summaryrefslogtreecommitdiff
path: root/libpod
diff options
context:
space:
mode:
Diffstat (limited to 'libpod')
-rw-r--r--libpod/container_config.go1
-rw-r--r--libpod/container_exec.go21
-rw-r--r--libpod/container_internal_linux.go3
-rw-r--r--libpod/define/volume_inspect.go6
-rw-r--r--libpod/events.go10
-rw-r--r--libpod/events/config.go4
-rw-r--r--libpod/events/events.go4
-rw-r--r--libpod/events/journal_linux.go2
-rw-r--r--libpod/healthcheck.go15
-rw-r--r--libpod/oci_conmon_linux.go201
-rw-r--r--libpod/plugin/volume_api.go18
-rw-r--r--libpod/pod_internal.go2
-rw-r--r--libpod/runtime_ctr.go6
-rw-r--r--libpod/runtime_pod_linux.go32
-rw-r--r--libpod/runtime_volume.go2
-rw-r--r--libpod/runtime_volume_linux.go98
-rw-r--r--libpod/stats.go24
-rw-r--r--libpod/util_linux.go27
18 files changed, 415 insertions, 61 deletions
diff --git a/libpod/container_config.go b/libpod/container_config.go
index 45ff03d58..544c45a8c 100644
--- a/libpod/container_config.go
+++ b/libpod/container_config.go
@@ -424,7 +424,6 @@ type InfraInherit struct {
CapDrop []string `json:"cap_drop,omitempty"`
HostDeviceList []spec.LinuxDevice `json:"host_device_list,omitempty"`
ImageVolumes []*specgen.ImageVolume `json:"image_volumes,omitempty"`
- InfraResources *spec.LinuxResources `json:"resource_limits,omitempty"`
Mounts []spec.Mount `json:"mounts,omitempty"`
NoNewPrivileges bool `json:"no_new_privileges,omitempty"`
OverlayVolumes []*specgen.OverlayVolume `json:"overlay_volumes,omitempty"`
diff --git a/libpod/container_exec.go b/libpod/container_exec.go
index be00c6fbe..b112273d0 100644
--- a/libpod/container_exec.go
+++ b/libpod/container_exec.go
@@ -277,9 +277,13 @@ func (c *Container) ExecStart(sessionID string) error {
return c.save()
}
+func (c *Container) ExecStartAndAttach(sessionID string, streams *define.AttachStreams, newSize *define.TerminalSize) error {
+ return c.execStartAndAttach(sessionID, streams, newSize, false)
+}
+
// ExecStartAndAttach starts and attaches to an exec session in a container.
// newSize resizes the tty to this size before the process is started, must be nil if the exec session has no tty
-func (c *Container) ExecStartAndAttach(sessionID string, streams *define.AttachStreams, newSize *define.TerminalSize) error {
+func (c *Container) execStartAndAttach(sessionID string, streams *define.AttachStreams, newSize *define.TerminalSize, isHealthcheck bool) error {
if !c.batched {
c.lock.Lock()
defer c.lock.Unlock()
@@ -315,7 +319,12 @@ func (c *Container) ExecStartAndAttach(sessionID string, streams *define.AttachS
return err
}
- c.newContainerEvent(events.Exec)
+ if isHealthcheck {
+ c.newContainerEvent(events.HealthStatus)
+ } else {
+ c.newContainerEvent(events.Exec)
+ }
+
logrus.Debugf("Successfully started exec session %s in container %s", session.ID(), c.ID())
var lastErr error
@@ -743,10 +752,14 @@ func (c *Container) ExecResize(sessionID string, newSize define.TerminalSize) er
return c.ociRuntime.ExecAttachResize(c, sessionID, newSize)
}
+func (c *Container) Exec(config *ExecConfig, streams *define.AttachStreams, resize <-chan define.TerminalSize) (int, error) {
+ return c.exec(config, streams, resize, false)
+}
+
// Exec emulates the old Libpod exec API, providing a single call to create,
// run, and remove an exec session. Returns exit code and error. Exit code is
// not guaranteed to be set sanely if error is not nil.
-func (c *Container) Exec(config *ExecConfig, streams *define.AttachStreams, resize <-chan define.TerminalSize) (int, error) {
+func (c *Container) exec(config *ExecConfig, streams *define.AttachStreams, resize <-chan define.TerminalSize, isHealthcheck bool) (int, error) {
sessionID, err := c.ExecCreate(config)
if err != nil {
return -1, err
@@ -780,7 +793,7 @@ func (c *Container) Exec(config *ExecConfig, streams *define.AttachStreams, resi
}()
}
- if err := c.ExecStartAndAttach(sessionID, streams, size); err != nil {
+ if err := c.execStartAndAttach(sessionID, streams, size, isHealthcheck); err != nil {
return -1, err
}
diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go
index 77b598b16..0f4bf0f55 100644
--- a/libpod/container_internal_linux.go
+++ b/libpod/container_internal_linux.go
@@ -870,6 +870,7 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) {
if err != nil {
return nil, err
}
+
g.SetLinuxCgroupsPath(cgroupPath)
// Warning: CDI may alter g.Config in place.
@@ -1141,7 +1142,7 @@ func (c *Container) addCheckpointImageMetadata(importBuilder *buildah.Builder) e
return fmt.Errorf("getting host info: %v", err)
}
- criuVersion, err := criu.GetCriuVestion()
+ criuVersion, err := criu.GetCriuVersion()
if err != nil {
return fmt.Errorf("getting criu version: %v", err)
}
diff --git a/libpod/define/volume_inspect.go b/libpod/define/volume_inspect.go
index fac179176..aaa23b4fc 100644
--- a/libpod/define/volume_inspect.go
+++ b/libpod/define/volume_inspect.go
@@ -57,3 +57,9 @@ type InspectVolumeData struct {
// UID/GID.
