Commit 9852ae3fe529 ("mm, memcg: consider subtrees in memory.events") made memory.events recursive: all events are propagated upwards by the tree. It was a change in semantics.
It broke the oom group leaf events test: it assumes that after an OOM the oom_kill counter is zero on parent's level.
Let's adjust the test: it should have similar expectations for the child and parent levels.
The test passes after this fix.
Signed-off-by: Roman Gushchin guro@fb.com Cc: Chris Down chris@chrisdown.name Cc: Johannes Weiner hannes@cmpxchg.org --- tools/testing/selftests/cgroup/test_memcontrol.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/tools/testing/selftests/cgroup/test_memcontrol.c b/tools/testing/selftests/cgroup/test_memcontrol.c index c19a97dd02d4..60bfe53c0289 100644 --- a/tools/testing/selftests/cgroup/test_memcontrol.c +++ b/tools/testing/selftests/cgroup/test_memcontrol.c @@ -1002,7 +1002,8 @@ static int test_memcg_sock(const char *root) /* * This test disables swapping and tries to allocate anonymous memory * up to OOM with memory.group.oom set. Then it checks that all - * processes in the leaf (but not the parent) were killed. + * processes in the leaf were killed. It also checks that oom_events + * were propagated to the parent level. */ static int test_memcg_oom_group_leaf_events(const char *root) { @@ -1045,7 +1046,7 @@ static int test_memcg_oom_group_leaf_events(const char *root) if (cg_read_key_long(child, "memory.events", "oom_kill ") <= 0) goto cleanup;
- if (cg_read_key_long(parent, "memory.events", "oom_kill ") != 0) + if (cg_read_key_long(parent, "memory.events", "oom_kill ") <= 0) goto cleanup;
ret = KSFT_PASS;
After commit 0e4b01df8659 ("mm, memcg: throttle allocators when failing reclaim over memory.high") allocating memory over memory.high became very time consuming. But it's exactly what the memory.high test from cgroup kselftests is doing: it tries to allocate 100M with 30M memory.high value. It takes forever to complete.
In order to keep it passing (or failing) in a reasonable amount of time let's try to allocate only a little over 30M: 31M to be precise.
With this change test_memcontrol finishes in a reasonable amount of time: $ time ./test_memcontrol ok 1 test_memcg_subtree_control ok 2 test_memcg_current ok 3 test_memcg_min ok 4 test_memcg_low ok 5 test_memcg_high ok 6 test_memcg_max ok 7 test_memcg_oom_events ok 8 test_memcg_swap_max ok 9 test_memcg_sock ok 10 test_memcg_oom_group_leaf_events ok 11 test_memcg_oom_group_parent_events ok 12 test_memcg_oom_group_score_events
real 0m2.273s user 0m0.064s sys 0m0.739s
Signed-off-by: Roman Gushchin guro@fb.com Cc: Chris Down chris@chrisdown.name Cc: Johannes Weiner hannes@cmpxchg.org --- tools/testing/selftests/cgroup/test_memcontrol.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tools/testing/selftests/cgroup/test_memcontrol.c b/tools/testing/selftests/cgroup/test_memcontrol.c index 60bfe53c0289..739fa90448f6 100644 --- a/tools/testing/selftests/cgroup/test_memcontrol.c +++ b/tools/testing/selftests/cgroup/test_memcontrol.c @@ -606,7 +606,7 @@ static int test_memcg_high(const char *root) if (cg_write(memcg, "memory.high", "30M")) goto cleanup;
- if (cg_run(memcg, alloc_anon, (void *)MB(100))) + if (cg_run(memcg, alloc_anon, (void *)MB(31))) goto cleanup;
if (!cg_run(memcg, alloc_pagecache_50M_check, NULL))
Roman Gushchin writes:
After commit 0e4b01df8659 ("mm, memcg: throttle allocators when failing reclaim over memory.high") allocating memory over memory.high became very time consuming. But it's exactly what the memory.high test from cgroup kselftests is doing: it tries to allocate 100M with 30M memory.high value. It takes forever to complete.
In order to keep it passing (or failing) in a reasonable amount of time let's try to allocate only a little over 30M: 31M to be precise.
With this change test_memcontrol finishes in a reasonable amount of time: $ time ./test_memcontrol ok 1 test_memcg_subtree_control ok 2 test_memcg_current ok 3 test_memcg_min ok 4 test_memcg_low ok 5 test_memcg_high ok 6 test_memcg_max ok 7 test_memcg_oom_events ok 8 test_memcg_swap_max ok 9 test_memcg_sock ok 10 test_memcg_oom_group_leaf_events ok 11 test_memcg_oom_group_parent_events ok 12 test_memcg_oom_group_score_events
real 0m2.273s user 0m0.064s sys 0m0.739s
Signed-off-by: Roman Gushchin guro@fb.com Cc: Chris Down chris@chrisdown.name Cc: Johannes Weiner hannes@cmpxchg.org
Thanks, this makes sense.
Reviewed-by: Chris Down chris@chrisdown.name
Roman Gushchin writes:
Commit 9852ae3fe529 ("mm, memcg: consider subtrees in memory.events") made memory.events recursive: all events are propagated upwards by the tree. It was a change in semantics.
It broke the oom group leaf events test: it assumes that after an OOM the oom_kill counter is zero on parent's level.
Let's adjust the test: it should have similar expectations for the child and parent levels.
The test passes after this fix.
Thanks! Hopefully b59b1baa will also avoid this going unnoticed in future.
Reviewed-by: Chris Down chris@chrisdown.name
linux-kselftest-mirror@lists.linaro.org