use the mesos slave id as swan agent id

agent/agent.go

@@ -5,10 +5,12 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"io/ioutil"
 	"net"
 	"net/http"
 	"net/url"
 	"os"
+	"strings"
 	"time"
 
 	log "github.com/Sirupsen/logrus"
@@ -171,11 +173,14 @@ func (agent *Agent) Join() error {
 		return err
 	}
 
+	// detect mesos slave id
+	id, err := agent.detectMesosSlaveID(masterURL)
+	if err != nil {
+		return err
+	}
+
 	// setup & join
-	agent.clusterNode = mole.NewAgent(&mole.Config{
-		Role:   mole.RoleAgent,
-		Master: masterURL,
-	})
+	agent.clusterNode = mole.NewAgent(id, masterURL)
 
 	return agent.clusterNode.Join()
 }
@@ -257,3 +262,55 @@ func (agent *Agent) detectLeaderAddr() (string, error) {
 
 	return "", errors.New("all of swan manager unavailable")
 }
+
+// try best to detect the mesos slave id running on the same host
+// by querying against to the swan master
+func (agent *Agent) detectMesosSlaveID(masterAddr *url.URL) (string, error) {
+	// obtian from env
+	if env := os.Getenv("MESOS_SALVE_ID"); env != "" {
+		return env, nil
+	}
+
+	// obtain from remote swan master by query ip addresses
+
+	var (
+		queryIPs []string
+		queryURL = masterAddr.String() + "/v1/agents/query_id?ips="
+	)
+
+	if env := os.Getenv("MESOS_SLAVE_IPS"); env != "" {
+		// obtain local ips from env
+		queryIPs = strings.Split(env, ",")
+
+	} else {
+		// obtain local ips from sysinfo
+		info, err := Gather()
+		if err != nil {
+			return "", err
+		}
+		for inet, ips := range info.IPs {
+			if inet == "docker0" {
+				continue
+			}
+			queryIPs = append(queryIPs, ips...)
+		}
+	}
+
+	// query id against swan master
+	resp, err := http.Get(queryURL + strings.Join(queryIPs, ","))
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+
+	body, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		return "", err
+	}
+
+	if resp.StatusCode == http.StatusOK {
+		return string(body), nil
+	}
+
+	return "", fmt.Errorf("query id against master got %d - %s", resp.StatusCode, string(body))
+}

api/cluster.go

@@ -7,6 +7,7 @@ import (
 	"io"
 	"io/ioutil"
 	"net/http"
+	"strings"
 
 	"github.com/Dataman-Cloud/swan/types"
 	log "github.com/Sirupsen/logrus"
@@ -26,6 +27,38 @@ func (r *Server) listAgents(w http.ResponseWriter, req *http.Request) {
 	writeJSON(w, http.StatusOK, ret)
 }
 
+func (r *Server) queryAgentID(w http.ResponseWriter, req *http.Request) {
+	if err := req.ParseForm(); err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	var (
+		ips = req.Form.Get("ips")
+	)
+	if ips == "" {
+		http.Error(w, "query parameter ips required", http.StatusBadRequest)
+		return
+	}
+
+	state, err := r.driver.MesosState()
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	for _, slave := range state.Slaves {
+		for _, ip := range strings.Split(ips, ",") {
+			if ip == slave.Hostname {
+				w.Write([]byte(slave.ID))
+				return
+			}
+		}
+	}
+
+	http.Error(w, "not found matched mesos slaves", http.StatusNotFound)
+}
+
 func (r *Server) getAgent(w http.ResponseWriter, req *http.Request) {
 	var (
 		id = mux.Vars(req)["agent_id"]
@@ -61,22 +94,22 @@ func (r *Server) getAgentConfigs(w http.ResponseWriter, req *http.Request) {
 }
 
 func (r *Server) redirectAgentDocker(w http.ResponseWriter, req *http.Request) {
-	n := len(`/v1/agents/docker/`) + 16
+	n := len(`/v1/agents/docker/`) + 39 // FIX LATER
 	r.redirectAgent(n, w, req)
 }
 
 func (r *Server) redirectAgentProxy(w http.ResponseWriter, req *http.Request) {
-	n := len(`/v1/agents/`) + 16
+	n := len(`/v1/agents/`) + 39
 	r.redirectAgent(n, w, req)
 }
 
 func (r *Server) redirectAgentDNS(w http.ResponseWriter, req *http.Request) {
-	n := len(`/v1/agents/`) + 16
+	n := len(`/v1/agents/`) + 39
 	r.redirectAgent(n, w, req)
 }
 
 func (r *Server) redirectAgentIPAM(w http.ResponseWriter, req *http.Request) {
-	n := len(`/v1/agents/`) + 16
+	n := len(`/v1/agents/`) + 39
 	r.redirectAgent(n, w, req)
 }
 

api/driver.go

@@ -6,6 +6,7 @@ import (
 	"github.com/Dataman-Cloud/swan/mesos"
 	"github.com/Dataman-Cloud/swan/mole"
 	"github.com/Dataman-Cloud/swan/types"
+	"github.com/andygrunwald/megos"
 )
 
 type Driver interface {
@@ -22,6 +23,8 @@ type Driver interface {
 	ClusterAgent(id string) *mole.ClusterAgent
 	CloseClusterAgent(id string)
 