NeedsChown bool `json:"NeedsChown,omitempty"`
}
+
+type VolumeReload struct {
+ Added []string
+ Removed []string
+ Errors []error
+}
diff --git a/libpod/events.go b/libpod/events.go
index 021b3b53c..bb50df92d 100644
--- a/libpod/events.go
+++ b/libpod/events.go
@@ -33,6 +33,16 @@ func (c *Container) newContainerEvent(status events.Status) {
Attributes: c.Labels(),
}
+ // if the current event is a HealthStatus event, we need to get the current
+ // status of the container to pass to the event
+ if status == events.HealthStatus {
+ containerHealthStatus, err := c.healthCheckStatus()
+ if err != nil {
+ e.HealthStatus = fmt.Sprintf("%v", err)
+ }
+ e.HealthStatus = containerHealthStatus
+ }
+
if err := c.runtime.eventer.Write(e); err != nil {
logrus.Errorf("Unable to write pod event: %q", err)
}
diff --git a/libpod/events/config.go b/libpod/events/config.go
index 2e7016136..a678baa2d 100644
--- a/libpod/events/config.go
+++ b/libpod/events/config.go
@@ -40,6 +40,8 @@ type Event struct {
Time time.Time
// Type of event that occurred
Type Type
+ // Health status of the current container
+ HealthStatus string `json:"health_status,omitempty"`
Details
}
@@ -141,6 +143,8 @@ const (
Exited Status = "died"
// Export ...
Export Status = "export"
+ // HealthStatus ...
+ HealthStatus Status = "health_status"
// History ...
History Status = "history"
// Import ...
diff --git a/libpod/events/events.go b/libpod/events/events.go
index a30e0f1ca..a8001ab95 100644
--- a/libpod/events/events.go
+++ b/libpod/events/events.go
@@ -76,7 +76,7 @@ func (e *Event) ToHumanReadable(truncate bool) string {
}
switch e.Type {
case Container, Pod:
- humanFormat = fmt.Sprintf("%s %s %s %s (image=%s, name=%s", e.Time, e.Type, e.Status, id, e.Image, e.Name)
+ humanFormat = fmt.Sprintf("%s %s %s %s (image=%s, name=%s, health_status=%s", e.Time, e.Type, e.Status, id, e.Image, e.Name, e.HealthStatus)
// check if the container has labels and add it to the output
if len(e.Attributes) > 0 {
for k, v := range e.Attributes {
@@ -168,6 +168,8 @@ func StringToStatus(name string) (Status, error) {
return Exited, nil
case Export.String():
return Export, nil
+ case HealthStatus.String():
+ return HealthStatus, nil
case History.String():
return History, nil
case Import.String():
diff --git a/libpod/events/journal_linux.go b/libpod/events/journal_linux.go
index d21b60c68..036638d34 100644
--- a/libpod/events/journal_linux.go
+++ b/libpod/events/journal_linux.go
@@ -58,6 +58,7 @@ func (e EventJournalD) Write(ee Event) error {
}
m["PODMAN_LABELS"] = string(b)
}
+ m["PODMAN_HEALTH_STATUS"] = ee.HealthStatus
case Network:
m["PODMAN_ID"] = ee.ID
m["PODMAN_NETWORK_NAME"] = ee.Network
@@ -213,6 +214,7 @@ func newEventFromJournalEntry(entry *sdjournal.JournalEntry) (*Event, error) {
newEvent.Details = Details{Attributes: labels}
}
}
+ newEvent.HealthStatus = entry.Fields["PODMAN_HEALTH_STATUS"]
case Network:
newEvent.ID = entry.Fields["PODMAN_ID"]
newEvent.Network = entry.Fields["PODMAN_NETWORK_NAME"]
diff --git a/libpod/healthcheck.go b/libpod/healthcheck.go
index bd77e98c6..95c70b60e 100644
--- a/libpod/healthcheck.go
+++ b/libpod/healthcheck.go
@@ -90,7 +90,7 @@ func (c *Container) runHealthCheck() (define.HealthCheckStatus, error) {
hcResult := define.HealthCheckSuccess
config := new(ExecConfig)
config.Command = newCommand
- exitCode, hcErr := c.Exec(config, streams, nil)
+ exitCode, hcErr := c.exec(config, streams, nil, true)
if hcErr != nil {
errCause := errors.Cause(hcErr)
hcResult = define.HealthCheckFailure
@@ -232,18 +232,27 @@ func (c *Container) getHealthCheckLog() (define.HealthCheckResults, error) {
// HealthCheckStatus returns the current state of a container with a healthcheck
func (c *Container) HealthCheckStatus() (string, error) {
+ c.lock.Lock()
+ defer c.lock.Unlock()
+ return c.healthCheckStatus()
+}
+
+// Internal function to return the current state of a container with a healthcheck.
+// This function does not lock the container.
