[LTP] [PATCH] controllers/memcg: update regression tests to work under cgroup2
Luke Nowakowski-Krijger
luke.nowakowskikrijger@canonical.com
Wed Nov 17 22:23:01 CET 2021
Hi Richard,
I see what you mean, it would not work if both hierarchies exist. I will do
something similar to the patch you linked as that has cleared up a lot of
my misunderstandings about it. While I am there I will update the tests to
use the newer test APIs.
Thanks for the review.
- Luke
On Tue, Nov 16, 2021 at 2:09 AM Richard Palethorpe <rpalethorpe@suse.de>
wrote:
> Hello Luke,
>
> Luke Nowakowski-Krijger <luke.nowakowskikrijger@canonical.com> writes:
>
> > Some tests no longer make sense under cgroup2, while other tests just
> > needed to be updated to use new parameters introduced by the cgroup2
> > interface.
> >
> > Signed-off-by: Luke Nowakowski-Krijger <
> luke.nowakowskikrijger@canonical.com>
> > ---
> > .../memcg/regression/memcg_regression_test.sh | 41 ++++++++++++++++---
> > .../memcg/regression/memcg_test_1.c | 12 +++++-
> > .../memcg/regression/memcg_test_3.c | 8 ++++
> > .../memcg/regression/memcg_test_4.sh | 18 ++++++--
> > 4 files changed, 68 insertions(+), 11 deletions(-)
> >
> > diff --git
> a/testcases/kernel/controllers/memcg/regression/memcg_regression_test.sh
> b/testcases/kernel/controllers/memcg/regression/memcg_regression_test.sh
> > index c91a4069e..ad88d49d1 100755
> > ---
> a/testcases/kernel/controllers/memcg/regression/memcg_regression_test.sh
> > +++
> b/testcases/kernel/controllers/memcg/regression/memcg_regression_test.sh
> > @@ -103,7 +103,12 @@ check_kernel_bug()
> > test_1()
> > {
> > mkdir memcg/0/
> > - echo 0 > memcg/0/memory.limit_in_bytes
> > +
> > + if [ "$cgroup_ver" = "cgroup2" ]; then
> > + echo 0 > memcg/0/memory.max
> > + else
> > + echo 0 > memcg/0/memory.limit_in_bytes
> > + fi
> >
> > ./memcg_test_1
> >
> > @@ -124,6 +129,16 @@ test_1()
> >
> #---------------------------------------------------------------------------
> > test_2()
> > {
> > + # for cgroup2 the default behaivor is to check the new memory
> limit and
> > + # then to start killing processes if oom. This test then doesen't
> > + # make sense as we don't expect EBUSY to be returned. The shrink
> > + # operation (write to memory.max in cgroup2) would kill the pid1
> process
> > + # and exit.
> > + if [ "$cgroup_ver" = "cgroup2" ]; then
> > + tst_resm TCONF "cgroup2 found, skipping test"
> > + return
> > + fi
> > +
> > ./memcg_test_2 &
> > pid1=$!
> > sleep 1
> > @@ -177,12 +192,20 @@ test_2()
> > test_3()
> > {
> > mkdir memcg/0
> > - for pid in `cat memcg/tasks`; do
> > - echo $pid > memcg/0/tasks 2> /dev/null
> > + if [ "$cgroup_ver" = "cgroup2" ]; then
> > + memcg_procs=memcg/cgroup.procs
> > + memcg_subprocs=memcg/0/cgroup.procs
> > + else
> > + memcg_procs=memcg/tasks
> > + memcg_subprocs=memcg/0/tasks
> > + fi
> > +
> > + for pid in `cat $memcg_procs`; do
> > + echo $pid > $memcg_subprocs 2> /dev/null
> > done
> >
> > - for pid in `cat memcg/0/tasks`; do
> > - echo $pid > memcg/tasks 2> /dev/null
> > + for pid in `cat $memcg_subprocs`; do
> > + echo $pid > $memcg_procs 2> /dev/null
> > done
> > rmdir memcg/0
> >
> > @@ -218,13 +241,19 @@ test_4()
> > }
> >
> > # main
> > +cgroup_ver=$(grep "/sys/fs/cgroup" /proc/mounts | cut -d' ' -f1)
> > failed=0
> > mkdir memcg/
> >
> > for cur in $(seq 1 $TST_TOTAL); do
> > export TST_COUNT=$cur
> >
> > - mount -t cgroup -o memory xxx memcg/
> > + if [ "$cgroup_ver" = "cgroup2" ]; then
> > + mount -t cgroup2 xxx memcg/
> > + else
> > + mount -t cgroup -o memory xxx memcg/
> > + fi
> > +
