Fix(deploy): wait all chunkserver online time (opencurve#114).

Signed-off-by: Wine93 <[email protected]>

cli/command/deploy.go

@@ -200,13 +200,20 @@ func precheckBeforeDeploy(curveadm *cli.CurveAdm,
 	return nil
 }
 
+func calcNumOfChunkserver(curveadm *cli.CurveAdm, dcs []*topology.DeployConfig) int {
+	services := curveadm.FilterDeployConfigByRole(dcs, topology.ROLE_CHUNKSERVER)
+	return len(services)
+}
+
 func genDeployPlaybook(curveadm *cli.CurveAdm,
 	dcs []*topology.DeployConfig,
 	options deployOptions) (*playbook.Playbook, error) {
+	var steps []int
 	kind := dcs[0].GetKind()
-	steps := CURVEFS_DEPLOY_STEPS
 	if kind == topology.KIND_CURVEBS {
 		steps = CURVEBS_DEPLOY_STEPS
+	} else {
+		steps = CURVEFS_DEPLOY_STEPS
 	}
 	steps = skipDeploySteps(steps, options)
 
@@ -230,6 +237,7 @@ func genDeployPlaybook(curveadm *cli.CurveAdm,
 			options[comm.KEY_CREATE_POOL_TYPE] = comm.POOL_TYPE_PHYSICAL
 		} else if step == CREATE_LOGICAL_POOL {
 			options[comm.KEY_CREATE_POOL_TYPE] = comm.POOL_TYPE_LOGICAL
+			options[comm.KEY_NUMBER_OF_CHUNKSERVER] = calcNumOfChunkserver(curveadm, dcs)
 		}
 
 		pb.AddStep(&playbook.PlaybookStep{

cli/command/migrate.go

@@ -214,6 +214,7 @@ func genMigratePlaybook(curveadm *cli.CurveAdm,
 			options[comm.KEY_CREATE_POOL_TYPE] = comm.POOL_TYPE_LOGICAL
 			options[comm.KEY_MIGRATE_SERVERS] = migrates
 			options[comm.KEY_NEW_TOPOLOGY_DATA] = data
+			options[comm.KEY_NUMBER_OF_CHUNKSERVER] = calcNumOfChunkserver(curveadm, dcs)
 		case playbook.UPDATE_TOPOLOGY:
 			options[comm.KEY_NEW_TOPOLOGY_DATA] = data
 		}

cli/command/scale_out.go

@@ -23,6 +23,8 @@
 package command
 
 import (
+	"time"
+
 	"github.com/fatih/color"
 	"github.com/opencurve/curveadm/cli/cli"
 	comm "github.com/opencurve/curveadm/internal/common"
@@ -33,7 +35,6 @@ import (
 	cliutil "github.com/opencurve/curveadm/internal/utils"
 	utils "github.com/opencurve/curveadm/internal/utils"
 	"github.com/spf13/cobra"
-	"time"
 )
 
 var (
@@ -334,6 +335,8 @@ func genScaleOutPlaybook(curveadm *cli.CurveAdm,
 			options[comm.KEY_CREATE_POOL_TYPE] = comm.POOL_TYPE_LOGICAL
 			options[comm.KEY_SCALE_OUT_CLUSTER] = dcs2scaleOut
 			options[comm.KEY_NEW_TOPOLOGY_DATA] = data
+			options[comm.KEY_NUMBER_OF_CHUNKSERVER] = calcNumOfChunkserver(curveadm, dcs) +
+				calcNumOfChunkserver(curveadm, dcs2scaleOut)
 		case playbook.UPDATE_TOPOLOGY:
 			options[comm.KEY_NEW_TOPOLOGY_DATA] = data
 		}

internal/common/common.go

@@ -41,10 +41,11 @@ var (
 // task options
 const (
 	// common
-	KEY_ALL_DEPLOY_CONFIGS = "ALL_DEPLOY_CONFIGS"
-	KEY_CREATE_POOL_TYPE   = "POOL_TYPE"
-	POOL_TYPE_LOGICAL      = "logicalpool"
-	POOL_TYPE_PHYSICAL     = "physicalpool"
+	KEY_ALL_DEPLOY_CONFIGS    = "ALL_DEPLOY_CONFIGS"
+	KEY_CREATE_POOL_TYPE      = "POOL_TYPE"
+	POOL_TYPE_LOGICAL         = "logicalpool"
+	POOL_TYPE_PHYSICAL        = "physicalpool"
+	KEY_NUMBER_OF_CHUNKSERVER = "NUMBER_OF_CHUNKSERVER"
 
