Re: [Devel] [PATCH] cgroup/cpuset: emulate cgroup in container
Personally I don't like these as we still have no unswer to "If cpusets are optional for docker, why k8s can't work without them?" it seem there is not enough explanation in VZAP-31. We also need to revert the patch below to show cpuset in CT: commit 5160bd34c9bd ("ve/proc/cpuset: do not show cpuset in CT") It seem I can still attach a process to a nested cgroup in CT with these patch: CT-6ecd9be1 /# cat /proc/cgroups | grep cpuset cpuset 16 1 1 CT-6ecd9be1 /# ls /sys/fs/cgroup/cpuset/cpuset.cpus /sys/fs/cgroup/cpuset/cpuset.cpus CT-6ecd9be1 /# mkdir /sys/fs/cgroup/cpuset/test CT-6ecd9be1 /# sleep 1000 & [1] 678 CT-6ecd9be1 /# echo 678 > /sys/fs/cgroup/cpuset/test/tasks CT-6ecd9be1 /# cat /sys/fs/cgroup/cpuset/test/tasks 678 On 12/13/2017 01:37 PM, Stanislav Kinsburskiy wrote: Any changes to this cgroup are skipped in container, but success code is returned. The idea is to fool Docker/Kubernetes. https://jira.sw.ru/browse/PSBM-58423 This patch obsoletes "ve/proc/cpuset: do not show cpuset in CT" Signed-off-by: Stanislav Kinsburskiy --- kernel/cpuset.c |9 + 1 file changed, 9 insertions(+) diff --git a/kernel/cpuset.c b/kernel/cpuset.c index 26d88eb..dfac505 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -1441,6 +1441,9 @@ static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) struct task_struct *task; int ret; + if (!ve_is_super(get_exec_env())) + return 0; + mutex_lock(&cpuset_mutex); ret = -ENOSPC; @@ -1470,6 +1473,9 @@ static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) static void cpuset_cancel_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) { + if (!ve_is_super(get_exec_env())) + return; + mutex_lock(&cpuset_mutex); cgroup_cs(cgrp)->attach_in_progress--; mutex_unlock(&cpuset_mutex); @@ -1494,6 +1500,9 @@ static void cpuset_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) struct cpuset *cs = cgroup_cs(cgrp); struct cpuset *oldcs = cgroup_cs(oldcgrp); + if (!ve_is_super(get_exec_env())) + return; + mutex_lock(&cpuset_mutex); /* prepare for attach */ -- Best regards, Tikhomirov Pavel Software Developer, Virtuozzo. ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
Re: [Devel] [PATCH] cgroup/cpuset: emulate cgroup in container
Hi Pavel, please, see my comments/question below 13.12.2017 13:43, Pavel Tikhomirov пишет: > Personally I don't like these as we still have no unswer to "If cpusets are > optional for docker, why k8s can't work without them?" it seem there is not > enough explanation in VZAP-31. > Well... I have to admit that I don't like it either. I (frankly) don't like the whole idea of putting kuber into container. And the reason is so simple: CT is a cheating technique. Those, who like "Russian dolls" should use VMs with nested virtualization. But the truth is that we do care about our customers (not sure why, since they pay us less and less). And because of this we've put so much various sh*t into our kernel already, that I got tired to trow it away on each major rebase a long time ago. But this all is lyrics. There is a task - and that's the fix. > We also need to revert the patch below to show cpuset in CT: > commit 5160bd34c9bd ("ve/proc/cpuset: do not show cpuset in CT") > Sure! It's mentioned in the patch description. > It seem I can still