)"
> +
> +	return 0
> +}
> +
> +check_limits()
> +{
> +	local real_uid="$SUDO_UID"
> +	local tasksneeded=$((max * nbprocess + 100))
> +
> +	if [ "$real_uid" = "" ]; then
> +		real_uid=`id -u`
> +		test $? -eq 0 || return 0
> +	fi
> +
> +	local tasksmax=`systemctl show user-${real_uid}.slice | awk -F '=' '/TasksMax/ {print $2}'`
> +	test $? -eq 0 || return 0
> +
> +	# Check for number for cases like TasksMax=infinity
> +	[ -n "$tasksmax" ] && [ "$tasksmax" -eq "$tasksmax" ] 2>/dev/null
> +	test $? -eq 0 || return 0

Well we do have a C implementation of this now, it would be better to
have the logic in one place and just call the C function to collect the
number of pids.

If we add tst_get_free_pids.c to the testcases/lib/ directory with:

#define TST_NO_DEFAULT_MAIN
#include <stdio.h>
#include <tst_test.h>

extern struct tst_test *tst_test;

static struct tst_test test = {
};

int main(void)
{
	/* force messages to be printed from new library */
        tst_test = &test;

        printf("%i\n", tst_get_free_pids());

        return 0;
}

We can do FREE_PIDS=$(tst_get_free_pids) here and then just compare the numbers.

> +	if [ $tasksmax -le $tasksneeded ]; then
> +		tst_brk TCONF "limit of tasks is too low (approximate need $tasksneeded, limit $tasksmax)"
> +	fi
> +	tst_res TINFO "task limit fulfilled (approximate need $tasksneeded, limit $tasksmax)"
> +
> +	return 0
> +}
> +
>  setup()
>  {
>  	if ! grep -q -w cpuacct /proc/cgroups; then
>  		tst_brk TCONF "cpuacct not supported on this system"
>  	fi
>  
> +	check_limits
> +	# Don't bother with memory limit checks on smaller tests
> +	if [ $max -ge 100 ] && [ $nbprocess -ge 100 ]; then
> +		check_free_memory
> +	fi
> +
>  	mount_point=`grep -w cpuacct /proc/mounts | cut -f 2 | cut -d " " -f2`
>  	tst_res TINFO "cpuacct: $mount_point"
>  	if [ "$mount_point" = "" ]; then
> -- 
> 2.30.2
> 
> 
> -- 
> Mailing list info: https://lists.linux.it/listinfo/ltp

On 01/09/2021 15:17, Cyril Hrubis wrote:
> Hi!
>> +check_free_memory()
>> +{
>> +	local memneeded
>> +	local memfree=`awk '/MemAvailable/ {print $2}' /proc/meminfo`
>> +
>> +	if [ $? -ne 0 ]; then
>> +		local memcached
>> +
>> +		memfree=`awk '/MemFree/ {print $2}' /proc/meminfo`
>> +		test $? -eq 0 || return 0
>> +
>> +		memcached=`awk '/MemCached/ {print $2}' /proc/meminfo`
>> +		test $? -eq 0 || return 0
>> +
>> +		memfree=$((memfree + memcached))
>> +	fi
>> +
>> +	# On x86_64, each 100 of processes were using ~16 MB of memory,
>> +	# so try to estimate the needed free memory based on this.
>> +	memneeded=$((max * nbprocess * 16384 / 100))
>> +
>> +	if [ $memfree -lt $memneeded ]; then
>> +		tst_brk TCONF "not enough of free memory on this system (approximate need $memneeded kB, free $memfree kB)"
>> +	fi
>> +	tst_res TINFO "memory requirements fulfilled (approximate need $memneeded kB, free $memfree kB)"
>> +
>> +	return 0
>> +}
>> +
>> +check_limits()
>> +{
>> +	local real_uid="$SUDO_UID"
>> +	local tasksneeded=$((max * nbprocess + 100))
>> +
>> +	if [ "$real_uid" = "" ]; then
>> +		real_uid=`id -u`
>> +		test $? -eq 0 || return 0
>> +	fi
>> +
>> +	local tasksmax=`systemctl show user-${real_uid}.slice | awk -F '=' '/TasksMax/ {print $2}'`
>> +	test $? -eq 0 || return 0
>> +
>> +	# Check for number for cases like TasksMax=infinity
>> +	[ -n "$tasksmax" ] && [ "$tasksmax" -eq "$tasksmax" ] 2>/dev/null
>> +	test $? -eq 0 || return 0
> 
> Well we do have a C implementation of this now, it would be better to
> have the logic in one place and just call the C function to collect the
> number of pids.
> 
> If we add tst_get_free_pids.c to the testcases/lib/ directory with:
> 
> #define TST_NO_DEFAULT_MAIN
> #include <stdio.h>
> #include <tst_test.h>
> 
> extern struct tst_test *tst_test;
> 
> static struct tst_test test = {
> };
> 
> int main(void)
> {
> 	/* force messages to be printed from new library */
>         tst_test = &test;
> 
>         printf("%i\n", tst_get_free_pids());
> 
>         return 0;
> }
> 
> We can do FREE_PIDS=$(tst_get_free_pids) here and then just compare the numbers.
> 

I'll move this piece, as you suggested. The free memory limits is okay
to stay here?


Best regards,
Krzysztof

Hi!
> I'll move this piece, as you suggested. The free memory limits is okay
> to stay here?

There is no such API in the C library yet, so unless you want to add it
it can stay where it is.