daemon: stop setting container resources to zero

Many of the fields in LinuxResources struct are pointers to scalars for
some reason, presumably to differentiate between set-to-zero and unset
when unmarshaling from JSON, despite zero being outside the acceptable
range for the corresponding kernel tunables. When creating the OCI spec
for a container, the daemon sets the container's OCI spec CPUShares and
BlkioWeight parameters to zero when the corresponding Docker container
configuration values are zero, signifying unset, despite the minimum
acceptable value for CPUShares being two, and BlkioWeight ten. This has
gone unnoticed as runC does not distingiush set-to-zero from unset as it
also uses zero internally to represent unset for those fields. However,
kata-containers v3.2.0-alpha.3 tries to apply the explicit-zero resource
parameters to the container, exactly as instructed, and fails loudly.
The OCI runtime-spec is silent on how the runtime should handle the case
when those parameters are explicitly set to out-of-range values and
kata's behaviour is not unreasonable, so the daemon must therefore be in
the wrong.

Translate unset values in the Docker container's resources HostConfig to
omit the corresponding fields in the container's OCI spec when starting
and updating a container in order to maximize compatibility with
runtimes.

Signed-off-by: Cory Snider <csnider@mirantis.com>
(cherry picked from commit dea870f4ea)
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
This commit is contained in:
Cory Snider 2023-06-05 18:44:51 -04:00 committed by Sebastiaan van Stijn
parent 0a6a5a9140
commit f50cb0c7bd
No known key found for this signature in database
GPG key ID: 76698F39D527CE8C
7 changed files with 98 additions and 24 deletions

View file

@ -105,7 +105,10 @@ func getMemoryResources(config containertypes.Resources) *specs.LinuxMemory {
memory.KernelTCP = &config.KernelMemoryTCP
}
return &memory
if memory != (specs.LinuxMemory{}) {
return &memory
}
return nil
}
func getPidsLimit(config containertypes.Resources) *specs.LinuxPids {
@ -127,7 +130,7 @@ func getCPUResources(config containertypes.Resources) (*specs.LinuxCPU, error) {
if config.CPUShares < 0 {
return nil, fmt.Errorf("shares: invalid argument")
}
if config.CPUShares >= 0 {
if config.CPUShares > 0 {
shares := uint64(config.CPUShares)
cpu.Shares = &shares
}
@ -168,7 +171,10 @@ func getCPUResources(config containertypes.Resources) (*specs.LinuxCPU, error) {
cpu.RealtimeRuntime = &c
}
return &cpu, nil
if cpu != (specs.LinuxCPU{}) {
return &cpu, nil
}
return nil, nil
}
func getBlkioWeightDevices(config containertypes.Resources) ([]specs.LinuxWeightDevice, error) {

View file

@ -954,13 +954,11 @@ func WithResources(c *container.Container) coci.SpecOpts {
if err != nil {
return err
}
blkioWeight := r.BlkioWeight
specResources := &specs.LinuxResources{
Memory: memoryRes,
CPU: cpuRes,
BlockIO: &specs.LinuxBlockIO{
Weight: &blkioWeight,
WeightDevice: weightDevices,
ThrottleReadBpsDevice: readBpsDevice,
ThrottleWriteBpsDevice: writeBpsDevice,
@ -969,6 +967,10 @@ func WithResources(c *container.Container) coci.SpecOpts {
},
Pids: getPidsLimit(r),
}
if r.BlkioWeight != 0 {
w := r.BlkioWeight
specResources.BlockIO.Weight = &w
}
if s.Linux.Resources != nil && len(s.Linux.Resources.Devices) > 0 {
specResources.Devices = s.Linux.Resources.Devices