+func (c *Container) healthCheckStatus() (string, error) {
if !c.HasHealthCheck() {
return "", errors.Errorf("container %s has no defined healthcheck", c.ID())
}
- c.lock.Lock()
- defer c.lock.Unlock()
+
if err := c.syncContainer(); err != nil {
return "", err
}
+
results, err := c.getHealthCheckLog()
if err != nil {
return "", errors.Wrapf(err, "unable to get healthcheck log for %s", c.ID())
}
+
return results.Status, nil
}
diff --git a/libpod/oci_conmon_linux.go b/libpod/oci_conmon_linux.go
index d417626dc..7a9ae7ee5 100644
--- a/libpod/oci_conmon_linux.go
+++ b/libpod/oci_conmon_linux.go
@@ -23,6 +23,9 @@ import (
"text/template"
"time"
+ runcconfig "github.com/opencontainers/runc/libcontainer/configs"
+ "github.com/opencontainers/runc/libcontainer/devices"
+
"github.com/containers/common/pkg/cgroups"
"github.com/containers/common/pkg/config"
conmonConfig "github.com/containers/conmon/runner/config"
@@ -1433,9 +1436,14 @@ func (r *ConmonOCIRuntime) moveConmonToCgroupAndSignal(ctr *Container, cmd *exec
// TODO: This should be a switch - we are not guaranteed that
// there are only 2 valid cgroup managers
cgroupParent := ctr.CgroupParent()
+ cgroupPath := filepath.Join(ctr.config.CgroupParent, "conmon")
+ Resource := ctr.Spec().Linux.Resources
+ cgroupResources, err := GetLimits(Resource)
+ if err != nil {
+ logrus.StandardLogger().Log(logLevel, "Could not get ctr resources")
+ }
if ctr.CgroupManager() == config.SystemdCgroupsManager {
unitName := createUnitName("libpod-conmon", ctr.ID())
-
realCgroupParent := cgroupParent
splitParent := strings.Split(cgroupParent, "/")
if strings.HasSuffix(cgroupParent, ".slice") && len(splitParent) > 1 {
@@ -1447,8 +1455,7 @@ func (r *ConmonOCIRuntime) moveConmonToCgroupAndSignal(ctr *Container, cmd *exec
logrus.StandardLogger().Logf(logLevel, "Failed to add conmon to systemd sandbox cgroup: %v", err)
}
} else {
- cgroupPath := filepath.Join(ctr.config.CgroupParent, "conmon")
- control, err := cgroups.New(cgroupPath, &spec.LinuxResources{})
+ control, err := cgroups.New(cgroupPath, &cgroupResources)
if err != nil {
logrus.StandardLogger().Logf(logLevel, "Failed to add conmon to cgroupfs sandbox cgroup: %v", err)
} else if err := control.AddPid(cmd.Process.Pid); err != nil {
@@ -1730,3 +1737,191 @@ func httpAttachNonTerminalCopy(container *net.UnixConn, http *bufio.ReadWriter,
}
}
}
+
+// GetLimits converts spec resource limits to cgroup consumable limits
+func GetLimits(resource *spec.LinuxResources) (runcconfig.Resources, error) {
+ if resource == nil {
+ resource = &spec.LinuxResources{}
+ }
+ final := &runcconfig.Resources{}
+ devs := []*devices.Rule{}
+
+ // Devices
+ for _, entry := range resource.Devices {
+ if entry.Major == nil || entry.Minor == nil {
+ continue
+ }
+ runeType := 'a'
+ switch entry.Type {
+ case "b":
+ runeType = 'b'
+ case "c":
+ runeType = 'c'
+ }
+
+ devs = append(devs, &devices.Rule{
+ Type: devices.Type(runeType),
+ Major: *entry.Major,
+ Minor: *entry.Minor,
+ Permissions: devices.Permissions(entry.Access),
+ Allow: entry.Allow,
+ })
+ }
+ final.Devices = devs
+
+ // HugepageLimits
+ pageLimits := []*runcconfig.HugepageLimit{}
+ for _, entry := range resource.HugepageLimits {
+ pageLimits = append(pageLimits, &runcconfig.HugepageLimit{
+ Pagesize: entry.Pagesize,
+ Limit: entry.Limit,
+ })
+ }
+ final.HugetlbLimit = pageLimits
+
+ // Networking
+ netPriorities := []*runcconfig.IfPrioMap{}
+ if resource.Network != nil {
+ for _, entry := range resource.Network.Priorities {
+ netPriorities = append(netPriorities, &runcconfig.IfPrioMap{
+ Interface: entry.Name,
+ Priority: int64(entry.Priority),
+ })
+ }
+ }
+ final.NetPrioIfpriomap = netPriorities
+ rdma := make(map[string]runcconfig.LinuxRdma)
+ for name, entry := range resource.Rdma {
+ rdma[name] = runcconfig.LinuxRdma{HcaHandles: entry.HcaHandles, HcaObjects: entry.HcaObjects}
+ }
+ final.Rdma = rdma
+
+ // Memory
+ if resource.Memory != nil {
+ if resource.Memory.Limit != nil {
+ final.Memory = *resource.Memory.Limit
+ }
+ if resource.Memory.Reservation != nil {
+ final.MemoryReservation = *resource.Memory.Reservation
+ }
+ if resource.Memory.Swap != nil {
+ final.MemorySwap = *resource.Memory.Swap
+ }
+ if resource.Memory.Swappiness != nil {
+ final.MemorySwappiness = resource.Memory.Swappiness
+ }
+ }
+
+ // CPU
+ if resource.CPU != nil {
+ if resource.CPU.Period != nil {
+ final.CpuPeriod = *resource.CPU.Period
+ }
+ if resource.CPU.Quota != nil {
+ final.CpuQuota = *resource.CPU.Quota
+ }
+ if resource.CPU.RealtimePeriod != nil {
+ final.CpuRtPeriod = *resource.CPU.RealtimePeriod
+ }
+ if resource.CPU.RealtimeRuntime != nil {
+ final.CpuRtRuntime = *resource.CPU.RealtimeRuntime
+ }
+ if resource.CPU.Shares != nil {
+ final.CpuShares = *resource.CPU.Shares
