create hosts before creating vc

Author: FirmlyReality

conf/container.batch.conf

@@ -45,7 +45,7 @@ lxc.cgroup.cpu.cfs_quota_us = %CONTAINER_CPU%
 lxc.cap.drop = sys_admin net_admin mac_admin mac_override sys_time sys_module
 
 lxc.mount.entry = %FS_PREFIX%/global/users/%USERNAME%/data %ROOTFS%/root/nfs none bind,rw,create=dir 0 0
-#lxc.mount.entry = %FS_PREFIX%/global/users/%USERNAME%/hosts/%CLUSTERID%.hosts %ROOTFS%/etc/hosts none bind,ro,create=file 0 0
+#lxc.mount.entry = %FS_PREFIX%/global/users/%USERNAME%/hosts/batch-%TASKID%.hosts %ROOTFS%/etc/hosts none bind,ro,create=file 0 0
 lxc.mount.entry = %FS_PREFIX%/global/users/%USERNAME%/ssh %ROOTFS%/root/.ssh none bind,ro,create=dir 0 0
 lxc.mount.entry = %FS_PREFIX%/local/temp/%LXCNAME%/ %ROOTFS%/tmp none bind,rw,create=dir 0 0
 

src/master/taskmgr.py

@@ -1,7 +1,7 @@
 import threading
 import time
 import string
-import random, copy
+import random, copy, subprocess
 import json
 from functools import wraps
 
@@ -48,10 +48,30 @@ def __lt__(self, other):
         return self.priority < other.priority
 
     def gen_ips_from_base(self,base_ip):
+        if self.task_base_ip == None:
+            return
         self.ips = []
         for  i in range(task.max_size):
             self.ips.append(int_to_ip(base_ip + self.task_base_ip + i + 2))
 
+    def gen_hosts(self):
+        username = self.taskinfo.username
+        taskid = self.taskinfo.taskid
+        logger.info("Generate hosts for user(%s) task(%s) base_ip(%s)"%(username,taskid,str(self.task_base_ip)))
+        fspath = env.getenv('FS_PREFIX')
+        if not os.path.isdir("%s/global/users/%s" % (fspath,username)):
+            path = env.getenv('DOCKLET_LIB')
+            subprocess.call([path+"/master/userinit.sh", username])
+            logger.info("user %s directory not found, create it" % username)
+
+        hosts_file = open("%s/global/users/%s/%s.hosts" % (fspath,username,"batch-"+taskid),"w")
+        hosts_file.write("127.0.0.1 localhost\n")
+        i = 0
+        for ip in self.ips:
+            hosts_file.write(ip+" batch-"+str(i)+"\n")
+            i += 1
+        hosts_file.close()
+
     def get_one_resources_need(self):
         return self.vnodeinfo.vnode.instance
 
@@ -225,18 +245,20 @@ def task_processor(self, task, vnodes_workers):
 
         self.acquire_task_net(task)
         task.gen_ips_from_base(self.base_ip)
+        task.gen_hosts()
         #need to create hosts
         [success, gwip] = self.setup_tasknet(task,[w[1] for w in vnodes_workers])
         if not success:
+            self.release_task_ips(task)
             return [False, gwip]
-
         token = ''.join(random.sample(string.ascii_letters + string.digits, 8))
         placed_workers = []
 
         # start vc
         for vid, worker in vnodes_workers:
             vnodeinfo = copy.copy(task.vnodeinfo)
             vnodeinfo.vnodeid = vid
+            vnodeinfo.vnode.hostname = "batch-"+str(vid%task.max_size)
             vnode = task.subtask_list[vid]
             vnode['status'] = RUNNING
             vnode['try_count'] += 1
@@ -253,7 +275,7 @@ def task_processor(self, task, vnodes_workers):
             ipaddr = task.ips[vid%task.max_size]
             brname = "docklet-batch-%s-%s"%(username, taskid)
             networkinfo = Network(ipaddr=ipaddr, gateway=gwip, masterip=self.masterip, brname=brname)
-            vnode.network = networkinfo
+            vnodeinfo.vnode.network = networkinfo
 
             try:
                 self.logger.info('[task_processor] starting vnode for task [%s] instance [%d]' % (task.vnodeinfo.id, vid))

src/protos/rpc_pb2.py

@@ -178,9 +178,10 @@ def start_vnode(self, request, context):
         gateway = request.vnode.network.gateway
         brname = request.vnode.network.brname
         masterip = request.vnode.network.masterip
+        hostname = request.vnode.hostname
 
         #create container
-        [success, msg] = self.create_container(taskid, vnodeid, username, image, lxcname, instance_type, ipaddr, gateway, brname)
+        [success, msg] = self.create_container(taskid, vnodeid, username, image, lxcname, instance_type, ipaddr, gateway, brname, hostname)
         if not success:
             return rpc_pb2.Reply(status=rpc_pb2.Reply.REFUSED, message=msg)
 
@@ -285,7 +286,7 @@ def stop_vnode(self, request, context):
 
 
     #accquire ip and create a container
-    def create_container(self,taskid,vnodeid,username,image,lxcname,quota,ipaddr,gateway,brname):
+    def create_container(self,taskid,vnodeid,username,image,lxcname,quota,ipaddr,gateway,brname,hostname):
         # prepare image and filesystem
         status = self.imgmgr.prepareFS(username,image,lxcname,str(quota.disk))
         if not status:
@@ -302,7 +303,8 @@ def create_container(self,taskid,vnodeid,username,image,lxcname,quota,ipaddr,gat
 
         def config_prepare(content):
             content = content.replace("%ROOTFS%",rootfs)
-            content = content.replace("%HOSTNAME%","batch-%s" % str(vnodeid))
+            content = content.replace("%HOSTNAME%",hostname)
+            content = content.replace("%TASKID%",taskid)
             content = content.replace("%CONTAINER_MEMORY%",str(quota.memory))
             content = content.replace("%CONTAINER_CPU%",str(quota.cpu*100000))
             content = content.replace("%FS_PREFIX%",self.fspath)