Merge pull request #9 from beepfd/ghostbaby

feat: update Go version and enhance scheduling metrics

Author: Ghostbaby

.github/workflows/push.yml

@@ -15,7 +15,7 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update && sudo apt-get install -y \
-          golang-1.22 \
+          golang-1.23 \
           git \
           make \
           gcc \

bpf/trace.c

@@ -15,14 +15,15 @@ extern int LINUX_KERNEL_VERSION __kconfig;
 // 定义数据结构来存储调度延迟信息
 struct sched_latency_t
 {
-    __u32 pid;               // 进程ID
-    __u32 tid;               // 线程ID
-    __u64 delay_ns;          // 调度延迟(纳秒)
-    __u64 ts;                // 时间戳
-    __u32 preempted_pid;     // 被抢占的进程ID
-    char preempted_comm[16]; // 被抢占的进程名
-    __u64 is_preempt;        // 是否抢占(0: 否, 1: 是)
-    char comm[16];           // 进程名
+    __u32 pid;                 // 进程ID
+    __u32 tid;                 // 线程ID
+    __u64 delay_ns;            // 调度延迟(纳秒)
+    __u64 ts;                  // 时间戳
+    __u32 preempted_pid;       // 被抢占的进程ID
+    char preempted_comm[16];   // 被抢占的进程名
+    __u64 is_preempt;          // 是否抢占(0: 否, 1: 是)
+    char comm[16];             // 进程名
+    __u32 preempted_pid_state; // 被抢占的进程状态
 } __attribute__((packed));
 
 struct sched_latency_t *unused_sched_latency_t __attribute__((unused));
@@ -201,6 +202,7 @@ static __always_inline void handle_sched_switch(u32 prev_pid, u32 prev_tgid,
         .tid = next_pid,
         .delay_ns = delay,
         .ts = now,
+        .preempted_pid_state = prev_state,
     };
 
     bpf_probe_read_kernel_str(&latency.comm, sizeof(latency.comm), next_comm);

cmd/config.yaml

@@ -5,7 +5,7 @@ btf:
   kernel: "/sys/kernel/btf/vmlinux"
 
 output:
-  type: file
+  type: clickhouse
   clickhouse:
     host: "192.168.200.201"
     port: "9000"

deploy/sql/clickhouse/sched.ck

@@ -20,6 +20,8 @@ CREATE TABLE shepherd.sched_latency
 
     `date` Date DEFAULT today(),
 
+    `preempted_pid_state` UInt32,
+
     `datetime` DateTime64(9) DEFAULT now64(9)
 )
 ENGINE = MergeTree

go.mod

@@ -1,7 +1,6 @@
 module github.com/cen-ngc5139/shepherd
 
-go 1.22.4
-toolchain go1.24.1
+go 1.23.4
 
 require (
 	github.com/ClickHouse/clickhouse-go/v2 v2.30.0

internal/output/output.go

@@ -58,7 +58,8 @@ func (o *Output) InitSinkCli(cfg config.OutputConfig) (err error) {
 			INSERT INTO sched_latency (
 				pid, tid, delay_ns, ts, 
 				preempted_pid, preempted_comm, 
-				is_preempt, comm
+				is_preempt, comm,
+				preempted_pid_state
 			)
 		`)
 		if err != nil {

internal/output/sched_delay.go

@@ -113,6 +113,7 @@ func insertSchedMetrics(ctx context.Context, conn clickhouse.Conn, batch driver.
 		sanitizeString(convertInt8ToString(event.PreemptedComm[:])),
 		event.IsPreempt,
 		sanitizeString(convertInt8ToString(event.Comm[:])),
+		event.PreemptedPidState,
 	)
 	if err != nil {
 		log.Errorf("failed to append to batch: %v", err)
@@ -132,7 +133,8 @@ func insertSchedMetrics(ctx context.Context, conn clickhouse.Conn, batch driver.
 			INSERT INTO sched_latency (
 				pid, tid, delay_ns, ts, 
 				preempted_pid, preempted_comm, 
-				is_preempt, comm
+				is_preempt, comm,
+				preempted_pid_state
 			)
 		`)
 		if err != nil {

internal/output/utils.go

@@ -67,3 +67,67 @@ func filterNonASCII(data []byte) string {
 func sanitizeString(s string) string {
 	return strings.TrimSpace(s)
 }
+
+
+// 线程状态常量
+const (
+	TASK_RUNNING           = 0x00000000
+	TASK_INTERRUPTIBLE     = 0x00000001
+	TASK_UNINTERRUPTIBLE   = 0x00000002
+	TASK_STOPPED           = 0x00000004
+	TASK_TRACED            = 0x00000008
+	EXIT_DEAD              = 0x00000010
+	EXIT_ZOMBIE            = 0x00000020
+	EXIT_TRACE             = EXIT_ZOMBIE | EXIT_DEAD
+	TASK_PARKED            = 0x00000040
+	TASK_DEAD              = 0x00000080
+	TASK_WAKEKILL          = 0x00000100
+	TASK_WAKING            = 0x00000200
+	TASK_NOLOAD            = 0x00000400
+	TASK_NEW               = 0x00000800
+	TASK_RTLOCK_WAIT       = 0x00001000
+	TASK_FREEZABLE         = 0x00002000
+	TASK_FREEZABLE_UNSAFE  = 0x00004000 // 取决于: IS_ENABLED(CONFIG_LOCKDEP)
+	TASK_FROZEN            = 0x00008000
+	TASK_STATE_MAX         = 0x00010000 // 截至 Linux 内核 6.9
+)
+
+// 任务状态映射表
+var taskStates = map[uint32]string{
+	0x00000000: "R",  // "RUNNING"
+	0x00000001: "S",  // "INTERRUPTIBLE"
+	0x00000002: "D",  // "UNINTERRUPTIBLE"
+	0x00000004: "T",  // "STOPPED"
+	0x00000008: "t",  // "TRACED"
+	0x00000010: "X",  // "EXIT_DEAD"
+	0x00000020: "Z",  // "EXIT_ZOMBIE"
+	0x00000040: "P",  // "PARKED"
+	0x00000080: "dd", // "DEAD"
+	0x00000100: "wk", // "WAKEKILL"
+	0x00000200: "wg", // "WAKING"
+	0x00000400: "I",  // "NOLOAD"
+	0x00000800: "N",  // "NEW"
+	0x00001000: "rt", // "RTLOCK_WAIT"
+	0x00002000: "fe", // "FREEZABLE"
+	0x00004000: "fu", // "__TASK_FREEZABLE_UNSAFE = (0x00004000 * IS_ENABLED(CONFIG_LOCKDEP))"
+	0x00008000: "fo", // "FROZEN"
+}
+
+// GetTaskStateName 将内核任务状态位掩码转换为可读字符串
+func GetTaskStateName(taskState uint32) string {
+	if taskState == 0 {
+		return "R"
+	}
+	if taskState&TASK_NOLOAD != 0 { // 空闲内核线程等待工作
+		return "I"
+	}
+
+	var names []string
+	for state, name := range taskStates {
+		if taskState&state != 0 {
+			names = append(names, name)
+		}
+	}
+
+	return strings.Join(names, "+")
+}