+ }
+ final.CpusetCpus = resource.CPU.Cpus
+ final.CpusetMems = resource.CPU.Mems
+ }
+
+ // BlkIO
+ if resource.BlockIO != nil {
+ if len(resource.BlockIO.ThrottleReadBpsDevice) > 0 {
+ for _, entry := range resource.BlockIO.ThrottleReadBpsDevice {
+ throttle := &runcconfig.ThrottleDevice{}
+ dev := &runcconfig.BlockIODevice{
+ Major: entry.Major,
+ Minor: entry.Minor,
+ }
+ throttle.BlockIODevice = *dev
+ throttle.Rate = entry.Rate
+ final.BlkioThrottleReadBpsDevice = append(final.BlkioThrottleReadBpsDevice, throttle)
+ }
+ }
+ if len(resource.BlockIO.ThrottleWriteBpsDevice) > 0 {
+ for _, entry := range resource.BlockIO.ThrottleWriteBpsDevice {
+ throttle := &runcconfig.ThrottleDevice{}
+ dev := &runcconfig.BlockIODevice{
+ Major: entry.Major,
+ Minor: entry.Minor,
+ }
+ throttle.BlockIODevice = *dev
+ throttle.Rate = entry.Rate
+ final.BlkioThrottleWriteBpsDevice = append(final.BlkioThrottleWriteBpsDevice, throttle)
+ }
+ }
+ if len(resource.BlockIO.ThrottleReadIOPSDevice) > 0 {
+ for _, entry := range resource.BlockIO.ThrottleReadIOPSDevice {
+ throttle := &runcconfig.ThrottleDevice{}
+ dev := &runcconfig.BlockIODevice{
+ Major: entry.Major,
+ Minor: entry.Minor,
+ }
+ throttle.BlockIODevice = *dev
+ throttle.Rate = entry.Rate
+ final.BlkioThrottleReadIOPSDevice = append(final.BlkioThrottleReadIOPSDevice, throttle)
+ }
+ }
+ if len(resource.BlockIO.ThrottleWriteIOPSDevice) > 0 {
+ for _, entry := range resource.BlockIO.ThrottleWriteIOPSDevice {
+ throttle := &runcconfig.ThrottleDevice{}
+ dev := &runcconfig.BlockIODevice{
+ Major: entry.Major,
+ Minor: entry.Minor,
+ }
+ throttle.BlockIODevice = *dev
+ throttle.Rate = entry.Rate
+ final.BlkioThrottleWriteIOPSDevice = append(final.BlkioThrottleWriteIOPSDevice, throttle)
+ }
+ }
+ if resource.BlockIO.LeafWeight != nil {
+ final.BlkioLeafWeight = *resource.BlockIO.LeafWeight
+ }
+ if resource.BlockIO.Weight != nil {
+ final.BlkioWeight = *resource.BlockIO.Weight
+ }
+ if len(resource.BlockIO.WeightDevice) > 0 {
+ for _, entry := range resource.BlockIO.WeightDevice {
+ weight := &runcconfig.WeightDevice{}
+ dev := &runcconfig.BlockIODevice{
+ Major: entry.Major,
+ Minor: entry.Minor,
+ }
+ if entry.Weight != nil {
+ weight.Weight = *entry.Weight
+ }
+ if entry.LeafWeight != nil {
+ weight.LeafWeight = *entry.LeafWeight
+ }
+ weight.BlockIODevice = *dev
+ final.BlkioWeightDevice = append(final.BlkioWeightDevice, weight)
+ }
+ }
+ }
+
+ // Pids
+ if resource.Pids != nil {
+ final.PidsLimit = resource.Pids.Limit
+ }
+
+ // Networking
+ if resource.Network != nil {
+ if resource.Network.ClassID != nil {
+ final.NetClsClassid = *resource.Network.ClassID
+ }
+ }
+
+ // Unified state
+ final.Unified = resource.Unified
+
+ return *final, nil
+}
diff --git a/libpod/plugin/volume_api.go b/libpod/plugin/volume_api.go
index f997ccf22..2de7db32c 100644
--- a/libpod/plugin/volume_api.go
+++ b/libpod/plugin/volume_api.go
@@ -197,13 +197,13 @@ func (p *VolumePlugin) verifyReachable() error {
// Send a request to the volume plugin for handling.
// Callers *MUST* close the response when they are done.
-func (p *VolumePlugin) sendRequest(toJSON interface{}, hasBody bool, endpoint string) (*http.Response, error) {
+func (p *VolumePlugin) sendRequest(toJSON interface{}, endpoint string) (*http.Response, error) {
var (
reqJSON []byte
err error
)
- if hasBody {
+ if toJSON != nil {
reqJSON, err = json.Marshal(toJSON)
if err != nil {
return nil, errors.Wrapf(err, "error marshalling request JSON for volume plugin %s endpoint %s", p.Name, endpoint)
@@ -274,7 +274,7 @@ func (p *VolumePlugin) CreateVolume(req *volume.CreateRequest) error {
logrus.Infof("Creating volume %s using plugin %s", req.Name, p.Name)
- resp, err := p.sendRequest(req, true, createPath)
+ resp, err := p.sendRequest(req, createPath)
if err != nil {
return err
}
@@ -291,7 +291,7 @@ func (p *VolumePlugin) ListVolumes() ([]*volume.Volume, error) {
logrus.Infof("Listing volumes using plugin %s", p.Name)
- resp, err := p.sendRequest(nil, false, listPath)
+ resp, err := p.sendRequest(nil, listPath)
if err != nil {
return nil, err
}
@@ -326,7 +326,7 @@ func (p *VolumePlugin) GetVolume(req *volume.GetRequest) (*volume.Volume, error)
logrus.Infof("Getting volume %s using plugin %s", req.Name, p.Name)
- resp, err := p.sendRequest(req, true, getPath)
+ resp, err := p.sendRequest(req, getPath)
if err != nil {
return nil, err
}
@@ -361,7 +361,7 @@ func (p *VolumePlugin) RemoveVolume(req *volume.RemoveRequest) error {
logrus.Infof("Removing volume %s using plugin %s", req.Name, p.Name)
- resp, err := p.sendRequest(req, true, removePath)
+ resp, err := p.sendRequest(req, removePath)
if err != nil {
return err
}
@@ -382,7 +382,7 @@ func (p *VolumePlugin) GetVolumePath(req *volume.PathRequest) (string, error) {
