Merge pull request #1807 from bensmrs/instance-state-cpu-time

Add allocated CPU time to instance state

Author: stgraber

doc/api-extensions.md

@@ -2748,3 +2748,7 @@ Adds `acme.http.port` to control an alternative HTTP port for `HTTP-01` validati
 ## `network_ovn_ipv4_dhcp_expiry`
 
 Introduces `ipv4.dhcp.expiry` for OVN networks.
+
+## `instance_state_cpu_time`
+
+This adds an `allocated_time` field below `CPU` in the instance state API.

doc/rest-api.yaml

@@ -2306,6 +2306,12 @@ definitions:
         x-go-package: github.com/lxc/incus/v6/shared/api
     InstanceStateCPU:
         properties:
+            allocated_time:
+                description: CPU time available per second, in nanoseconds
+                example: 4000000000
+                format: int64
+                type: integer
+                x-go-name: AllocatedTime
             usage:
                 description: CPU usage in nanoseconds
                 example: 3637691016

internal/server/cgroup/abstraction.go

@@ -749,6 +749,65 @@ func (cg *CGroup) SetCPUCfsLimit(limitPeriod int64, limitQuota int64) error {
 	return ErrUnknownVersion
 }
 
+// GetCPUCfsLimit gets the quota and duration in ms for each scheduling period.
+func (cg *CGroup) GetCPUCfsLimit() (int64, int64, error) {
+	version := cgControllers["cpu"]
+	switch version {
+	case Unavailable:
+		return -1, -1, ErrControllerMissing
+	case V1:
+		limitQuotaStr, err := cg.rw.Get(version, "cpu", "cpu.cfs_quota_us")
+		if err != nil {
+			return -1, -1, err
+		}
+
+		limitQuota, err := strconv.ParseInt(limitQuotaStr, 10, 64)
+		if err != nil {
+			return -1, -1, err
+		}
+
+		limitPeriodStr, err := cg.rw.Get(version, "cpu", "cpu.cfs_period_us")
+		if err != nil {
+			return -1, -1, err
+		}
+
+		limitPeriod, err := strconv.ParseInt(limitPeriodStr, 10, 64)
+		if err != nil {
+			return -1, -1, err
+		}
+
+		return limitPeriod, limitQuota, nil
+	case V2:
+		cpuMax, err := cg.rw.Get(version, "cpu", "cpu.max")
+		if err != nil {
+			return -1, -1, err
+		}
+
+		cpuMaxFields := strings.Split(cpuMax, " ")
+		if len(cpuMaxFields) != 2 {
+			return -1, -1, errors.New("Couldn't parse CFS limits")
+		}
+
+		if cpuMaxFields[0] == "max" {
+			return -1, -1, nil
+		}
+
+		limitQuota, err := strconv.ParseInt(cpuMaxFields[0], 10, 64)
+		if err != nil {
+			return -1, -1, err
+		}
+
+		limitPeriod, err := strconv.ParseInt(cpuMaxFields[1], 10, 64)
+		if err != nil {
+			return -1, -1, err
+		}
+
+		return limitPeriod, limitQuota, nil
+	}
+
+	return -1, -1, ErrUnknownVersion
+}
+
 // SetHugepagesLimit applies a limit to the number of processes.
 func (cg *CGroup) SetHugepagesLimit(pageType string, limit int64) error {
 	version := cgControllers["hugetlb"]

internal/server/instance/drivers/driver_lxc.go

@@ -7521,6 +7521,19 @@ func (d *lxc) Exec(req api.InstanceExecPost, stdin *os.File, stdout *os.File, st
 	return instCmd, nil
 }
 
+func (d *lxc) cpuStateUsage(cg *cgroup.CGroup) (int64, bool) {
+	if !d.state.OS.CGInfo.Supports(cgroup.CPUAcct, cg) {
+		return -1, false
+	}
+
+	value, err := cg.GetCPUAcctUsage()
+	if err != nil {
+		return -1, true
+	}
+
+	return value, true
+}
+
 func (d *lxc) cpuState() api.InstanceStateCPU {
 	cpu := api.InstanceStateCPU{}
 
