Skip to content

D
Distributed Computing and Network Fusion System
Commit 3b84df6c authored by wutu's avatar wutu
Browse files
Options

修复一个监控关闭失效的bug

parent 63dd43bf
Hide whitespace changes
Inline Side-by-side
Showing with 171 additions and 48 deletions
bishe-client-starter/src/main/java/top/ninwoo/bishe/starter/service/NetworkService.java
View file @ 3b84df6c
......@@ -20,5 +20,7 @@ public interface NetworkService {
String enableNetworkMonitor(Long clusterId, String appName);
String cancelNetworkMonitor(Long clusterId, String appName);
List<ContainerMonitor> getContainerMonitors(Long clusterId, String appName);
}
bishe-client-starter/src/main/java/top/ninwoo/bishe/starter/service/NetworkServiceImpl.java
View file @ 3b84df6c
package top.ninwoo.bishe.starter.service;
import lombok.extern.slf4j.Slf4j;
import org.springframework.core.ParameterizedTypeReference;
import org.springframework.http.HttpMethod;
import org.springframework.http.ResponseEntity;
import org.springframework.stereotype.Service;
import org.springframework.web.client.RestTemplate;
......@@ -10,6 +12,7 @@ import top.ninwoo.common.entity.ContainerMonitorInfo;
import top.ninwoo.common.entity.NetworkTopology;
import javax.annotation.Resource;
import java.awt.*;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
......@@ -24,9 +27,10 @@ public class NetworkServiceImpl implements NetworkService {
private final static String GET_IP_LIST_BY_APP_NAME = "/network/getIpListByAppName?clusterId={clusterId}&appName={appName}";
private final static String ENABLE_NETWORK_MONITOR = "/enableNetworkMonitor?clusterId={clusterId}&appName={appName}";
private final static String ENABLE_NETWORK_MONITOR = "/network/enableNetworkMonitor?clusterId={clusterId}&appName={appName}";
private final static String GET_NETWORK_MONITOR_INFO = "/getNetworkMonitorInfo?clusterId={clusterId}&appName={appName}";
private final static String CANCEL_NETWORK_MONITOR = "/network/cancelNetworkMonitor?clusterId={clusterId}&appName={appName}";
private final static String GET_NETWORK_MONITOR_INFO = "/network/getNetworkMonitorInfo?clusterId={clusterId}&appName={appName}";
@Resource
ClientProperties clientProperties;
......@@ -95,6 +99,18 @@ public class NetworkServiceImpl implements NetworkService {
return restTemplate.getForObject("http://" + cloudUrl + ENABLE_NETWORK_MONITOR, String.class, param);
}
@Override
public String cancelNetworkMonitor(Long clusterId, String appName) {
String cloudUrl = clientProperties.getCloudUrl();
Map<String, Object> param = new HashMap<>();
param.put("clusterId", clusterId);
param.put("appName", appName);
return restTemplate.getForObject("http://" + cloudUrl + CANCEL_NETWORK_MONITOR, String.class, param);
}
/**
* 获取容器监控的钩子程序
* @param clusterId
......@@ -110,12 +126,16 @@ public class NetworkServiceImpl implements NetworkService {
param.put("clusterId", clusterId);
param.put("appName", appName);
// 这里需要构建这个ContainerMonitor
List<ContainerMonitorInfo> containerMonitorInfos
= restTemplate.getForObject("http://" + cloudUrl + GET_NETWORK_MONITOR_INFO, List.class, param);
//List<ContainerMonitorInfo> containerMonitorInfos
// = restTemplate.getForObject("http://" + cloudUrl + GET_NETWORK_MONITOR_INFO, List.class, param);
ResponseEntity<List<ContainerMonitorInfo>> responseEntity
= restTemplate.exchange("http://" + cloudUrl + GET_NETWORK_MONITOR_INFO,
HttpMethod.GET, null, new ParameterizedTypeReference<List<ContainerMonitorInfo>>() {}, clusterId, appName);
List<ContainerMonitorInfo> containerMonitorInfos = responseEntity.getBody();
List<ContainerMonitor> containerMonitors = new ArrayList<>();
containerMonitorInfos.forEach(c -> {
containerMonitors.add(new ContainerMonitor(c, restTemplate));
});
for (ContainerMonitorInfo containerMonitorInfo : containerMonitorInfos) {
containerMonitors.add(new ContainerMonitor(containerMonitorInfo, restTemplate));
}
return containerMonitors;
}
}
\ No newline at end of file
bishe-cloud-center/src/main/java/top/ninwoo/cloudcenter/controller/NetworkController.java
View file @ 3b84df6c
......@@ -34,6 +34,11 @@ public class NetworkController {
return cloudService.enableNetworkMonitor(clusterId, appName);
}
@RequestMapping(value = "/cancelNetworkMonitor")
public String cancelNetworkMonitor(Long clusterId, String appName) {
return cloudService.cancelNetworkMonitor(clusterId, appName);
}
@RequestMapping(value = "/getNetworkMonitorInfo")
public List<ContainerMonitorInfo> getNetworkMonitor(Long clusterId, String appName) {
// 这里可以理解为是返回一个网络监控的钩子程序，具体的查询还需要通过钩子进行执行
......
