feat(delivery): support multi-instance scheduling on a single host

- Per-host instance limit (DELIVERY_HOST_INSTANCE_LIMIT, default 4) - Optional target_host to pin a host from the candidate pool - Regenerate swagger docs
2026-07-27 15:19:44 +08:00
parent bfb4f0df4d
commit 41b2f45420
7 changed files with 355 additions and 439 deletions
@@ -24,6 +24,9 @@ import (

 var dnsLabelPattern = regexp.MustCompile(`^[a-z0-9](?:[-a-z0-9]*[a-z0-9])?$`)

+// AWX inventory 主机名：允许字母数字、中划线与点（FQDN）。
+var hostNamePattern = regexp.MustCompile(`^[a-zA-Z0-9](?:[-a-zA-Z0-9.]*[a-zA-Z0-9])?$`)
+
 type MySQLDeliveryInput struct {
 	BusinessLineID uint64 `json:"business_line_id" binding:"required"`
 	TargetID       uint64 `json:"target_id" binding:"required"`
@@ -33,9 +36,11 @@ type MySQLDeliveryInput struct {
 	Topology       string `json:"topology"`
 	MySQLPort      int    `json:"mysql_port"`
 	DataDisk       string `json:"data_disk"`
-	CPUMilli       int64  `json:"cpu_milli" binding:"required"`
-	MemoryMi       int64  `json:"memory_mi" binding:"required"`
-	StorageGi      int64  `json:"storage_gi" binding:"required"`
+	// 调度控制（选填）：点名候选池内主机跳过自动选机，端口/配额/实机守卫照常执行
+	TargetHost string `json:"target_host"`
+	CPUMilli   int64  `json:"cpu_milli" binding:"required"`
+	MemoryMi   int64  `json:"memory_mi" binding:"required"`
+	StorageGi  int64  `json:"storage_gi" binding:"required"`
 	// 数据库配置（选填，缺省由 playbook 基线兜底）
 	Timezone            string `json:"timezone"`
 	LowerCaseTableNames *int   `json:"lower_case_table_names"`
@@ -93,20 +98,31 @@ func parseTargetMetadata(raw string) targetMetadata {
 	return meta
 }

-// firstFreeHost 返回候选池中第一个未被占用的节点。
-func firstFreeHost(hosts []targetHost, occupied []string) *targetHost {
-	taken := make(map[string]bool, len(occupied))
-	for _, h := range occupied {
-		taken[h] = true
+// firstFreeHost 返回候选池中非失败任务数未达单机实例上限的第一个节点；
+// limit < 1 时按 1 兜底（退化为旧的一机一实例语义）。顺序遍历天然形成"先摊开、摊满一轮再叠加"。
+func firstFreeHost(hosts []targetHost, occupied []string, limit int) *targetHost {
+	if limit < 1 {
+		limit = 1
 	}
 	for i := range hosts {
-		if !taken[hosts[i].Name] {
+		if hostTaskCount(occupied, hosts[i].Name) < limit {
 			return &hosts[i]
 		}
 	}
 	return nil
 }