> > if [ $? -ne 0 ]; then
> > tst_resm TFAIL "failed to mount memory subsystem"
> > failed=1
> > diff --git
> a/testcases/kernel/controllers/memcg/regression/memcg_test_1.c
> b/testcases/kernel/controllers/memcg/regression/memcg_test_1.c
> > index c7fb948fe..b9277e633 100644
> > --- a/testcases/kernel/controllers/memcg/regression/memcg_test_1.c
> > +++ b/testcases/kernel/controllers/memcg/regression/memcg_test_1.c
> > @@ -33,6 +33,16 @@
> >
> > #define FORKED_PROC_COUNT 10
> >
> > +static int open_cgroup_procs(void)
> > +{
> > + int fd = open("memcg/0/tasks", O_WRONLY);
> > + if (fd >= 0)
> > + return fd;
> > +
> > + fd = open("memcg/0/cgroup.procs", O_WRONLY);
> > + return fd;
> > +}
> > +
> > int main(void)
> > {
> > char buf[10];
> > @@ -40,7 +50,7 @@ int main(void)
> > int loop;
> > int pid;
> > int size = getpagesize();
> > - int fd = open("memcg/0/tasks", O_WRONLY);
> > + int fd = open_cgroup_procs();
> >
> > if (fd < 0)
> > return 1;
> > diff --git
> a/testcases/kernel/controllers/memcg/regression/memcg_test_3.c
> b/testcases/kernel/controllers/memcg/regression/memcg_test_3.c
> > index 75a6e1545..d5531fe87 100644
> > --- a/testcases/kernel/controllers/memcg/regression/memcg_test_3.c
> > +++ b/testcases/kernel/controllers/memcg/regression/memcg_test_3.c
> > @@ -66,6 +66,14 @@ static void setup(void)
> > SAFE_MKDIR(MNTPOINT, 0644);
> >
> > ret = mount("memcg", MNTPOINT, "cgroup", 0, "memory");
> > +
> > + if (!ret) {
> > + mount_flag = 1;
> > + return;
> > + }
> > +
> > + ret = mount("memcg", MNTPOINT, "cgroup2", 0, NULL);
> > +
> > if (ret) {
> > if (errno == ENOENT)
> > tst_brk(TCONF | TERRNO, "memcg not supported");
> > diff --git
> a/testcases/kernel/controllers/memcg/regression/memcg_test_4.sh
> b/testcases/kernel/controllers/memcg/regression/memcg_test_4.sh
> > index 620031366..287864b81 100755
> > --- a/testcases/kernel/controllers/memcg/regression/memcg_test_4.sh
> > +++ b/testcases/kernel/controllers/memcg/regression/memcg_test_4.sh
> > @@ -22,9 +22,19 @@
> > ##
> ##
> >
> ################################################################################
> >
> > +cgroup_ver=$(grep "/sys/fs/cgroup" /proc/mounts | cut -d' ' -f1)
>
> Unfortunately this doesn't work when both V1 and V2 hierarchies are
> mounted. You need to check that the memory controller is enabled on
> V2. Also it's possible to mount CGroups at alternate locations.
>
> Please see the recent patch from Masayoshi:
>
> https://patchwork.ozlabs.org/project/ltp/patch/20211113041706.12893-1-msys.mizuma@gmail.com/
>
> > +
> > +if [ "$cgroup_ver" = 'cgroup2' ]; then
> > + cgroup_proc=cgroup.procs
> > + cgroup_mem_limit=memory.max
> > +else
> > + cgroup_proc=tasks
> > + cgroup_mem_limit=memory.limit_in_bytes
> > +fi
> > +
> > # attach current task to memcg/0/
> > mkdir memcg/0
> > -echo $$ > memcg/0/tasks
> > +echo $$ > memcg/0/${cgroup_proc}
> >
> > ./memcg_test_4 &
> > pid=$!
> > @@ -35,13 +45,13 @@ sleep 1
> > sleep 1
> >
> > # shrink memory, and then 80M will be swapped
> > -echo 40M > memcg/0/memory.limit_in_bytes
> > +echo 40M > memcg/0/${cgroup_mem_limit}
> >
> > # turn off swap, and swapoff will be killed
> > swapoff -a
> > sleep 1
> > -echo $pid > memcg/tasks 2> /dev/null
> > -echo $$ > memcg/tasks 2> /dev/null
> > +echo $pid > memcg/${cgroup_proc} 2> /dev/null
> > +echo $$ > memcg/${cgroup_proc} 2> /dev/null
> >
> > # now remove the cgroup
> > rmdir memcg/0
> > --
> > 2.32.0
>
>
> --
> Thank you,
> Richard.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20211117/8a481418/attachment-0001.htm>
More information about the ltp
mailing list