I1UY6Q 添加参数校验：用户，用户组，版本号一致性等

src/manager/om/script/gs_expansion

@@ -20,12 +20,8 @@
 #############################################################################
 
 import os
-import pwd
 import sys
-import threading
-import uuid
-import subprocess
-import weakref
+
 
 sys.path.append(sys.path[0])
 from gspylib.common.DbClusterInfo import dbClusterInfo, \
@@ -61,6 +57,7 @@ class Expansion(ParallelBaseOM):
                     os.path.join(os.path.realpath(__file__), "../../"))
 
         self.standbyLocalMode = False
+        self.envFile = DefaultValue.getEnv("MPPDB_ENV_SEPARATE_PATH")
 
     def usage(self):
         """
@@ -142,6 +139,13 @@ General options:
         backIpList = clusterInfo.getClusterBackIps()
         nodeNameList = clusterInfo.getClusterNodeNames()
 
+        # only support single az now.
+        azNames = clusterInfo.getazNames()
+        self.azName = "AZ1"
+        if len(azNames) > 0:
+            self.azName = azNames[0]
+        
+        self.localIp = backIpList[0]
         self.nodeNameList = nodeNameList
         self.backIpNameMap = {}
         for backip in backIpList:

src/manager/om/script/gspylib/common/ErrorCode.py

@@ -1108,7 +1108,11 @@ class ErrorCode():
                        "database on node is abnormal. \n"
                        "node [%s], user [%s], dataNode [%s]. \n"
                        "You can use command \"gs_ctl query -D %s\" for more "
-                       "detail."
+                       "detail.",
+        "GAUSS_35704": "[GAUSS-35704] %s [%s] does not exist on node [%s].",
+        "GAUSS_35705": "[GAUSS-35705] Error, the database version is "
+                       "inconsistent in %s: %s" 
+                       
     }
 
 

src/manager/om/script/impl/expansion/ExpansionImpl.py

@@ -25,8 +25,12 @@ import os
 import getpass
 import pwd
 import datetime
+import weakref
 from random import sample
 import time
+import grp
+import socket
+import stat
 from multiprocessing import Process, Value
 
 sys.path.append(sys.path[0] + "/../../../../")
@@ -134,9 +138,10 @@ class ExpansionImpl():
         for host in newHosts:
             # create single deploy xml file for each standby node
             xmlContent = self.__generateXml(host)
-            fo = open("%s" % tempXmlFile, "w")
-            fo.write( xmlContent )
-            fo.close()
+            with os.fdopen(os.open("%s" % tempXmlFile, os.O_WRONLY | os.O_CREAT,
+             stat.S_IWUSR | stat.S_IRUSR),'w') as fo:
+                fo.write( xmlContent )
+                fo.close()
             # send single deploy xml file to each standby node
             sshTool = SshTool(host)
             retmap, output = sshTool.getSshStatusOutput("mkdir -p %s" % 
@@ -180,7 +185,7 @@ class ExpansionImpl():
     <DEVICELIST>
         <DEVICE sn="1000001">
             <PARAM name="name" value="{nodeName}"/>
-            <PARAM name="azName" value="AZ1"/>
+            <PARAM name="azName" value="{azName}"/>
             <PARAM name="azPriority" value="1"/>
             <PARAM name="backIp1" value="{backIp}"/>
             <PARAM name="sshIp1" value="{sshIp}"/>
@@ -193,7 +198,7 @@ class ExpansionImpl():
 </ROOT>
         """.format(nodeName=nodeName,backIp=backIp,appPath=appPath,
         logPath=logPath,toolPath=toolPath,corePath=corePath,
-        sshIp=sshIp,port=port,dataNode=dataNode)
+        sshIp=sshIp,port=port,dataNode=dataNode,azName=self.context.azName)
         return xmlConfig
 
     def changeUser(self):
@@ -582,10 +587,11 @@ retry for %s times" % start_retry_num)
             subprocess.getstatusoutput("mkdir -m a+x -p %s; touch %s; \
                 cat /dev/null > %s" % \
                     (self.tempFileDir, tempShFile, tempShFile))
-            fo = open("%s" % tempShFile, "w")
-            fo.write("#bash\n")
-            fo.write( command )
-            fo.close()
+            with os.fdopen(os.open("%s" % tempShFile, os.O_WRONLY | os.O_CREAT,
+             stat.S_IWUSR | stat.S_IRUSR),'w') as fo:
+                fo.write("#bash\n")
+                fo.write( command )
+                fo.close()
 