+// hostTaskCount 统计某主机在占用清单（非失败任务的 target_host 列表，含重复）中的出现次数。
+func hostTaskCount(occupied []string, name string) int {
+	count := 0
+	for _, h := range occupied {
+		if h == name {
+			count++
+		}
+	}
+	return count
+}
+
 // 端口池 13306-13999：混合分配模型，用户留空时自动分配，可覆盖为池内指定端口。
 const (
 	mysqlPortPoolStart = 13306
@@ -344,6 +360,9 @@ func validateDeliveryInput(input MySQLDeliveryInput, dataDisks []string) error {
 			return fmt.Errorf("data_disk %q is not in the allowed mount point list %v", input.DataDisk, allowed)
 		}
 	}
+	if input.TargetHost != "" && (len(input.TargetHost) > 253 || !hostNamePattern.MatchString(input.TargetHost)) {
+		return fmt.Errorf("target_host must be a valid inventory host name")
+	}
 	if input.Timezone != "" && !timezonePattern.MatchString(input.Timezone) {
 		return fmt.Errorf("timezone must be an offset like +08:00, SYSTEM, or a named zone like Asia/Shanghai")
 	}
@@ -498,7 +517,25 @@ func (s *DeliveryService) claimAndReserve(ctx context.Context) (*model.DeliveryT
 		if err := tx.Model(&model.DeliveryTask{}).Where("target_id = ? AND target_host <> ? AND status NOT IN ?", task.TargetID, "", occupiedExclude).Pluck("target_host", &occupied).Error; err != nil {
 			return err
 		}
-		host := firstFreeHost(meta.Hosts, occupied)
+		limit := s.cfg.DeliveryHostInstanceLimit
+		var host *targetHost
+		if payload.TargetHost != "" {
+			// 点名主机：仅校验池内存在性与单机实例上限，端口/配额/实机守卫照常执行。
+			for i := range meta.Hosts {
+				if meta.Hosts[i].Name == payload.TargetHost {
+					host = &meta.Hosts[i]
+					break
+				}
+			}
+			if host == nil {
+				return s.failInTransaction(tx, &task, model.TaskValidationFailed, fmt.Sprintf("target_host %q is not in the candidate host pool", payload.TargetHost))
+			}
+			if effective := max(limit, 1); hostTaskCount(occupied, host.Name) >= effective {
+				return fmt.Errorf("defer: pinned host %s reached the per-host instance limit %d", host.Name, effective)
+			}
+		} else {
+			host = firstFreeHost(meta.Hosts, occupied, limit)
+		}
 		if host == nil {
 			return fmt.Errorf("defer: no free host available on target")
 		}
@@ -15,6 +15,7 @@ func TestValidateDeliveryInput(t *testing.T) {
 	full.Topology = "standalone"
 	full.MySQLPort = 13306
 	full.DataDisk = "/disk1"
+	full.TargetHost = "k8s-server-03"
 	full.MemoryMi = 8192
 	full.StorageGi = 2000
 	full.Timezone = "+08:00"
@@ -55,6 +56,7 @@ func TestValidateDeliveryInput(t *testing.T) {
 		"port below pool":       func(in *MySQLDeliveryInput) { in.MySQLPort = 3307 },
 		"port above pool":       func(in *MySQLDeliveryInput) { in.MySQLPort = 14000 },
 		"data disk not in list": func(in *MySQLDeliveryInput) { in.DataDisk = "/mnt/other" },
+		"bad target host":       func(in *MySQLDeliveryInput) { in.TargetHost = "-bad-host" },
 		"bad timezone":          func(in *MySQLDeliveryInput) { in.Timezone = "UTC+8" },
 		"bad lower case":        func(in *MySQLDeliveryInput) { in.LowerCaseTableNames = intPtr(2) },
 		"bad charset":           func(in *MySQLDeliveryInput) { in.CharacterSet = "big5" },
@@ -77,6 +79,30 @@ func TestValidateDeliveryInput(t *testing.T) {
 	}
 }

+func TestFirstFreeHost(t *testing.T) {
+	hosts := []targetHost{{Name: "node-a"}, {Name: "node-b"}}
+	if h := firstFreeHost(hosts, nil, 1); h == nil || h.Name != "node-a" {
+		t.Fatalf("expected node-a on empty occupancy, got %+v", h)
+	}
+	if h := firstFreeHost(hosts, []string{"node-a"}, 1); h == nil || h.Name != "node-b" {
+		t.Fatalf("expected node-b when node-a is full at limit 1, got %+v", h)
+	}
+	if h := firstFreeHost(hosts, []string{"node-a", "node-b"}, 1); h != nil {
+		t.Fatalf("limit 1 with all hosts taken should return nil, got %+v", h)
+	}
+	// 摊满一轮后回到首台叠加第二个实例
+	if h := firstFreeHost(hosts, []string{"node-a", "node-b"}, 2); h == nil || h.Name != "node-a" {
+		t.Fatalf("expected node-a for second round at limit 2, got %+v", h)
+	}
+	if h := firstFreeHost(hosts, []string{"node-a", "node-a", "node-b", "node-b"}, 2); h != nil {
+		t.Fatalf("limit 2 with all hosts saturated should return nil, got %+v", h)
+	}
+	// limit < 1 退化为一机一实例
+	if h := firstFreeHost(hosts, []string{"node-a"}, 0); h == nil || h.Name != "node-b" {
+		t.Fatalf("limit 0 should degrade to 1, got %+v", h)
+	}
+}
+
 func TestAllocatePort(t *testing.T) {
 	if port, err := allocatePort(0, nil); err != nil || port != mysqlPortPoolStart {
 		t.Fatalf("expected first pool port %d, got %d err=%v", mysqlPortPoolStart, port, err)