attach a process to a nested cgroup in CT with these > patch: > > CT-6ecd9be1 /# cat /proc/cgroups | grep cpuset > cpuset 16 1 1 > CT-6ecd9be1 /# ls /sys/fs/cgroup/cpuset/cpuset.cpus > /sys/fs/cgroup/cpuset/cpuset.cpus > CT-6ecd9be1 /# mkdir /sys/fs/cgroup/cpuset/test > CT-6ecd9be1 /# sleep 1000 & > [1] 678 > CT-6ecd9be1 /# echo 678 > /sys/fs/cgroup/cpuset/test/tasks > CT-6ecd9be1 /# cat /sys/fs/cgroup/cpuset/test/tasks > 678 > Isn't it wonderful? :) Poor me, I have to admit, that I didn't know, that this task will be even visible in the nest cgroup... :( I thought, that emulation would be less effective. Nevertheless, if you have some better way to solve the issue, please, share. > On 12/13/2017 01:37 PM, Stanislav Kinsburskiy wrote: >> Any changes to this cgroup are skipped in container, but success code is >> returned. >> The idea is to fool Docker/Kubernetes. >> >> https://jira.sw.ru/browse/PSBM-58423 >> >> This patch obsoletes "ve/proc/cpuset: do not show cpuset in CT" >> >> Signed-off-by: Stanislav Kinsburskiy >> --- >> kernel/cpuset.c | 9 + >> 1 file changed, 9 insertions(+) >> >> diff --git a/kernel/cpuset.c b/kernel/cpuset.c >> index 26d88eb..dfac505 100644 >> --- a/kernel/cpuset.c >> +++ b/kernel/cpuset.c >> @@ -1441,6 +1441,9 @@ static int cpuset_can_attach(struct cgroup *cgrp, >> struct cgroup_taskset *tset) >> struct task_struct *task; >> int ret; >> + if (!ve_is_super(get_exec_env())) >> + return 0; >> + >> mutex_lock(&cpuset_mutex); >> ret = -ENOSPC; >> @@ -1470,6 +1473,9 @@ static int cpuset_can_attach(struct cgroup *cgrp, >> struct cgroup_taskset *tset) >> static void cpuset_cancel_attach(struct cgroup *cgrp, >> struct cgroup_taskset *tset) >> { >> + if (!ve_is_super(get_exec_env())) >> + return; >> + >> mutex_lock(&cpuset_mutex); >> cgroup_cs(cgrp)->attach_in_progress--; >> mutex_unlock(&cpuset_mutex); >> @@ -1494,6 +1500,9 @@ static void cpuset_attach(struct cgroup *cgrp, struct >> cgroup_taskset *tset) >> struct cpuset *cs = cgroup_cs(cgrp); >> struct cpuset *oldcs = cgroup_cs(oldcgrp); >> + if (!ve_is_super(get_exec_env())) >> + return; >> + >> mutex_lock(&cpuset_mutex); >> /* prepare for attach */ >> > ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
Re: [Devel] [PATCH] cgroup/cpuset: emulate cgroup in container
On 12/13/2017 04:24 PM, Stanislav Kinsburskiy wrote: Hi Pavel, please, see my comments/question below 13.12.2017 13:43, Pavel Tikhomirov пишет: Personally I don't like these as we still have no unswer to "If cpusets are optional for docker, why k8s can't work without them?" it seem there is not enough explanation in VZAP-31. Well... I have to admit that I don't like it either. I (frankly) don't like the whole idea of putting kuber into container. And the reason is so simple: CT is a cheating technique. Those, who like "Russian dolls" should use VMs with nested virtualization. But the truth is that we do care about our customers (not sure why, since they pay us less and less). And because of this we've put so much various sh*t into our kernel already, that I got tired to trow it away on each major rebase a long time ago. But this all is lyrics. There is a task - and that's the fix. Sure, I agree. We also need to revert the patch below to show cpuset in CT: commit 5160bd34c9bd ("ve/proc/cpuset: do not show cpuset in