bishe-cloud-center/src/main/java/top/ninwoo/cloudcenter/service/CloudService.java
View file @ 3b84df6c
......@@ -27,4 +27,6 @@ public interface CloudService {
String enableNetworkMonitor(Long clusterId, String appName);
List<ContainerMonitorInfo> getContainerMonitorInfoByAppName(Long clusterId, String appName);
String cancelNetworkMonitor(Long clusterId, String appName);
}
bishe-cloud-center/src/main/java/top/ninwoo/cloudcenter/service/impl/CloudServiceImpl.java
View file @ 3b84df6c
......@@ -24,7 +24,7 @@ public class CloudServiceImpl implements CloudService {
private static final String CANCEL_DROP_DOCKER_NETWORK = "/cancelDropDockerNetwork?clusterId={clusterId}&appName={appName}&ipList={ipList}";
private static final String REMOTE_IP_LIST_BY_APPNAME = "/getIpListByAppName?clusterId={clusterId}&appName={appName}";
private static final String ENABLE_NETWORK_MONITOR = "/enableContainerMonitor?clusterId={clusterId}&containerName={appName}";
private static final String CANCEL_NETWORK_MONITOR = "/cancelNetworkMonitor?clusterId={clusterId}&containerName={appName}";
@Value("${bs.ipservice.url}")
private String ipServiceUrl;
// 全部的逻辑拓扑
......@@ -33,10 +33,11 @@ public class CloudServiceImpl implements CloudService {
// 用于存储配置文件
private ConcurrentHashMap<Long, List<SeparatedClusterConfig>> allClusterConfig
= new ConcurrentHashMap<>();
// 存储app在哪一个边缘节点
private ConcurrentHashMap<Long, Map<String, Set<String>>> allAppStatus
= new ConcurrentHashMap<>();
private ConcurrentHashMap<Long, Map<String, Map<String, Set<String>>>> allContainerIds = new ConcurrentHashMap<>();
@Resource
private CloudRegisterCenter cloudRegisterCenter;
......@@ -225,12 +226,30 @@ public class CloudServiceImpl implements CloudService {
c.setEdgeNodeId(edgeNode);
}
// 下发配置
ResponseEntity<Long> response = restTemplate.postForEntity("http://" + c.getEdgeNodeId() + CREATE_CLUSTER, c.getClusterConfig(), Long.class);
ResponseEntity<Map> response = restTemplate.postForEntity("http://" + c.getEdgeNodeId() + CREATE_CLUSTER, c.getClusterConfig(), Map.class);
if(!response.getStatusCode().is2xxSuccessful()) {
LOG.error("集群配置下发失败:{}:{}", c.getEdgeNodeId(), response.getBody());
}
// 更新配置
//c.setClusterConfig(response.getBody());
// TODO: 存储appName和containerId的映射
if (!allContainerIds.containsKey(c.getClusterConfig().getId())) {
allContainerIds.put(c.getClusterConfig().getId(), new HashMap<>());
}
Map<String, Map<String, Set<String>>> allNames = allContainerIds.get(c.getClusterConfig().getId());
if (!allNames.containsKey(c.getEdgeNodeId())) {
allNames.put(c.getEdgeNodeId(), new HashMap<>());
}
Map<String, Set<String>> edgeContainerIds = allNames.get(c.getEdgeNodeId());
Map<String, List<String>> nameToIds = response.getBody();
nameToIds.forEach((name, v) -> {
if(!edgeContainerIds.containsKey(name)) {
edgeContainerIds.put(name, new HashSet<>());
}
edgeContainerIds.get(name).addAll(v);
});
// TODO: 对应需要设计一个删除这个配置的操作
// 构建初始化的逻辑拓扑
for (String appName : c.getClusterConfig().getTopology().getAppNames()) {
if(!appName.startsWith("br")) {
......@@ -428,32 +447,64 @@ public class CloudServiceImpl implements CloudService {
return allLogicalTopo.get(clusterId);
}
@Override