@@ -7529,23 +7542,31 @@ func (d *lxc) cpuState() api.InstanceStateCPU {
 		return cpu
 	}
 
-	// CPU usage in seconds
 	cg, err := d.cgroup(cc, true)
 	if err != nil {
 		return cpu
 	}
 
-	if !d.state.OS.CGInfo.Supports(cgroup.CPUAcct, cg) {
+	cpuUsage, ok := d.cpuStateUsage(cg)
+	if ok {
+		cpu.Usage = cpuUsage
+	}
+
+	cpuCount, err := cg.GetEffectiveCPUs()
+	if err != nil {
 		return cpu
 	}
 
-	value, err := cg.GetCPUAcctUsage()
+	limitPeriod, limitQuota, err := cg.GetCPUCfsLimit()
 	if err != nil {
-		cpu.Usage = -1
 		return cpu
 	}
 
-	cpu.Usage = value
+	if limitQuota == -1 {
+		cpu.AllocatedTime = int64(cpuCount) * 1_000_000_000
+	} else {
+		cpu.AllocatedTime = 1_000_000_000 * limitQuota / limitPeriod
+	}
 
 	return cpu
 }

internal/server/instance/drivers/driver_qemu.go

@@ -8138,6 +8138,17 @@ func (d *qemu) renderState(statusCode api.StatusCode) (*api.InstanceState, error
 			}
 		}
 
+		// Populate the CPU time allocation
+		limitsCPU, ok := d.expandedConfig["limits.cpu"]
+		if ok {
+			cpuCount, err := strconv.ParseInt(limitsCPU, 10, 64)
+			if err != nil {
+				status.CPU.AllocatedTime = cpuCount * 1_000_000_000
+			}
+		} else {
+			status.CPU.AllocatedTime = qemudefault.CPUCores * 1_000_000_000
+		}
+
 		// Populate host_name for network devices.
 		for k, m := range d.ExpandedDevices() {
 			// We only care about nics.

internal/version/api.go

@@ -471,6 +471,7 @@ var APIExtensions = []string{
 	"network_dns_nameservers",
 	"acme_http01_port",
 	"network_ovn_ipv4_dhcp_expiry",
+	"instance_state_cpu_time",
 }
 
 // APIExtensionsCount returns the number of available API extensions.

shared/api/instance_state.go

@@ -98,6 +98,12 @@ type InstanceStateCPU struct {
 	// CPU usage in nanoseconds
 	// Example: 3637691016
 	Usage int64 `json:"usage" yaml:"usage"`
+
+	// CPU time available per second, in nanoseconds
+	// Example: 4000000000
+	//
+	// API extension: instance_state_cpu_time
+	AllocatedTime int64 `json:"allocated_time" yaml:"allocated_time"`
 }
 
 // InstanceStateMemory represents the memory information section of an instance's state.

test/suites/basic.sh

@@ -399,6 +399,18 @@ test_basic_usage() {
   [ "$(incus config get test-limits limits.cpu)" = "1" ]
   [ "$(incus config get test-limits limits.cpu.allowance)" = "50%" ]
   [ "$(incus config get test-limits limits.memory)" = "204MiB" ]
+
+  # Test CPU allocation information
+  [ "$(incus query /1.0/instances/test-limits/state | jq -r '.cpu.allocated_time')" = "1000000000" ]
+  incus config set test-limits limits.cpu.allowance 100ms/200ms
+  [ "$(incus query /1.0/instances/test-limits/state | jq -r '.cpu.allocated_time')" = "500000000" ]
+  incus config set test-limits limits.cpu 2
+  [ "$(incus query /1.0/instances/test-limits/state | jq -r '.cpu.allocated_time')" = "500000000" ]
+  incus config unset test-limits limits.cpu.allowance
+  [ "$(incus query /1.0/instances/test-limits/state | jq -r '.cpu.allocated_time')" = "2000000000" ]
+  incus config unset test-limits limits.cpu
+  [ "$(incus query /1.0/instances/test-limits/state | jq -r '.cpu.allocated_time')" = "$(nproc)000000000" ]
+
   incus delete -f test-limits
 
   # Test last_used_at field is working properly