CT") Sure! It's mentioned in the patch description. Sorry, missed it, now I see. Thanks for pointing that out! It seem I can still attach a process to a nested cgroup in CT with these patch: CT-6ecd9be1 /# cat /proc/cgroups | grep cpuset cpuset 16 1 1 CT-6ecd9be1 /# ls /sys/fs/cgroup/cpuset/cpuset.cpus /sys/fs/cgroup/cpuset/cpuset.cpus CT-6ecd9be1 /# mkdir /sys/fs/cgroup/cpuset/test CT-6ecd9be1 /# sleep 1000 & [1] 678 CT-6ecd9be1 /# echo 678 > /sys/fs/cgroup/cpuset/test/tasks CT-6ecd9be1 /# cat /sys/fs/cgroup/cpuset/test/tasks 678 Isn't it wonderful? :) Poor me, I have to admit, that I didn't know, that this task will be even visible in the nest cgroup... :( I thought, that emulation would be less effective. May be I don't understand your patch completely, but my guess was - you want to fake attaching to cpuset cgroups so that attach does not actually moves process in it but just says OK. Thus all configurations of nested cgroups does not matter as there are no tasks in them and we are OK. Now, AFAIKS, we have a partial state when task is moved to other cgroup but it's cpuset is left unchanged. I can still apply cpuset to a task from CT (sorry for a lot of lines): 1) Host is idle. [root@silo ~]# mpstat 1 1 -P ALL Linux 3.10.0-693.11.1.ovz.39.4 (silo.sw.ru) 12/13/2017 _x86_64_ (4 CPU) 06:02:08 PM CPU%usr %nice%sys %iowait%irq %soft %steal %guest %gnice %idle 06:02:09 PM all0.000.000.000.000.000.250.00 0.000.00 99.75 06:02:09 PM00.000.000.000.000.000.000.00 0.000.00 100.00 06:02:09 PM10.000.000.000.000.000.000.00 0.000.00 100.00 06:02:09 PM20.000.000.000.000.000.000.00 0.000.00 100.00 06:02:09 PM30.000.000.000.000.000.000.00 0.000.00 100.00 Average: CPU%usr %nice%sys %iowait%irq %soft %steal %guest %gnice %idle Average: all0.000.000.000.000.000.250.00 0.000.00 99.75 Average: 00.000.000.000.000.000.000.00 0.000.00 100.00 Average: 10.000.000.000.000.000.000.00 0.000.00 100.00 Average: 20.000.000.000.000.000.000.00 0.000.00 100.00 Average: 30.000.000.000.000.000.000.00 0.000.00 100.00 2) Run something in CT: CT-6ecd9be1 /# dd if=/dev/zero of=/dev/null & [1] 758 3) It burns cpu0: [root@silo ~]# mpstat 1 1 -P ALL Linux 3.10.0-693.11.1.ovz.39.4 (silo.sw.ru) 12/13/2017 _x86_64_ (4 CPU) 06:02:18 PM CPU%usr %nice%sys %iowait%irq %soft %steal %guest %gnice %idle 06:02:19 PM all9.750.00 15.500.000.000.000.00 0.000.00 74.75 06:02:19 PM0 38.610.00 61.390.000.000.000.00 0.000.000.00 06:02:19 PM10.000.000.000.000.000.000.00 0.000.00 100.00 06:02:19 PM20.000.000.000.000.000.000.00 0.000.00 100.00 06:02:19 PM30.000.000.000.000.000.000.00 0.000.00 100.00 Average: CPU%usr %nice%sys %iowait%irq %soft %steal %guest %gnice %idle Average: all9.750.00 15.500.000.000.000.00 0.000.00 74.75 Average: 0 38.610.00 61.390.000.000.000.00 0.000.000.00 Average: 10.000.000.000.000.000.000.00 0.000.00 100.00 Average: 20.000.000.000.000.000.000.00 0.000.00 100.00 Average: 30.000.000.000.000.000.000.00 0.000.00 100.00 4) Create
Re: [Devel] [PATCH] cgroup/cpuset: emulate cgroup in container
Nice catch, thanks! I'll first try to tweak this gently, so it will look like cpuset cgroup works, but it won't. 13.12.2017 16:18, Pavel Tikhomirov пишет: > So change to cpuset.cpus is actually applied to the task, that's what I mean. > Correct me if I'm wrong. ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel