Add ctr metrics support for Windows/LCOW containers

Signed-off-by: Justin Terry (VM) <juterry@microsoft.com>

cmd/ctr/commands/tasks/metrics.go

@@ -25,6 +25,7 @@ import (
 	"os"
 	"text/tabwriter"
 
+	wstats "github.com/Microsoft/hcsshim/cmd/containerd-shim-runhcs-v1/stats"
 	v1 "github.com/containerd/cgroups/stats/v1"
 	"github.com/containerd/containerd/cmd/ctr/commands"
 	"github.com/containerd/typeurl"
@@ -77,9 +78,17 @@ var metricsCommand = cli.Command{
 		if err != nil {
 			return err
 		}
-		data, ok := anydata.(*v1.Metrics)
+		var (
-		if !ok {
+			data         *v1.Metrics
-			return errors.New("cannot convert metric data to cgroups.Metrics")
+			windowsStats *wstats.Statistics
+		)
+		switch v := anydata.(type) {
+		case *v1.Metrics:
+			data = v
+		case *wstats.Statistics:
+			windowsStats = v
+		default:
+			return errors.New("cannot convert metric data to cgroups.Metrics or windows.Statistics")
 		}
 
 		switch context.String(formatFlag) {
@@ -87,20 +96,18 @@ var metricsCommand = cli.Command{
 			w := tabwriter.NewWriter(os.Stdout, 1, 8, 4, ' ', 0)
 			fmt.Fprintf(w, "ID\tTIMESTAMP\t\n")
 			fmt.Fprintf(w, "%s\t%s\t\n\n", metric.ID, metric.Timestamp)
-
+			if data != nil {
-			fmt.Fprintf(w, "METRIC\tVALUE\t\n")
+				printCgroupMetricsTable(w, data)
-			if data.Memory != nil {
+			} else {
-				fmt.Fprintf(w, "memory.usage_in_bytes\t%d\t\n", data.Memory.Usage.Usage)
+				if windowsStats.GetLinux() != nil {
-				fmt.Fprintf(w, "memory.limit_in_bytes\t%d\t\n", data.Memory.Usage.Limit)
+					printCgroupMetricsTable(w, windowsStats.GetLinux())
-				fmt.Fprintf(w, "memory.stat.cache\t%d\t\n", data.Memory.TotalCache)
+				} else if windowsStats.GetWindows() != nil {
-			}
+					printWindowsContainerStatistics(w, windowsStats.GetWindows())
-			if data.CPU != nil {
+				}
-				fmt.Fprintf(w, "cpuacct.usage\t%d\t\n", data.CPU.Usage.Total)
+				// Print VM stats if its isolated
-				fmt.Fprintf(w, "cpuacct.usage_percpu\t%v\t\n", data.CPU.Usage.PerCPU)
+				if windowsStats.VM != nil {
-			}
+					printWindowsVMStatistics(w, windowsStats.VM)
-			if data.Pids != nil {
+				}
-				fmt.Fprintf(w, "pids.current\t%v\t\n", data.Pids.Current)
-				fmt.Fprintf(w, "pids.limit\t%v\t\n", data.Pids.Limit)
 			}
 			return w.Flush()
 		case formatJSON:
@@ -115,3 +122,61 @@ var metricsCommand = cli.Command{
 		}
 	},
 }
+
+func printCgroupMetricsTable(w *tabwriter.Writer, data *v1.Metrics) {
+	fmt.Fprintf(w, "METRIC\tVALUE\t\n")
+	if data.Memory != nil {
+		fmt.Fprintf(w, "memory.usage_in_bytes\t%d\t\n", data.Memory.Usage.Usage)
+		fmt.Fprintf(w, "memory.limit_in_bytes\t%d\t\n", data.Memory.Usage.Limit)
+		fmt.Fprintf(w, "memory.stat.cache\t%d\t\n", data.Memory.TotalCache)
+	}