View file

@ -11,6 +11,8 @@ import (
"github.com/docker/docker/daemon/config"
"github.com/docker/docker/daemon/network"
"github.com/docker/docker/libnetwork"
"github.com/google/go-cmp/cmp/cmpopts"
"github.com/opencontainers/runtime-spec/specs-go"
"golang.org/x/sys/unix"
"gotest.tools/v3/assert"
is "gotest.tools/v3/assert/cmp"
@ -214,3 +216,38 @@ func TestGetSourceMount(t *testing.T) {
_, _, err = getSourceMount(cwd)
assert.NilError(t, err)
}
func TestDefaultResources(t *testing.T) {
skip.If(t, os.Getuid() != 0, "skipping test that requires root") // TODO: is this actually true? I'm guilty of following the cargo cult here.
c := &container.Container{
HostConfig: &containertypes.HostConfig{
IpcMode: containertypes.IPCModeNone,
},
}
d := setupFakeDaemon(t, c)
s, err := d.createSpec(context.Background(), c)
assert.NilError(t, err)
checkResourcesAreUnset(t, s.Linux.Resources)
}
func checkResourcesAreUnset(t *testing.T, r *specs.LinuxResources) {
t.Helper()
if r != nil {
if r.Memory != nil {
assert.Check(t, is.DeepEqual(r.Memory, &specs.LinuxMemory{}))
}
if r.CPU != nil {
assert.Check(t, is.DeepEqual(r.CPU, &specs.LinuxCPU{}))
}
assert.Check(t, is.Nil(r.Pids))
if r.BlockIO != nil {
assert.Check(t, is.DeepEqual(r.BlockIO, &specs.LinuxBlockIO{}, cmpopts.EquateEmpty()))
}
if r.Network != nil {
assert.Check(t, is.DeepEqual(r.Network, &specs.LinuxNetwork{}, cmpopts.EquateEmpty()))
}
}
}

View file

@ -11,15 +11,19 @@ import (
func toContainerdResources(resources container.Resources) *libcontainerdtypes.Resources {
var r libcontainerdtypes.Resources
r.BlockIO = &specs.LinuxBlockIO{
Weight: &resources.BlkioWeight,
if resources.BlkioWeight != 0 {
r.BlockIO = &specs.LinuxBlockIO{
Weight: &resources.BlkioWeight,
}
}
shares := uint64(resources.CPUShares)
r.CPU = &specs.LinuxCPU{
Shares: &shares,
Cpus: resources.CpusetCpus,
Mems: resources.CpusetMems,
cpu := specs.LinuxCPU{
Cpus: resources.CpusetCpus,
Mems: resources.CpusetMems,
}
if resources.CPUShares != 0 {
shares := uint64(resources.CPUShares)
cpu.Shares = &shares
}
var (
@ -37,17 +41,33 @@ func toContainerdResources(resources container.Resources) *libcontainerdtypes.Re
period = uint64(resources.CPUPeriod)
}
r.CPU.Period = &period
r.CPU.Quota = &quota
r.Memory = &specs.LinuxMemory{
Limit: &resources.Memory,
Reservation: &resources.MemoryReservation,
Kernel: &resources.KernelMemory,
if period != 0 {
cpu.Period = &period
}
if quota != 0 {
cpu.Quota = &quota
}
if cpu != (specs.LinuxCPU{}) {
r.CPU = &cpu
}
var memory specs.LinuxMemory
if resources.Memory != 0 {
memory.Limit = &resources.Memory
}
if resources.MemoryReservation != 0 {
memory.Reservation = &resources.MemoryReservation
}
if resources.KernelMemory != 0 {
memory.Kernel = &resources.KernelMemory
}
if resources.MemorySwap > 0 {
r.Memory.Swap = &resources.MemorySwap
memory.Swap = &resources.MemorySwap
}
if memory != (specs.LinuxMemory{}) {
r.Memory = &memory
}
r.Pids = getPidsLimit(resources)

View file

@ -0,0 +1,11 @@
package daemon // import "github.com/docker/docker/daemon"
import (
"testing"
"github.com/docker/docker/api/types/container"
)
func TestToContainerdResources_Defaults(t *testing.T) {
checkResourcesAreUnset(t, toContainerdResources(container.Resources{}))
}

View file

@ -21,9 +21,7 @@ func summaryFromInterface(i interface{}) (*libcontainerdtypes.Summary, error) {
}
func (t *task) UpdateResources(ctx context.Context, resources *libcontainerdtypes.Resources) error {
// go doesn't like the alias in 1.8, this means this need to be
// platform specific
return t.Update(ctx, containerd.WithResources((*specs.LinuxResources)(resources)))
return t.Update(ctx, containerd.WithResources(resources))
}
func hostIDFromMap(id uint32, mp []specs.LinuxIDMapping) int {

View file

@ -27,7 +27,7 @@ func InterfaceToStats(read time.Time, v interface{}) *Stats {
}
// Resources defines updatable container resource values. TODO: it must match containerd upcoming API
type Resources specs.LinuxResources
type Resources = specs.LinuxResources
// Checkpoints contains the details of a checkpoint
type Checkpoints struct{}