 	// format
 	KEY_ALL_FORMAT_STATUS = "ALL_FORMAT_STATUS"

internal/task/scripts/shell/wait_chunkserver.sh

@@ -1,23 +1,23 @@
 #!/usr/bin/env bash
 
-# Usage: no args, just run it in bash
+# Usage: wait_chunkserver 3
 # Created Date: 2022-03-09
 # Author: aspirer
 
 
-# FIXME(P0): wait not works, return -12
+g_total="$1"
+
 wait=0
 while ((wait<60))
 do
-    status=$(curve_ops_tool chunkserver-status |grep "offline")
-    total=$(echo ${status} | grep -c "total num = 0")
-    offline=$(echo ${status} | grep -c "offline = 0")
-    if [ ${total} -eq 0 ] && [ ${offline} -eq 1 ]; then
-        echo "CURVEADM_SUCCESS"
+    online=$(curve_ops_tool chunkserver-status | sed -nr 's/.*online = ([0-9]+).*/\1/p')
+    if [[ $online -eq $g_total ]]; then
         exit 0
     fi
+
     sleep 0.5s
-    wait=$(expr ${wait} + 1)
+    wait=$((wait+1))
 done
-echo "CURVEADM_TIMEOUT"
+
+echo "wait all chunkserver online timeout, total=$g_total, online=$online"
 exit 1

internal/task/task/common/create_pool.go

@@ -187,8 +187,8 @@ func NewCreateTopologyTask(curveadm *cli.CurveAdm, dc *topology.DeployConfig) (*
 
 	// new task
 	pooltype := curveadm.MemStorage().Get(comm.KEY_CREATE_POOL_TYPE).(string)
-	name := utils.Choose(pooltype == comm.POOL_TYPE_LOGICAL, "Create Logical Pool",
-		"Create Physical Pool")
+	name := utils.Choose(pooltype == comm.POOL_TYPE_LOGICAL,
+		"Create Logical Pool", "Create Physical Pool")
 	subname := fmt.Sprintf("host=%s role=%s containerId=%s",
 		dc.GetHost(), dc.GetRole(), tui.TrimContainerId(containerId))
 	t := task.NewTask(name, subname, hc.GetSSHConfig())
@@ -244,6 +244,7 @@ func NewCreateTopologyTask(curveadm *cli.CurveAdm, dc *topology.DeployConfig) (*
 	if dc.GetKind() == topology.KIND_CURVEBS && pooltype == comm.POOL_TYPE_LOGICAL {
 		waitChunkserversScript := scripts.WAIT_CHUNKSERVERS
 		waitChunkserversScriptPath := fmt.Sprintf("%s/wait_chunkservers.sh", layout.ToolsBinDir)
+		nchunkserver := curveadm.MemStorage().Get(comm.KEY_NUMBER_OF_CHUNKSERVER).(int)
 		t.AddStep(&step.InstallFile{ // install wait_chunkservers script
 			ContainerId:       &containerId,
 			ContainerDestPath: waitChunkserversScriptPath,
@@ -252,7 +253,7 @@ func NewCreateTopologyTask(curveadm *cli.CurveAdm, dc *topology.DeployConfig) (*
 		})
 		t.AddStep(&step.ContainerExec{ // wait all chunkservers online before create logical pool
 			ContainerId: &containerId,
-			Command:     fmt.Sprintf("bash %s", waitChunkserversScriptPath),
+			Command:     fmt.Sprintf("bash %s %d", waitChunkserversScriptPath, nchunkserver),
 			Success:     &success,
 			Out:         &out,
 			ExecOptions: curveadm.ExecOptions(),