+	MesosState() (*megos.State, error)
+
 	// for debug convenience
 	Dump() interface{}
 	Load() map[string]interface{}

api/route_test.go

@@ -1,5 +1,6 @@
 package api
 
+/*
 import (
 	"reflect"
 	"testing"
@@ -199,3 +200,4 @@ func TestNewPrefixRoute(t *testing.T) {
 		})
 	}
 }
+*/

api/router.go

@@ -64,6 +64,7 @@ func (s *Server) setupRoutes(mux *mux.Router) {
 		NewRoute("DELETE", "/v1/debug", s.disableDebug),
 
 		NewRoute("GET", "/v1/agents", s.listAgents),
+		NewRoute("GET", "/v1/agents/query_id", s.queryAgentID),
 		NewRoute("GET", "/v1/agents/{agent_id}", s.getAgent),
 		NewRoute("DELETE", "/v1/agents/{agent_id}", s.closeAgent),
 		NewRoute("GET", "/v1/agents/{agent_id}/sysinfo", s.getAgent),

api/server_test.go

@@ -1,5 +1,6 @@
 package api
 
+/*
 import (
 	"net"
 	"net/http"
@@ -425,3 +426,4 @@ func TestServer_forwardRequest(t *testing.T) {
 		})
 	}
 }
+*/

api/utils_test.go

@@ -1,5 +1,6 @@
 package api
 
+/*
 import (
 	"io"
 	"net/http"
@@ -80,3 +81,4 @@ func Test_decode(t *testing.T) {
 		})
 	}
 }
+*/

docker-compose.yml

@@ -105,6 +105,7 @@ services:
       - SWAN_JOIN_ADDRS=127.0.0.1:9999
       - SWAN_LOG_LEVEL=debug
       - SWAN_START_DELAY=5
+      - MESOS_SLAVE_IPS=127.0.0.1   # should keep same as mesos-slave.MESOS_HOSTNAME
     healthcheck:
       test: ["CMD", "wget", "-s", "-q", "http://127.0.0.1:10000/ping"]
       interval: 5s

docs/api.md

@@ -48,6 +48,7 @@
 
 + agents
   - [GET /v1/agents](#list-agents) *List all agents*
+  - [GET /v1/agents/query_id](#query-agent-id) *Query mesos slave id by ip addresses (internal use)*
   - [GET /v1/agents/{agent_id}](#get-agent) *Get specified agent*
   - [DELETE /v1/agents/{agent_id}](#close-agent) *Disconnect specified agent*
   - [GET /v1/agents/{agent_id}/dns](#get-agent-dns) *Get dns records on specified agent*
@@ -1448,13 +1449,13 @@ swan上可以直接请求 节点上的 [Docker Remote API 1.21](https://docs.doc
 
 Example:
 ```
-/v1/agents/3981314045636649/docker/containers/json
+/v1/agents/212c92eb-f594-43d5-89da-7820a56e8570-S0/docker/containers/json
 
-/v1/agents/3981314045636649/docker/containers/54b701f325fe1f229dd9174fb90123057e933d2f564d4e5f90e0a69ee6461770/json
+/v1/agents/212c92eb-f594-43d5-89da-7820a56e8570-S0/docker/containers/54b701f325fe1f229dd9174fb90123057e933d2f564d4e5f90e0a69ee6461770/json
 
-/v1/agents/3981314045636649/docker/images/json
+/v1/agents/212c92eb-f594-43d5-89da-7820a56e8570-S0/docker/images/json
 
-/v1/agents/3981314045636649/docker/networks
+/v1/agents/212c92eb-f594-43d5-89da-7820a56e8570-S0/docker/networks
 ```
 