logrus.Infof("Getting volume %s path using plugin %s", req.Name, p.Name)
- resp, err := p.sendRequest(req, true, hostVirtualPath)
+ resp, err := p.sendRequest(req, hostVirtualPath)
if err != nil {
return "", err
}
@@ -419,7 +419,7 @@ func (p *VolumePlugin) MountVolume(req *volume.MountRequest) (string, error) {
logrus.Infof("Mounting volume %s using plugin %s for container %s", req.Name, p.Name, req.ID)
- resp, err := p.sendRequest(req, true, mountPath)
+ resp, err := p.sendRequest(req, mountPath)
if err != nil {
return "", err
}
@@ -455,7 +455,7 @@ func (p *VolumePlugin) UnmountVolume(req *volume.UnmountRequest) error {
logrus.Infof("Unmounting volume %s using plugin %s for container %s", req.Name, p.Name, req.ID)
- resp, err := p.sendRequest(req, true, unmountPath)
+ resp, err := p.sendRequest(req, unmountPath)
if err != nil {
return err
}
diff --git a/libpod/pod_internal.go b/libpod/pod_internal.go
index 41f745e6c..1502bcb06 100644
--- a/libpod/pod_internal.go
+++ b/libpod/pod_internal.go
@@ -69,7 +69,7 @@ func (p *Pod) refresh() error {
if p.config.UsePodCgroup {
switch p.runtime.config.Engine.CgroupManager {
case config.SystemdCgroupsManager:
- cgroupPath, err := systemdSliceFromPath(p.config.CgroupParent, fmt.Sprintf("libpod_pod_%s", p.ID()))
+ cgroupPath, err := systemdSliceFromPath(p.config.CgroupParent, fmt.Sprintf("libpod_pod_%s", p.ID()), p.ResourceLim())
if err != nil {
logrus.Errorf("Creating Cgroup for pod %s: %v", p.ID(), err)
}
diff --git a/libpod/runtime_ctr.go b/libpod/runtime_ctr.go
index a9ae9d1db..459514d47 100644
--- a/libpod/runtime_ctr.go
+++ b/libpod/runtime_ctr.go
@@ -502,7 +502,7 @@ func (r *Runtime) setupContainer(ctx context.Context, ctr *Container) (_ *Contai
volOptions = append(volOptions, parsedOptions...)
}
}
- newVol, err := r.newVolume(volOptions...)
+ newVol, err := r.newVolume(false, volOptions...)
if err != nil {
return nil, errors.Wrapf(err, "error creating named volume %q", vol.Name)
}
@@ -805,7 +805,7 @@ func (r *Runtime) removeContainer(ctx context.Context, c *Container, force, remo
if !volume.Anonymous() {
continue
}
- if err := runtime.removeVolume(ctx, volume, false, timeout); err != nil && errors.Cause(err) != define.ErrNoSuchVolume {
+ if err := runtime.removeVolume(ctx, volume, false, timeout, false); err != nil && errors.Cause(err) != define.ErrNoSuchVolume {
if errors.Cause(err) == define.ErrVolumeBeingUsed {
// Ignore error, since podman will report original error
volumesFrom, _ := c.volumesFrom()
@@ -963,7 +963,7 @@ func (r *Runtime) evictContainer(ctx context.Context, idOrName string, removeVol
if !volume.Anonymous() {
continue
}
- if err := r.removeVolume(ctx, volume, false, timeout); err != nil && err != define.ErrNoSuchVolume && err != define.ErrVolumeBeingUsed {
+ if err := r.removeVolume(ctx, volume, false, timeout, false); err != nil && err != define.ErrNoSuchVolume && err != define.ErrVolumeBeingUsed {
logrus.Errorf("Cleaning up volume (%s): %v", v, err)
}
}
diff --git a/libpod/runtime_pod_linux.go b/libpod/runtime_pod_linux.go
index dcc3a044f..00017ca21 100644
--- a/libpod/runtime_pod_linux.go
+++ b/libpod/runtime_pod_linux.go
@@ -17,7 +17,7 @@ import (
"github.com/containers/podman/v4/libpod/events"
"github.com/containers/podman/v4/pkg/rootless"
"github.com/containers/podman/v4/pkg/specgen"
- spec "github.com/opencontainers/runtime-spec/specs-go"
+ runcconfig "github.com/opencontainers/runc/libcontainer/configs"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
)
@@ -66,6 +66,7 @@ func (r *Runtime) NewPod(ctx context.Context, p specgen.PodSpecGenerator, option
case config.CgroupfsCgroupsManager:
canUseCgroup := !rootless.IsRootless() || isRootlessCgroupSet(pod.config.CgroupParent)
if canUseCgroup {
+ // need to actually create parent here
if pod.config.CgroupParent == "" {
pod.config.CgroupParent = CgroupfsDefaultCgroupParent
} else if strings.HasSuffix(path.Base(pod.config.CgroupParent), ".slice") {
@@ -73,12 +74,26 @@ func (r *Runtime) NewPod(ctx context.Context, p specgen.PodSpecGenerator, option
}
// If we are set to use pod cgroups, set the cgroup parent that
// all containers in the pod will share
- // No need to create it with cgroupfs - the first container to
- // launch should do it for us
if pod.config.UsePodCgroup {
pod.state.CgroupPath = filepath.Join(pod.config.CgroupParent, pod.ID())
if p.InfraContainerSpec != nil {
p.InfraContainerSpec.CgroupParent = pod.state.CgroupPath
+ res, err := GetLimits(p.InfraContainerSpec.ResourceLimits)
+ if err != nil {
+ return nil, err
+ }
+ // Need to both create and update the cgroup
+ // rather than create a new path in c/common for pod cgroup creation
+ // just create as if it is a ctr and then update figures out that we need to