+	if data.CPU != nil {
+		fmt.Fprintf(w, "cpuacct.usage\t%d\t\n", data.CPU.Usage.Total)
+		fmt.Fprintf(w, "cpuacct.usage_percpu\t%v\t\n", data.CPU.Usage.PerCPU)
+	}
+	if data.Pids != nil {
+		fmt.Fprintf(w, "pids.current\t%v\t\n", data.Pids.Current)
+		fmt.Fprintf(w, "pids.limit\t%v\t\n", data.Pids.Limit)
+	}
+}
+
+func printWindowsContainerStatistics(w *tabwriter.Writer, stats *wstats.WindowsContainerStatistics) {
+	fmt.Fprintf(w, "METRIC\tVALUE\t\n")
+	fmt.Fprintf(w, "timestamp\t%s\t\n", stats.Timestamp)
+	fmt.Fprintf(w, "start_time\t%s\t\n", stats.ContainerStartTime)
+	fmt.Fprintf(w, "uptime_ns\t%d\t\n", stats.UptimeNS)
+	if stats.Processor != nil {
+		fmt.Fprintf(w, "cpu.total_runtime_ns\t%d\t\n", stats.Processor.TotalRuntimeNS)
+		fmt.Fprintf(w, "cpu.runtime_user_ns\t%d\t\n", stats.Processor.RuntimeUserNS)
+		fmt.Fprintf(w, "cpu.runtime_kernel_ns\t%d\t\n", stats.Processor.RuntimeUserNS)
+	}
+	if stats.Memory != nil {
+		fmt.Fprintf(w, "memory.commit_bytes\t%d\t\n", stats.Memory.MemoryUsageCommitBytes)
+		fmt.Fprintf(w, "memory.commit_peak_bytes\t%d\t\n", stats.Memory.MemoryUsageCommitPeakBytes)
+		fmt.Fprintf(w, "memory.private_working_set_bytes\t%d\t\n", stats.Memory.MemoryUsagePrivateWorkingSetBytes)
+	}
+	if stats.Storage != nil {
+		fmt.Fprintf(w, "storage.read_count_normalized\t%d\t\n", stats.Storage.ReadCountNormalized)
+		fmt.Fprintf(w, "storage.read_size_bytes\t%d\t\n", stats.Storage.ReadSizeBytes)
+		fmt.Fprintf(w, "storage.write_count_normalized\t%d\t\n", stats.Storage.WriteCountNormalized)
+		fmt.Fprintf(w, "storage.write_size_bytes\t%d\t\n", stats.Storage.WriteSizeBytes)
+	}
+}
+
+func printWindowsVMStatistics(w *tabwriter.Writer, stats *wstats.VirtualMachineStatistics) {
+	fmt.Fprintf(w, "METRIC\tVALUE\t\n")
+	if stats.Processor != nil {
+		fmt.Fprintf(w, "vm.cpu.total_runtime_ns\t%d\t\n", stats.Processor.TotalRuntimeNS)
+	}
+	if stats.Memory != nil {
+		fmt.Fprintf(w, "vm.memory.working_set_bytes\t%d\t\n", stats.Memory.WorkingSetBytes)
+		fmt.Fprintf(w, "vm.memory.virtual_node_count\t%d\t\n", stats.Memory.VirtualNodeCount)
+		fmt.Fprintf(w, "vm.memory.available\t%d\t\n", stats.Memory.VmMemory.AvailableMemory)
+		fmt.Fprintf(w, "vm.memory.available_buffer\t%d\t\n", stats.Memory.VmMemory.AvailableMemoryBuffer)
+		fmt.Fprintf(w, "vm.memory.reserved\t%d\t\n", stats.Memory.VmMemory.ReservedMemory)
+		fmt.Fprintf(w, "vm.memory.assigned\t%d\t\n", stats.Memory.VmMemory.AssignedMemory)
+		fmt.Fprintf(w, "vm.memory.slp_active\t%t\t\n", stats.Memory.VmMemory.SlpActive)
+		fmt.Fprintf(w, "vm.memory.balancing_enabled\t%t\t\n", stats.Memory.VmMemory.BalancingEnabled)
+		fmt.Fprintf(w, "vm.memory.dm_operation_in_progress\t%t\t\n", stats.Memory.VmMemory.DmOperationInProgress)
+	}
+}

vendor/github.com/Microsoft/hcsshim/cmd/containerd-shim-runhcs-v1/stats/doc.go

@@ -0,0 +1,6 @@
+package stats
+
+import (
+	// go mod will not vendor without an import for metrics.proto
+	_ "github.com/containerd/cgroups/stats/v1"
+)

vendor/github.com/Microsoft/hcsshim/cmd/containerd-shim-runhcs-v1/stats/stats.proto

@@ -0,0 +1,70 @@
+syntax = "proto3";
+
+package containerd.runhcs.stats.v1;
+
+import weak "gogoproto/gogo.proto";
+import "google/protobuf/timestamp.proto";
+import "github.com/containerd/cgroups/stats/v1/metrics.proto";
+
+option go_package = "github.com/Microsoft/hcsshim/cmd/containerd-shim-runhcs-v1/stats;stats";
+
+message Statistics {
+	oneof container {
+		WindowsContainerStatistics windows = 1;
+		io.containerd.cgroups.v1.Metrics linux = 2;
+	}
+	VirtualMachineStatistics vm = 3 [(gogoproto.customname) = "VM"];
+}
+
+message WindowsContainerStatistics {
+	google.protobuf.Timestamp timestamp = 1 [(gogoproto.stdtime) = true, (gogoproto.nullable) = false];
+	google.protobuf.Timestamp container_start_time = 2 [(gogoproto.stdtime) = true, (gogoproto.nullable) = false];
+	uint64 uptime_ns = 3 [(gogoproto.customname) = "UptimeNS"];
+	WindowsContainerProcessorStatistics processor = 4;
+	WindowsContainerMemoryStatistics memory = 5;
+	WindowsContainerStorageStatistics storage = 6;
+}
+
+message WindowsContainerProcessorStatistics {
+	uint64 total_runtime_ns = 1 [(gogoproto.customname) = "TotalRuntimeNS"];
+	uint64 runtime_user_ns = 2 [(gogoproto.customname) = "RuntimeUserNS"];
+	uint64 runtime_kernel_ns = 3 [(gogoproto.customname) = "RuntimeKernelNS"];
+}
+
+message WindowsContainerMemoryStatistics {
+	uint64 memory_usage_commit_bytes = 1;
+	uint64 memory_usage_commit_peak_bytes = 2;
+	uint64 memory_usage_private_working_set_bytes = 3;
+}
+
+message WindowsContainerStorageStatistics {
+	uint64 read_count_normalized = 1;
+	uint64 read_size_bytes = 2;
+	uint64 write_count_normalized = 3;
+	uint64 write_size_bytes = 4;
+}
+
+message VirtualMachineStatistics {
+	VirtualMachineProcessorStatistics processor = 1;
+	VirtualMachineMemoryStatistics memory = 2;
+}
+
+message VirtualMachineProcessorStatistics {
+	uint64 total_runtime_ns = 1 [(gogoproto.customname) = "TotalRuntimeNS"];
+}
+
+message VirtualMachineMemoryStatistics {
+	uint64 working_set_bytes = 1;
+	uint32 virtual_node_count = 2;
+	VirtualMachineMemory vm_memory = 3;
+}
+
+message VirtualMachineMemory {
+	int32 available_memory = 1;
+	int32 available_memory_buffer = 2;
+	uint64 reserved_memory = 3;
+	uint64 assigned_memory = 4;
+	bool slp_active = 5;
+	bool balancing_enabled = 6;
+	bool dm_operation_in_progress = 7;
+}