 #### Reset
@@ -1476,15 +1477,14 @@ Example response:
 }
 ```
 
-
 #### list agents
 ```
 GET /v1/agents
 ```
 
 ```json
 {
-  "3264208446845635": {
+  "212c92eb-f594-43d5-89da-7820a56e8570-S0": {
     "hostname": "master",
     "os": "NAME=\"CentOS Linux\"\nVERSION=\"7 (Core)\"\nID=\"centos\"\nID_LIKE=\"rhel fedora\"\nVERSION_ID=\"7\"\nPRETTY_NAME=\"CentOS Linux 7 (Core)\"\nANSI_COLOR=\"0;31\"\nCPE_NAME=\"cpe:/o:centos:centos:7\"\nHOME_URL=\"https://www.centos.org/\"\nBUG_REPORT_URL=\"https://bugs.centos.org/\"\n\nCENTOS_MANTISBT_PROJECT=\"CentOS-7\"\nCENTOS_MANTISBT_PROJECT_VERSION=\"7\"\nREDHAT_SUPPORT_PRODUCT=\"centos\"\nREDHAT_SUPPORT_PRODUCT_VERSION=\"7\"\n\n",
     "uptime": "34909.000000",
@@ -1550,7 +1550,7 @@ GET /v1/agents
       2181
     ]
   },
-  "3981314045636649": {
+  "212c92eb-f594-43d5-89da-7820a56e8570-S1": {
     "hostname": "node1",
     "os": "NAME=\"CentOS Linux\"\nVERSION=\"7 (Core)\"\nID=\"centos\"\nID_LIKE=\"rhel fedora\"\nVERSION_ID=\"7\"\nPRETTY_NAME=\"CentOS Linux 7 (Core)\"\nANSI_COLOR=\"0;31\"\nCPE_NAME=\"cpe:/o:centos:centos:7\"\nHOME_URL=\"https://www.centos.org/\"\nBUG_REPORT_URL=\"https://bugs.centos.org/\"\n\nCENTOS_MANTISBT_PROJECT=\"CentOS-7\"\nCENTOS_MANTISBT_PROJECT_VERSION=\"7\"\nREDHAT_SUPPORT_PRODUCT=\"centos\"\nREDHAT_SUPPORT_PRODUCT_VERSION=\"7\"\n\n",
     "uptime": "34906.000000",
@@ -1592,6 +1592,15 @@ GET /v1/agents
 }
 ```
 
+#### query agent id
+```
+GET /v1/agents/query_id?ips=192.168.1.196,192.168.1.130,xxx
+```
+
+```plain
+212c92eb-f594-43d5-89da-7820a56e8570-S0
+```
+
 #### get agent
 ```
 GET /v1/agents/{agent_id}

integration-test/swan_api_canary_test.go

@@ -34,6 +34,7 @@ func (s *ApiSuite) TestCanaryUpdate(c *check.C) {
 
 	// verify proxy record
 	proxy := s.listAppProxies(id, c)
+	c.Assert(proxy, check.Not(check.IsNil))
 	c.Assert(proxy.Alias, check.Equals, "www.xxx.com")
 	c.Assert(len(proxy.Backends), check.Equals, 5)
 	c.Assert(proxy.Listen, check.Equals, "")
@@ -110,6 +111,7 @@ func (s *ApiSuite) TestCanaryUpdate(c *check.C) {
 	// verify proxy record
 	time.Sleep(time.Millisecond * 500)
 	proxy = s.listAppProxies(id, c)
+	c.Assert(proxy, check.Not(check.IsNil))
 	c.Assert(proxy.Alias, check.Equals, "www.xxx.com")
 	c.Assert(len(proxy.Backends), check.Equals, 5)
 	c.Assert(proxy.Listen, check.Equals, "")
@@ -196,6 +198,7 @@ func (s *ApiSuite) TestCanaryUpdate(c *check.C) {
 
 	// verify proxy record again
 	proxy = s.listAppProxies(id, c)
+	c.Assert(proxy, check.Not(check.IsNil))
 	c.Assert(proxy.Alias, check.Equals, "www.xxx.com")
 	c.Assert(len(proxy.Backends), check.Equals, 5)
 	c.Assert(proxy.Listen, check.Equals, "")

integration-test/swan_api_create_proxy.go

@@ -50,6 +50,7 @@ func (s *ApiSuite) TestCreateAppProxy(c *check.C) {
 
 	// verify proxy record
 	proxy := s.listAppProxies(id, c)
+	c.Assert(proxy, check.Not(check.IsNil))
 	c.Assert(proxy.Alias, check.Equals, "www.xxx.com")
 	c.Assert(len(proxy.Backends), check.Equals, 10)
 	c.Assert(proxy.Listen, check.Equals, "")

integration-test/swan_api_rollback_test.go

@@ -51,6 +51,7 @@ func (s *ApiSuite) TestRollBackApp(c *check.C) {
 
 	// verify proxy record
 	proxy := s.listAppProxies(id, c)
+	c.Assert(proxy, check.Not(check.IsNil))
 	c.Assert(proxy.Alias, check.Equals, "www.xxx.com")
 	c.Assert(len(proxy.Backends), check.Equals, 3)
 	c.Assert(proxy.Listen, check.Equals, "")
@@ -101,6 +102,7 @@ func (s *ApiSuite) TestRollBackApp(c *check.C) {
 
 	// verify proxy record
 	proxy = s.listAppProxies(id, c)
+	c.Assert(proxy, check.Not(check.IsNil))
 	c.Assert(proxy.Alias, check.Equals, "www.xxx.com")
 	c.Assert(len(proxy.Backends), check.Equals, 3)
 	c.Assert(proxy.Listen, check.Equals, "")
@@ -150,6 +152,7 @@ func (s *ApiSuite) TestRollBackApp(c *check.C) {
 
 	// verify proxy record
 	proxy = s.listAppProxies(id, c)
+	c.Assert(proxy, check.Not(check.IsNil))
 	c.Assert(proxy.Alias, check.Equals, "www.xxx.com")
 	c.Assert(len(proxy.Backends), check.Equals, 3)
 	c.Assert(proxy.Listen, check.Equals, "")