+ // populate the resource limits on the pod level
+ cgc, err := cgroups.New(pod.state.CgroupPath, &res)
+ if err != nil {
+ return nil, err
+ }
+ err = cgc.Update(&res)
+ if err != nil {
+ return nil, err
+ }
}
}
}
@@ -95,7 +110,7 @@ func (r *Runtime) NewPod(ctx context.Context, p specgen.PodSpecGenerator, option
// If we are set to use pod cgroups, set the cgroup parent that
// all containers in the pod will share
if pod.config.UsePodCgroup {
- cgroupPath, err := systemdSliceFromPath(pod.config.CgroupParent, fmt.Sprintf("libpod_pod_%s", pod.ID()))
+ cgroupPath, err := systemdSliceFromPath(pod.config.CgroupParent, fmt.Sprintf("libpod_pod_%s", pod.ID()), p.InfraContainerSpec.ResourceLimits)
if err != nil {
return nil, errors.Wrapf(err, "unable to create pod cgroup for pod %s", pod.ID())
}
@@ -239,9 +254,8 @@ func (r *Runtime) removePod(ctx context.Context, p *Pod, removeCtrs, force bool,
}
// New resource limits
- resLimits := new(spec.LinuxResources)
- resLimits.Pids = new(spec.LinuxPids)
- resLimits.Pids.Limit = 1 // Inhibit forks with very low pids limit
+ resLimits := new(runcconfig.Resources)
+ resLimits.PidsLimit = 1 // Inhibit forks with very low pids limit
// Don't try if we failed to retrieve the cgroup
if err == nil {
@@ -301,7 +315,7 @@ func (r *Runtime) removePod(ctx context.Context, p *Pod, removeCtrs, force bool,
if !volume.Anonymous() {
continue
}
- if err := r.removeVolume(ctx, volume, false, timeout); err != nil {
+ if err := r.removeVolume(ctx, volume, false, timeout, false); err != nil {
if errors.Cause(err) == define.ErrNoSuchVolume || errors.Cause(err) == define.ErrVolumeRemoved {
continue
}
@@ -321,7 +335,7 @@ func (r *Runtime) removePod(ctx context.Context, p *Pod, removeCtrs, force bool,
switch p.runtime.config.Engine.CgroupManager {
case config.SystemdCgroupsManager:
- if err := deleteSystemdCgroup(p.state.CgroupPath); err != nil {
+ if err := deleteSystemdCgroup(p.state.CgroupPath, p.ResourceLim()); err != nil {
if removalErr == nil {
removalErr = errors.Wrapf(err, "error removing pod %s cgroup", p.ID())
} else {
diff --git a/libpod/runtime_volume.go b/libpod/runtime_volume.go
index 21bf8aefc..6872db21d 100644
--- a/libpod/runtime_volume.go
+++ b/libpod/runtime_volume.go
@@ -33,7 +33,7 @@ func (r *Runtime) RemoveVolume(ctx context.Context, v *Volume, force bool, timeo
return nil
}
}
- return r.removeVolume(ctx, v, force, timeout)
+ return r.removeVolume(ctx, v, force, timeout, false)
}
// GetVolume retrieves a volume given its full name.
diff --git a/libpod/runtime_volume_linux.go b/libpod/runtime_volume_linux.go
index f8788e183..877f3a1fd 100644
--- a/libpod/runtime_volume_linux.go
+++ b/libpod/runtime_volume_linux.go
@@ -5,6 +5,7 @@ package libpod
import (
"context"
+ "fmt"
"os"
"path/filepath"
"strings"
@@ -25,11 +26,13 @@ func (r *Runtime) NewVolume(ctx context.Context, options ...VolumeCreateOption)
if !r.valid {
return nil, define.ErrRuntimeStopped
}
- return r.newVolume(options...)
+ return r.newVolume(false, options...)
}
-// newVolume creates a new empty volume
-func (r *Runtime) newVolume(options ...VolumeCreateOption) (_ *Volume, deferredErr error) {
+// newVolume creates a new empty volume with the given options.
+// The createPluginVolume can be set to true to make it not create the volume in the volume plugin,
+// this is required for the UpdateVolumePlugins() function. If you are not sure set this to false.
+func (r *Runtime) newVolume(noCreatePluginVolume bool, options ...VolumeCreateOption) (_ *Volume, deferredErr error) {
volume := newVolume(r)
for _, option := range options {
if err := option(volume); err != nil {
@@ -83,7 +86,7 @@ func (r *Runtime) newVolume(options ...VolumeCreateOption) (_ *Volume, deferredE
// Now we get conditional: we either need to make the volume in the
// volume plugin, or on disk if not using a plugin.
- if volume.plugin != nil {
+ if volume.plugin != nil && !noCreatePluginVolume {
// We can't chown, or relabel, or similar the path the volume is
// using, because it's not managed by us.
// TODO: reevaluate this once we actually have volume plugins in
@@ -164,6 +167,85 @@ func (r *Runtime) newVolume(options ...VolumeCreateOption) (_ *Volume, deferredE
return volume, nil
}
+// UpdateVolumePlugins reads all volumes from all configured volume plugins and
+// imports them into the libpod db. It also checks if existing libpod volumes
+// are removed in the plugin, in this case we try to remove it from libpod.
+// On errors we continue and try to do as much as possible. all errors are
+// returned as array in the returned struct.
+// This function has many race conditions, it is best effort but cannot guarantee
+// a perfect state since plugins can be modified from the outside at any time.
