I am using docker to containerize bunch of services. Some times, containerized services heavily swaps. I've changed vm.swappiness
to 1
via sysctl on my host system. But docker's memory cgroup still have old (default) value of 60. Therefore, all particular containers' cgroups have same value, as parent.
sysctl vm.swappiness
> vm.swappiness = 1
cat /sys/fs/cgroup/memory/docker/memory.swappiness
> 60
cat /sys/fs/cgroup/memory/docker/${CONTAINER_ID}/memory.swappiness
> 60
All attempts to change swappiness manually (by echoing desired value into memory.swappiness
file) fails with permission denied
.
Subject: How can I restrict containers swappiness?
I am using ubuntu 12.04
with kernel 3.13
, my docker version is 1.1.2
with native execution driver (not lxc) of version 0.2
. Kernel is loaded with cgroup_enable=memory swapaccount=1
.
See --memory-swap details. --memory-swappiness. By default, the host kernel can swap out a percentage of anonymous pages used by a container. You can set --memory-swappiness to a value between 0 and 100, to tune this percentage.
Set Maximum Memory Access To limit the maximum amount of memory usage for a container, add the --memory option to the docker run command. Alternatively, you can use the shortcut -m . Within the command, specify how much memory you want to dedicate to that specific container.
Restart policy detailsA restart policy only takes effect after a container starts successfully. In this case, starting successfully means that the container is up for at least 10 seconds and Docker has started monitoring it. This prevents a container which does not start at all from going into a restart loop.
Got it! Docker does not even touch this parameter. memory.swappines
for cgroups really changing according /proc/vm/swappiness
. All children inherits this value from parent. Docker does not even touch this parameter. Moreover, in some cases (and exactly in my own) there is no ability to write something in memory.swappines
. If memory cgroup uses hierarchy, or contains children, all attempts to write something to cgroups memory.swappiness
will fail.
Look there. This is from mm/memcontrol.c
.
static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
struct cftype *cft, u64 val)
{
struct mem_cgroup *memcg = mem_cgroup_from_css(css);
struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
if (val > 100 || !parent)
return -EINVAL;
mutex_lock(&memcg_create_mutex);
/* If under hierarchy, only empty-root can set this value */
if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
mutex_unlock(&memcg_create_mutex);
return -EINVAL;
}
memcg->swappiness = val;
mutex_unlock(&memcg_create_mutex);
return 0;
}
If you upgrade to a 3.18 kernel or later, the restriction preventing modification to the cgroup memory.swappiness
parameter in child/hierarchy cgroups is removed. The Linux kernel patch which removed this restriction can be seen here: https://github.com/torvalds/linux/commit/3dae7fec5e884a4e72e5416db0894de66f586201
Docker 1.8 will most likely include the following PR (https://github.com/docker/docker/pull/14004) which allows the container to set its own memory.swappiness
value allowing user control over this cgroup setting, as long as the Docker daemon host kernel has the patch noted above, or the host kernel is 3.18 or greater.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With