Merge pull request #370 from FirmlyReality/batch

Batch

bin/docklet-master

@@ -32,6 +32,8 @@ WEB_PORT=8888
 USER_PORT=9100
 #cluster net, default is 172.16.0.1/16
 CLUSTER_NET="172.16.0.1/16"
+# ip addresses range of containers for batch job, default is 10.0.3.0/24
+BATCH_NET="10.0.3.0/24"
 #configurable-http-proxy public port, default is 8000
 PROXY_PORT=8000
 #configurable-http-proxy api port, default is 8001
@@ -103,6 +105,7 @@ pre_start_master () {
     # iptables for NAT network for containers to access web
     iptables -t nat -F
     iptables -t nat -A POSTROUTING -s $CLUSTER_NET -j MASQUERADE
+    iptables -t nat -A POSTROUTING -s $BATCH_NET -j MASQUERADE
 
 }
 

bin/docklet-supermaster

@@ -36,6 +36,8 @@ WEB_PORT=8888
 USER_PORT=9100
 #cluster net, default is 172.16.0.1/16
 CLUSTER_NET="172.16.0.1/16"
+# ip addresses range of containers for batch job, default is 10.0.3.0/24
+BATCH_NET="10.0.3.0/24"
 
 . $DOCKLET_CONF/docklet.conf
 
@@ -103,6 +105,7 @@ pre_start_master () {
     # iptables for NAT network for containers to access web
     iptables -t nat -F
     iptables -t nat -A POSTROUTING -s $CLUSTER_NET -j MASQUERADE
+    iptables -t nat -A POSTROUTING -s $BATCH_NET -j MASQUERADE
 
 }
 

conf/container.batch.conf

@@ -27,6 +27,7 @@ lxc.network.link = lxcbr0
 lxc.network.flags = up
 lxc.network.ipv4 = %IP%
 lxc.network.ipv4.gateway = %GATEWAY%
+lxc.network.mtu = 1420
 
 lxc.cgroup.pids.max = 2048
 lxc.cgroup.memory.limit_in_bytes = %CONTAINER_MEMORY%M

src/worker/taskcontroller.py

@@ -16,6 +16,7 @@ import grpc
 #from utils import env
 import json,lxc,subprocess,threading,os,time,traceback
 from utils import imagemgr,etcdlib,gputools
+from utils.lvmtool import sys_run
 from worker import ossmounter
 from protos import rpc_pb2, rpc_pb2_grpc
 
@@ -272,8 +273,11 @@ class TaskController(rpc_pb2_grpc.WorkerServicer):
             conffile.write("\n"+ mount_str % (self.fspath, username, mount.remotePath, rootfs, mount.remotePath))
         conffile.close()
 
-        container = lxc.Container(lxcname)
+
-        if not container.start():
+        logger.info("Start container %s..." % lxcname)
+        #container = lxc.Container(lxcname)
+        ret = subprocess.run('lxc-start -n %s'%lxcname,stdout=subprocess.PIPE,stderr=subprocess.STDOUT, shell=True)
+        if ret.returncode != 0:
             logger.error('start container %s failed' % lxcname)
             self.release_ip(ip)
             self.imgmgr.deleteFS(lxcname)

web/templates/batch/batch_output.html

@@ -53,7 +53,7 @@
       var host = window.location.host;
       url = "//" + host + "/batch/job/output/" + "{{ masterip }}" + "/" + "{{ jobid }}" + "/" + "{{ taskid }}" + "/" + "{{ instid }}" + "/" + "{{ issue }}" + "/";
       $.post(url,{},function(data){
-        $("#output").html(data.data);
+        $("#output").text(String(data.data));
       },"json");
     }
     setInterval(updateOutput,2000);