+func (r *Runtime) UpdateVolumePlugins(ctx context.Context) *define.VolumeReload {
+ var (
+ added []string
+ removed []string
+ errs []error
+ allPluginVolumes = map[string]struct{}{}
+ )
+
+ for driverName, socket := range r.config.Engine.VolumePlugins {
+ driver, err := volplugin.GetVolumePlugin(driverName, socket)
+ if err != nil {
+ errs = append(errs, err)
+ continue
+ }
+ vols, err := driver.ListVolumes()
+ if err != nil {
+ errs = append(errs, fmt.Errorf("failed to read volumes from plugin %q: %w", driverName, err))
+ continue
+ }
+ for _, vol := range vols {
+ allPluginVolumes[vol.Name] = struct{}{}
+ if _, err := r.newVolume(true, WithVolumeName(vol.Name), WithVolumeDriver(driverName)); err != nil {
+ // If the volume exists this is not an error, just ignore it and log. It is very likely
+ // that the volume from the plugin was already in our db.
+ if !errors.Is(err, define.ErrVolumeExists) {
+ errs = append(errs, err)
+ continue
+ }
+ logrus.Infof("Volume %q already exists: %v", vol.Name, err)
+ continue
+ }
+ added = append(added, vol.Name)
+ }
+ }
+
+ libpodVolumes, err := r.state.AllVolumes()
+ if err != nil {
+ errs = append(errs, fmt.Errorf("cannot delete dangling plugin volumes: failed to read libpod volumes: %w", err))
+ }
+ for _, vol := range libpodVolumes {
+ if vol.UsesVolumeDriver() {
+ if _, ok := allPluginVolumes[vol.Name()]; !ok {
+ // The volume is no longer in the plugin, lets remove it from the libpod db.
+ if err := r.removeVolume(ctx, vol, false, nil, true); err != nil {
+ if errors.Is(err, define.ErrVolumeBeingUsed) {
+ // Volume is still used by at least one container. This is very bad,
+ // the plugin no longer has this but we still need it.
+ errs = append(errs, fmt.Errorf("volume was removed from the plugin %q but containers still require it: %w", vol.config.Driver, err))
+ continue
+ }
+ if errors.Is(err, define.ErrNoSuchVolume) || errors.Is(err, define.ErrVolumeRemoved) || errors.Is(err, define.ErrMissingPlugin) {
+ // Volume was already removed, no problem just ignore it and continue.
+ continue
+ }
+
+ // some other error
+ errs = append(errs, err)
+ continue
+ }
+ // Volume was successfully removed
+ removed = append(removed, vol.Name())
+ }
+ }
+ }
+
+ return &define.VolumeReload{
+ Added: added,
+ Removed: removed,
+ Errors: errs,
+ }
+}
+
// makeVolumeInPluginIfNotExist makes a volume in the given volume plugin if it
// does not already exist.
func makeVolumeInPluginIfNotExist(name string, options map[string]string, plugin *volplugin.VolumePlugin) error {
@@ -197,8 +279,10 @@ func makeVolumeInPluginIfNotExist(name string, options map[string]string, plugin
return nil
}
-// removeVolume removes the specified volume from state as well tears down its mountpoint and storage
-func (r *Runtime) removeVolume(ctx context.Context, v *Volume, force bool, timeout *uint) error {
+// removeVolume removes the specified volume from state as well tears down its mountpoint and storage.
+// ignoreVolumePlugin is used to only remove the volume from the db and not the plugin,
+// this is required when the volume was already removed from the plugin, i.e. in UpdateVolumePlugins().
+func (r *Runtime) removeVolume(ctx context.Context, v *Volume, force bool, timeout *uint, ignoreVolumePlugin bool) error {
if !v.valid {
if ok, _ := r.state.HasVolume(v.Name()); !ok {
return nil
@@ -263,7 +347,7 @@ func (r *Runtime) removeVolume(ctx context.Context, v *Volume, force bool, timeo
var removalErr error
// If we use a volume plugin, we need to remove from the plugin.
- if v.UsesVolumeDriver() {
+ if v.UsesVolumeDriver() && !ignoreVolumePlugin {
canRemove := true
// Do we have a volume driver?
diff --git a/libpod/stats.go b/libpod/stats.go
index d2ffc3b32..eaac9d7d0 100644
--- a/libpod/stats.go
+++ b/libpod/stats.go
@@ -9,6 +9,8 @@ import (
"syscall"
"time"
+ runccgroup "github.com/opencontainers/runc/libcontainer/cgroups"
+
"github.com/containers/common/pkg/cgroups"
"github.com/containers/podman/v4/libpod/define"
"github.com/pkg/errors"
@@ -69,29 +71,29 @@ func (c *Container) GetContainerStats(previousStats *define.ContainerStats) (*de
// If the current total usage in the cgroup is less than what was previously
// recorded then it means the container was restarted and runs in a new cgroup
- if previousStats.Duration > cgroupStats.CPU.Usage.Total {
+ if previousStats.Duration > cgroupStats.CpuStats.CpuUsage.TotalUsage {
previousStats = &define.ContainerStats{}
}
previousCPU := previousStats.CPUNano
now := uint64(time.Now().UnixNano())
- stats.Duration = cgroupStats.CPU.Usage.Total
+ stats.Duration = cgroupStats.CpuStats.CpuUsage.TotalUsage
stats.UpTime = time.Duration(stats.Duration)
stats.CPU = calculateCPUPercent(cgroupStats, previousCPU, now, previousStats.SystemNano)
// calc the average cpu usage for the time the container is running
stats.AvgCPU = calculateCPUPercent(cgroupStats, 0, now, uint64(c.state.StartedTime.UnixNano()))
- stats.MemUsage = cgroupStats.Memory.Usage.Usage
+ stats.MemUsage = cgroupStats.MemoryStats.Usage.Usage
stats.MemLimit = c.getMemLimit()
stats.MemPerc = (float64(stats.MemUsage) / float64(stats.MemLimit)) * 100
stats.PIDs = 0
if conState == define.ContainerStateRunning || conState == define.ContainerStatePaused {
- stats.PIDs = cgroupStats.Pids.Current
+ stats.PIDs = cgroupStats.PidsStats.Current
}
stats.BlockInput, stats.BlockOutput = calculateBlockIO(cgroupStats)
- stats.CPUNano = cgroupStats.CPU.Usage.Total
- stats.CPUSystemNano = cgroupStats.CPU.Usage.Kernel
+ stats.CPUNano = cgroupStats.CpuStats.CpuUsage.TotalUsage
+ stats.CPUSystemNano = cgroupStats.CpuStats.CpuUsage.UsageInKernelmode
stats.SystemNano = now
- stats.PerCPU = cgroupStats.CPU.Usage.PerCPU
+ stats.PerCPU = cgroupStats.CpuStats.CpuUsage.PercpuUsage
// Handle case where the container is not in a network namespace
if netStats != nil {
stats.NetInput = netStats.TxBytes
@@ -133,10 +135,10 @@ func (c *Container) getMemLimit() uint64 {
// previousCPU is the last value of stats.CPU.Usage.Total measured at the time previousSystem.