             # send guc command bashfile to each host and execute it.
             sshTool.scpFiles("%s" % tempShFile, "%s" % tempShFile, [host], 
@@ -619,14 +625,14 @@ retry for %s times" % start_retry_num)
                 remoteHostInfo = nodeDict[remoteHost]
       
                 guc_repl_template = """\
-                gs_guc set -D {dn} -c "replconninfo{index}=\
-                'localhost={localhost} localport={localport} \
-                localheartbeatport={localeHeartPort} \
-                localservice={localservice} \
-                remotehost={remoteNode} \
-                remoteport={remotePort} \
-                remoteheartbeatport={remoteHeartPort} \
-                remoteservice={remoteservice}'"
+gs_guc set -D {dn} -c "replconninfo{index}=\
+'localhost={localhost} localport={localport} \
+localheartbeatport={localeHeartPort} \
+localservice={localservice} \
+remotehost={remoteNode} \
+remoteport={remotePort} \
+remoteheartbeatport={remoteHeartPort} \
+remoteservice={remoteservice}'"
                 """.format(dn=localeHostInfo["dataNode"],
                 index=index,
                 localhost=localeHostInfo["sshIp"],
@@ -653,8 +659,13 @@ retry for %s times" % start_retry_num)
 
     def checkLocalModeOnStandbyHosts(self):
         """
+        expansion the installed standby node. check standby database.
+        1. if the database is normal
+        2. if the databases version are same before existing and new 
         """
         standbyHosts = self.context.newHostList
+        envfile = self.envFile
+        
         self.logger.log("Checking the database with locale mode.")
         for host in standbyHosts:
             hostName = self.context.backIpNameMap[host]
@@ -665,6 +676,39 @@ retry for %s times" % start_retry_num)
                 GaussLog.exitWithError(ErrorCode.GAUSS_357["GAUSS_35703"] % 
                 (hostName, self.user, dataNode, dataNode))
         
+        allHostIp = []
+        allHostIp.append(self.context.localIp)
+        versionDic = {}
+
+        for hostip in standbyHosts:
+            allHostIp.append(hostip)
+        sshTool= SshTool(allHostIp)
+        #get version in the nodes 
+        getversioncmd = "gaussdb --version"
+        resultMap, outputCollect = sshTool.getSshStatusOutput(getversioncmd,
+                                                               [], envfile)
+        self.cleanSshToolFile(sshTool)
+        versionLines = outputCollect.splitlines()
+        for verline in versionLines:
+            if verline[0:9] == '[SUCCESS]':
+               ipKey = verline[10:-1]
+               continue
+            else:
+                versionStr = "".join(verline)
+                preVersion = versionStr.split(' ')
+                versionInfo = preVersion[4]
+                versionDic[ipKey] = versionInfo[:-2]
+        for hostip in versionDic:
+            if hostip == self.context.localIp:
+               versionCompare = ""
+               versionCompare = versionDic[hostip]
+            else:
+                if versionDic[hostip] == versionCompare:
+                    continue
+                else:
+                    GaussLog.exitWithError(ErrorCode.GAUSS_357["GAUSS_35705"] \
+                       %(hostip, versionDic[hostip]))
+        
         self.logger.log("Successfully checked the database with locale mode.")
 
