Merge "Initial Support for Health Check service"

src/ifmap/ifmap_graph_walker.cc

@@ -370,6 +370,7 @@ void IFMapGraphWalker::AddNodesToWhitelist() {
     traversal_white_list_->include_vertex.insert("loadbalancer-member");
     traversal_white_list_->include_vertex.insert("loadbalancer-healthmonitor");
     traversal_white_list_->include_vertex.insert("subnet");
+    traversal_white_list_->include_vertex.insert("service-health-check");
 }
 
 void IFMapGraphWalker::AddLinksToWhitelist() {
@@ -473,5 +474,7 @@ void IFMapGraphWalker::AddLinksToWhitelist() {
     // getting the pool from. EG: public-network (might not have any VMs)
     traversal_white_list_->include_edge.insert(
         "source=floating-ip-pool,target=virtual-network");
+    traversal_white_list_->include_edge.insert(
+        "source=virtual-machine-interface,target=service-health-check");
 }
 

src/vnsw/agent/cmn/agent.cc

@@ -28,6 +28,8 @@
 #include <oper/operdb_init.h>
 #include <oper/config_manager.h>
 #include <oper/interface_common.h>
+#include <oper/health_check.h>
+#include <oper/metadata_ip.h>
 #include <oper/multicast.h>
 #include <oper/nexthop.h>
 #include <oper/mirror_table.h>
@@ -427,7 +429,7 @@ Agent::Agent() :
     instance_id_(g_vns_constants.INSTANCE_ID_DEFAULT),
     module_type_(Module::VROUTER_AGENT), db_(NULL),
     task_scheduler_(NULL), agent_init_(NULL), fabric_vrf_(NULL),
-    intf_table_(NULL),
+    intf_table_(NULL), health_check_table_(NULL), metadata_ip_allocator_(NULL),
     nh_table_(NULL), uc_rt_table_(NULL), mc_rt_table_(NULL), vrf_table_(NULL),
     vm_table_(NULL), vn_table_(NULL), sg_table_(NULL), mpls_table_(NULL),
     acl_table_(NULL), mirror_table_(NULL), vrf_assign_table_(NULL),
@@ -557,6 +559,22 @@ void Agent::set_flow_stats_manager(FlowStatsManager *aging_module) {
     flow_stats_manager_ = aging_module;
 }
 
+HealthCheckTable *Agent::health_check_table() const {
+    return health_check_table_;
+}
+
+void Agent::set_health_check_table(HealthCheckTable *table) {
+    health_check_table_ = table;
+}
+
+MetaDataIpAllocator *Agent::metadata_ip_allocator() const {
+    return metadata_ip_allocator_.get();
+}
+
+void Agent::set_metadata_ip_allocator(MetaDataIpAllocator *allocator) {
+    metadata_ip_allocator_.reset(allocator);
+}
+
 PktModule *Agent::pkt() const {
     return pkt_;
 }

src/vnsw/agent/cmn/agent.h

@@ -35,6 +35,7 @@ class AgentInit;
 class AgentStatsCollector;
 class FlowStatsCollector;
 class FlowStatsManager;
+class MetaDataIpAllocator;
 namespace OVSDB {
 class OvsdbClient;
 };
@@ -125,12 +126,18 @@ typedef boost::intrusive_ptr<const PhysicalDeviceVn> PhysicalDeviceVnConstRef;
 void intrusive_ptr_release(const PhysicalDeviceVn *p);
 void intrusive_ptr_add_ref(const PhysicalDeviceVn *p);
 
+class HealthCheckService;
+typedef boost::intrusive_ptr<HealthCheckService> HealthCheckServiceRef;
+void intrusive_ptr_release(const HealthCheckService* p);
+void intrusive_ptr_add_ref(const HealthCheckService* p);
+
 //class SecurityGroup;
 typedef std::vector<int> SecurityGroupList;
 typedef std::vector<std::string> CommunityList;
 
 class AgentDBTable;
 class InterfaceTable;
+class HealthCheckTable;
 class NextHopTable;
 class VmTable;
 class VnTable;
@@ -747,6 +754,12 @@ class Agent {
     FlowStatsManager *flow_stats_manager() const;
     void set_flow_stats_manager(FlowStatsManager *fsc);
 
+    HealthCheckTable *health_check_table() const;
+    void set_health_check_table(HealthCheckTable *table);
+
+    MetaDataIpAllocator *metadata_ip_allocator() const;
+    void set_metadata_ip_allocator(MetaDataIpAllocator *allocator);
+
     PktModule *pkt() const;
     void set_pkt(PktModule *pkt);
 
@@ -1002,6 +1015,8 @@ class Agent {
     AgentInit *agent_init_;
     VrfEntry *fabric_vrf_;
     InterfaceTable *intf_table_;
+    HealthCheckTable *health_check_table_;
+    std::auto_ptr<MetaDataIpAllocator> metadata_ip_allocator_;
     NextHopTable *nh_table_;
     InetUnicastAgentRouteTable *uc_rt_table_;
     Inet4MulticastAgentRouteTable *mc_rt_table_;

src/vnsw/agent/cmn/index_vector.h

@@ -58,6 +58,12 @@ class IndexVector {
             entries_.resize(index + kGrowSize);
         }
 
+        // TODO(prabhjot) need to enable assertion below
+        // currently disabled due to some issue with MPLS
+        // label allocation
+        // index should not be already in use
+        // assert(bitmap_[index] == 1);
+
         bitmap_.set(index, 0);
         entries_[index] = entry;
         return index;

src/vnsw/agent/contrail/SConscript

@@ -120,6 +120,8 @@ doc_files += env['AGENT_FLOW_DOC_FILES']
 
 # Please update sandesh/common/vns.sandesh on process name change
 env.Alias('install', env.Install(env['INSTALL_BIN'], contrail_vrouter_agent))
+env.Alias('install', env.Install(env['INSTALL_EXAMPLE'],
+          'contrail_vrouter_agent_health_check.py'))
 env.Alias('install', env.Install(env['INSTALL_CONF'],
           '../contrail-vrouter-agent.conf'))
 env.Alias('install',

src/vnsw/agent/contrail/contrail_vrouter_agent_health_check.py

@@ -0,0 +1,117 @@
+#!/usr/bin/python
+#
+# Copyright (c) 2016 Juniper Networks, Inc. All rights reserved.
+#
+
+import os
+import sys
+from threading import Event
+from subprocess import call
+from optparse import OptionParser
+
+class HealthCheckBase():
+    def __init__(self, event, ip, timer, timeout, retries, uri):
+        self.event = event
+        self.ip = ip
+        self.timer = timer
+        self.timeout = timeout
+        self.retries = retries
+        self.uri = uri;
+        self.retries_done = 0
+        self.initial = True
+        self.ret = 0
+        self.stop = False
+        self.dev_null_fd = open(os.devnull, 'w')
+
+    def start(self):
+        while (self.initial or (not self.event.wait(self.timer))):
+            if (self.stop == False):
+                ret = self.execute_util(self.ip, self.timeout, self.uri,
+                                        self.dev_null_fd)
+            self.retries_done += 1
+            if ret != 0 and self.retries_done != self.retries:
+                # ignore by restoring previous response,
+                # till number of retries are completed
+                ret = self.ret
+            else:
+                self.retries_done = 0
+                # TODO(prabhjot) need to add log for every transaction
+                if self.initial == True or ret != self.ret:
+                    # TODO(prabhjot) currently script uses failure to
+                    # write to stdout as a mechanism to identify parent
+                    # process close and exit, eventually we should
+                    # restore below line to optimize unneccessary message
+                    # exchange on the stdout pipe
+                    # self.initial = False
+                    self.ret = ret
+                    if self.ret == 0:
+                        sys.stdout.write("Success\n")
+                        sys.stdout.flush()
+                    else:
+                        sys.stdout.write("Failure\n")
+                        sys.stdout.flush()
+
+    def execute_util(self, ip, timeout, uri, out_fd):
+        sys.stderr.write("Error: health check method not defined!\n")
+        sys.stderr.flush()
+        # pause script and stop retrying
+        self.stop = True
+        self.retries = 1
+        return 1
+
+class HealthCheckPing(HealthCheckBase):
+    def execute_util(self, ip, timeout, uri, out_fd):
+        return call(["ping", "-c2", "-W" + str(timeout), ip], stdout=out_fd,
+                    stderr=out_fd)
+
+class HealthCheckHttp(HealthCheckBase):
+    def execute_util(self, ip, timeout, uri, out_fd):
+        return call(["curl", "-m" + str(timeout), "http://" + ip + "/" + uri],
+                    stdout=out_fd, stderr=out_fd)
+
+def HealthCheckService(x):
+    return {
+        "ping": HealthCheckPing,
+        "Ping": HealthCheckPing,
+        "PING": HealthCheckPing,
+        "http": HealthCheckHttp,
+        "Http": HealthCheckHttp,
+        "HTTP": HealthCheckHttp,
+    }.get(x, HealthCheckBase) #[x]
+
+parser = OptionParser()
+parser.add_option("-m", "--method", dest="method",
+                  help="method to do Health Check (Ping/Http)",
+                  metavar="METHOD")
+parser.add_option("-d", "--destip", dest="dest_ip",
+                  help="Destination IP to run Health Check", metavar="IP",
+                  default="")
+parser.add_option("-i", "--interval", dest="interval", type="int",
+                  help="Health Check interval in seconds", metavar="TIME",
+                  default="2")
+parser.add_option("-t", "--timeout", dest="timeout", type="int",
+                  help="Health Check timeout in seconds", metavar="TIME",
+                  default="1")
+parser.add_option("-r", "--retries", dest="retries", type="int",
+                  help="Number of retries to be done to declare error",
+                  metavar="COUNT", default="2")
+parser.add_option("-u", "--uri", dest="uri",
+                  help="Additional indentifier for Health Check object",
+                  metavar="STRING", default="")
+(option, args) = parser.parse_args()
+
+if option.interval < 1:
+    option.interval = 1
+
+if option.timeout < 1:
+    option.timeout = 1
+
+if option.retries < 1:
+    option.retries = 1
+
+timerEvent = Event()
+service = HealthCheckService(option.method)(timerEvent, option.dest_ip,
+                                            option.interval, option.timeout,
+                                            option.retries, option.uri)
+service.start()
+

src/vnsw/agent/oper/SConscript

@@ -37,6 +37,7 @@ vnswoperdb = env.Library('vnswoperdb',
                           'config_manager.cc',
                           'evpn_route.cc',
                           'global_vrouter.cc',
+                          'health_check.cc',
                           'ifmap_dependency_manager.cc',
                           'inet_interface.cc',
                           'inet4_multicast_route.cc',
@@ -47,6 +48,7 @@ vnswoperdb = env.Library('vnswoperdb',
                           'loadbalancer_pool.cc',
                           'loadbalancer_pool_info.cc',
                           'logical_interface.cc',
+                          'metadata_ip.cc',
                           'mirror_table.cc',
                           'mpls.cc',
                           'multicast.cc',

src/vnsw/agent/oper/agent.sandesh

@@ -103,6 +103,8 @@ struct ItfSandeshData {
    49: optional string ipv4_active;
    50: list<string> fixed_ip4_list;
    51: list<string> fixed_ip6_list;
+   52: optional string health_check_active;
+   53: optional string metadata_ip_active;
 }
 
 struct VnIpamData {
@@ -1326,3 +1328,39 @@ response sandesh IFMapNodePolicyResp {
 request sandesh IFMapNodePolicyReq {
     1: string node;
 }
+
+traceobject sandesh HealthCheckTrace {
+    1: string log;
+}
+
+struct HealthCheckInstanceSandeshData {
+    1: string vm_interface (link="ItfReq");
+    2: string metadata_ip;
+    3: string service_ip;
+    4: string health_check_ip;
+    5: bool active;
+    6: bool running;
+    7: string last_update_time;
+}
+
+struct HealthCheckSandeshData {
+    1: string uuid;
+    2: string name;
+    3: string monitor_type;
+    4: string http_method;
+    5: string url_path;
+    6: string expected_codes;
+    7: u32 delay;
+    8: u32 timeout;
+    9: u32 max_retries;
+    10: list<HealthCheckInstanceSandeshData> inst_list;
+}
+
+request sandesh HealthCheckSandeshReq {
+    1: string uuid;
+}
+
+response sandesh HealthCheckSandeshResp {
+    1: list<HealthCheckSandeshData> hc_list;
+}
+

src/vnsw/agent/oper/agent_sandesh.cc

@@ -10,6 +10,7 @@
 #include <oper/peer.h>
 #include <oper/vrf.h>
 #include <oper/interface_common.h>
+#include <oper/health_check.h>
 #include <oper/nexthop.h>
 #include <oper/vn.h>
 #include <oper/vm.h>
@@ -1058,3 +1059,36 @@ void VrouterObjectLimitsReq::HandleRequest() const {
    resp->set_vrouter_object_limit(vr_limits);
    resp->Response();
 }
+
+AgentHealthCheckSandesh::AgentHealthCheckSandesh(const std::string &context,
+                                                 const std::string &u) :
+        AgentSandesh(context, ""), uuid_str_(u) {
+    boost::system::error_code ec;
+    uuid_ = StringToUuid(u);
+}
+
+DBTable *AgentHealthCheckSandesh::AgentGetTable() {
+    return static_cast<DBTable *>(Agent::GetInstance()->health_check_table());
+}
+
+void AgentHealthCheckSandesh::Alloc() {
+    resp_ = new HealthCheckSandeshResp();
+}
+
+bool AgentHealthCheckSandesh::Filter(const DBEntryBase *entry) {
+    const HealthCheckService *service =
+        dynamic_cast<const HealthCheckService *>(entry);
+    assert(service);
+
+    if (MatchUuid(uuid_str_ , uuid_,  service->uuid()) == false)
+        return false;
+
+    return true;
+}
+
+bool AgentHealthCheckSandesh::FilterToArgs(AgentSandeshArguments *args) {
+    args->Add("uuid", uuid_str_);
+
+    return true;
+}
+

src/vnsw/agent/oper/agent_sandesh.h

@@ -458,4 +458,21 @@ class AgentLoadBalancerV2Sandesh : public AgentSandesh {
     void Alloc();
 };
 
+class AgentHealthCheckSandesh : public AgentSandesh {
+public:
+    AgentHealthCheckSandesh(const std::string &context, const std::string &u);
+    ~AgentHealthCheckSandesh() {}
+    virtual bool Filter(const DBEntryBase *entry);
+    virtual bool FilterToArgs(AgentSandeshArguments *args);
+
+private:
+    friend class VnListReq;
+    DBTable *AgentGetTable();
+    void Alloc();
+    std::string uuid_str_;
+
+    boost::uuids::uuid uuid_;
+
+};
+
 #endif // vnsw_agent_sandesh_h_