// (now - previousSystem) is the time delta in nanoseconds, between the measurement in previousCPU
// and the updated value in stats.
-func calculateCPUPercent(stats *cgroups.Metrics, previousCPU, now, previousSystem uint64) float64 {
+func calculateCPUPercent(stats *runccgroup.Stats, previousCPU, now, previousSystem uint64) float64 {
var (
cpuPercent = 0.0
- cpuDelta = float64(stats.CPU.Usage.Total - previousCPU)
+ cpuDelta = float64(stats.CpuStats.CpuUsage.TotalUsage - previousCPU)
systemDelta = float64(now - previousSystem)
)
if systemDelta > 0.0 && cpuDelta > 0.0 {
@@ -146,8 +148,8 @@ func calculateCPUPercent(stats *cgroups.Metrics, previousCPU, now, previousSyste
return cpuPercent
}
-func calculateBlockIO(stats *cgroups.Metrics) (read uint64, write uint64) {
- for _, blkIOEntry := range stats.Blkio.IoServiceBytesRecursive {
+func calculateBlockIO(stats *runccgroup.Stats) (read uint64, write uint64) {
+ for _, blkIOEntry := range stats.BlkioStats.IoServiceBytesRecursive {
switch strings.ToLower(blkIOEntry.Op) {
case "read":
read += blkIOEntry.Value
diff --git a/libpod/util_linux.go b/libpod/util_linux.go
index fe98056dc..414d1bff9 100644
--- a/libpod/util_linux.go
+++ b/libpod/util_linux.go
@@ -11,6 +11,7 @@ import (
"github.com/containers/common/pkg/cgroups"
"github.com/containers/podman/v4/libpod/define"
"github.com/containers/podman/v4/pkg/rootless"
+ spec "github.com/opencontainers/runtime-spec/specs-go"
"github.com/opencontainers/selinux/go-selinux/label"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
@@ -20,7 +21,7 @@ import (
// systemdSliceFromPath makes a new systemd slice under the given parent with
// the given name.
// The parent must be a slice. The name must NOT include ".slice"
-func systemdSliceFromPath(parent, name string) (string, error) {
+func systemdSliceFromPath(parent, name string, resources *spec.LinuxResources) (string, error) {
cgroupPath, err := assembleSystemdCgroupName(parent, name)
if err != nil {
return "", err
@@ -28,7 +29,7 @@ func systemdSliceFromPath(parent, name string) (string, error) {
logrus.Debugf("Created cgroup path %s for parent %s and name %s", cgroupPath, parent, name)
- if err := makeSystemdCgroup(cgroupPath); err != nil {
+ if err := makeSystemdCgroup(cgroupPath, resources); err != nil {
return "", errors.Wrapf(err, "error creating cgroup %s", cgroupPath)
}
@@ -45,8 +46,12 @@ func getDefaultSystemdCgroup() string {
}
// makeSystemdCgroup creates a systemd Cgroup at the given location.
-func makeSystemdCgroup(path string) error {
- controller, err := cgroups.NewSystemd(getDefaultSystemdCgroup())
+func makeSystemdCgroup(path string, resources *spec.LinuxResources) error {
+ res, err := GetLimits(resources)
+ if err != nil {
+ return err
+ }
+ controller, err := cgroups.NewSystemd(getDefaultSystemdCgroup(), &res)
if err != nil {
return err
}
@@ -54,12 +59,20 @@ func makeSystemdCgroup(path string) error {
if rootless.IsRootless() {
return controller.CreateSystemdUserUnit(path, rootless.GetRootlessUID())
}
- return controller.CreateSystemdUnit(path)
+ err = controller.CreateSystemdUnit(path)
+ if err != nil {
+ return err
+ }
+ return nil
}
// deleteSystemdCgroup deletes the systemd cgroup at the given location
-func deleteSystemdCgroup(path string) error {
- controller, err := cgroups.NewSystemd(getDefaultSystemdCgroup())
+func deleteSystemdCgroup(path string, resources *spec.LinuxResources) error {
+ res, err := GetLimits(resources)
+ if err != nil {
+ return err
+ }
+ controller, err := cgroups.NewSystemd(getDefaultSystemdCgroup(), &res)
if err != nil {
return err
}
generated by cgit v1.2.3-54-g00ecf (git 2.39.0) at 2025-01-16 03:47:55 +0000