     def preInstall(self):
@@ -686,12 +730,92 @@ standby nodes.")
         """
         self.logger.debug("start to delete temporary file")
         hostNames = self.context.nodeNameList
-        sshTool = SshTool(hostNames)
-        clearCmd = "source %s ; rm -rf %s" % (self.envFile, self.tempFileDir)
-        result, output = sshTool.getSshStatusOutput(clearCmd, 
-        hostNames, self.envFile)
-        self.logger.debug(output)
+    def checkNodesDetail(self):
+        """
+        """
+        self.checkUserAndGroupExists()
+        self.checkXmlFileAccessToUser()
+    
+    def checkXmlFileAccessToUser(self):
+        """
+        Check if the xml config file has readable access to user.
+        """
+        userInfo = pwd.getpwnam(self.user)
+        uid = userInfo.pw_uid
+        gid = userInfo.pw_gid
 
+        xmlFile = self.context.xmlFile
+        fstat = os.stat(xmlFile)
+        mode = fstat[stat.ST_MODE]
+        if (fstat[stat.ST_UID] == uid and (mode & stat.S_IRUSR > 0)) or \
+           (fstat[stat.ST_GID] == gid and (mode & stat.S_IRGRP > 0)):
+            pass
+        else:
+            self.logger.debug("User %s has no access right for file %s" \
+                 % (self.user, xmlFile))
+            os.chown(xmlFile, uid, gid)
+            os.chmod(xmlFile, stat.S_IRUSR)
+
+    def checkUserAndGroupExists(self):
+        """
+        check system user and group exists and be same 
+        on primary and standby nodes
+        """
+        inputUser = self.user
+        inputGroup = self.group
+        
+        user_group_id = ""
+        isUserExits = False
+        localHost = socket.gethostname()
+        for user in pwd.getpwall():
+            if user.pw_name == self.user:
+                user_group_id = user.pw_gid
+                isUserExits = True
+                break
+        if not isUserExits:
+            GaussLog.exitWithError(ErrorCode.GAUSS_357["GAUSS_35704"] \
+                % ("User", self.user, localHost))
+
+        isGroupExits = False
+        group_id = ""
+        for group in grp.getgrall():
+            if group.gr_name == self.group:
+                group_id = group.gr_gid
+                isGroupExits = True
+        if not isGroupExits:
+            GaussLog.exitWithError(ErrorCode.GAUSS_357["GAUSS_35704"] \
+                % ("Group", self.group, localHost))
+        if user_group_id != group_id:
+            GaussLog.exitWithError("User [%s] is not in the group [%s]."\
+                 % (self.user, self.group))
+        
+        hostNames = self.context.newHostList
+        envfile = self.envFile
+        sshTool = SshTool(hostNames)
+
+        #get username in the other standy nodes 
+        getUserNameCmd = "cat /etc/passwd | grep -w %s" % inputUser
+        resultMap, outputCollect = sshTool.getSshStatusOutput(getUserNameCmd, 
+        [], envfile)
+        
+        for hostKey in resultMap:
+            if resultMap[hostKey] == STATUS_FAIL:
+                self.cleanSshToolFile(sshTool)
+                GaussLog.exitWithError(ErrorCode.GAUSS_357["GAUSS_35704"] \
+                       % ("User", self.user, hostKey))
+        
+        #get groupname in the other standy nodes   
+        getGroupNameCmd = "cat /etc/group | grep -w %s" % inputGroup
+        resultMap, outputCollect = sshTool.getSshStatusOutput(getGroupNameCmd, 
+        [], envfile)
+        for hostKey in resultMap:
+            if resultMap[hostKey] == STATUS_FAIL:
+                self.cleanSshToolFile(sshTool)
+                GaussLog.exitWithError(ErrorCode.GAUSS_357["GAUSS_35704"] \
+                       % ("Group", self.group, hostKey))
+        self.cleanSshToolFile(sshTool)
+
+    
     def installAndExpansion(self):
         """
         install database and expansion standby node with db om user
@@ -727,6 +851,7 @@ Start to establish the primary-standby relationship.")
         """
         start expansion
         """
+        self.checkNodesDetail()
         # preinstall on standby nodes with root user.
         if not self.context.standbyLocalMode:
             self.preInstall()