public String cancelNetworkMonitor(Long clusterId, String appName) {
if(!allAppStatus.containsKey(clusterId)) {
LOG.warn("正在取消一个不存在集群的网络监控");
return "failed";
}
if (!allAppStatus.get(clusterId).containsKey(appName)) {
LOG.warn("正在取消一个不存在的AppName的监控");
return "failed";
}
Set<String> nodeIds = allAppStatus.get(clusterId).get(appName);
// 遍历clusterAppName下发，容器监控模块
nodeIds.forEach(c -> {
if (!"".equals(c)) {
// 下发启用monitor监控的指令
Map<String, Object> param = new HashMap<>();
param.put("clusterId", clusterId);
param.put("appName", appName);
ResponseEntity<Boolean> response
= restTemplate.getForEntity("http://" + c + CANCEL_NETWORK_MONITOR, Boolean.class, param);
if(!response.getStatusCode().is2xxSuccessful()) {
LOG.warn("集群{}通信失败", c);
}
if(!response.getBody()) {
LOG.warn("集群{},appName{}指令下发失败", c, appName);
}
}
});
return "success";
}
@Override
public String enableNetworkMonitor(Long clusterId, String appName) {
if(!allClusterConfig.containsKey(clusterId)) {
if(!allAppStatus.containsKey(clusterId)) {
LOG.warn("正在启用一个不存在集群的网络监控");
return "failed";
}
if (!allClusterConfig.containsKey(appName)) {
if (!allAppStatus.get(clusterId).containsKey(appName)) {
LOG.warn("正在启用一个不存在的AppName的监控");
return "failed";
}
List<SeparatedClusterConfig> separatedClusterConfigs = allClusterConfig.get(appName);
Set<String> nodeIds = allAppStatus.get(clusterId).get(appName);
// 遍历clusterAppName下发，容器监控模块
separatedClusterConfigs.forEach(c -> {
if (c.getClusterConfig().getTopology().containsAppName(appName)) {
nodeIds.forEach(c -> {
if (!"".equals(c)) {
// 下发启用monitor监控的指令
Map<String, Object> param = new HashMap<>();
param.put("clusterId", clusterId);
param.put("appName", appName);
ResponseEntity<Boolean> response
= restTemplate.getForEntity("http://" + c.getEdgeNodeId() + ENABLE_NETWORK_MONITOR, Boolean.class, param);
= restTemplate.getForEntity("http://" + c + ENABLE_NETWORK_MONITOR, Boolean.class, param);
if(!response.getStatusCode().is2xxSuccessful()) {
LOG.warn("集群{}通信失败", c.getEdgeNodeId());
LOG.warn("集群{}通信失败", c);
}
if(!response.getBody()) {
LOG.warn("集群{},appName{}指令下发失败", c.getEdgeNodeId(), appName);
LOG.warn("集群{},appName{}指令下发失败", c, appName);
}
}
});
......@@ -468,19 +519,21 @@ public class CloudServiceImpl implements CloudService {
return containerMonitorInfos;
}
List<SeparatedClusterConfig> separatedClusterConfigs = allClusterConfig.get(clusterId);
if(!separatedClusterConfigs.contains(appName)) {
return containerMonitorInfos;
}
// TODO: 这里边没有需要的containerId
separatedClusterConfigs.forEach( c -> {
if(c.getClusterConfig().getTopology().containsAppName(appName)) {
for (ContainerDescription docker : c.getClusterConfig().getDockers()) {
ContainerMonitorInfo monitorInfo = new ContainerMonitorInfo();
monitorInfo.setAppName(appName);
monitorInfo.setEdgeNodeId(c.getEdgeNodeId());
monitorInfo.setClusterId(clusterId);
monitorInfo.setContainerId(docker.getDockerContainer().getId());
containerMonitorInfos.add(monitorInfo);
// 从找到对应容器id
Set<String> cids = allContainerIds.get(clusterId).get(c.getEdgeNodeId()).get(appName);
cids.forEach( cid -> {
ContainerMonitorInfo monitorInfo = new ContainerMonitorInfo();
monitorInfo.setAppName(appName);
monitorInfo.setEdgeNodeId(c.getEdgeNodeId());
monitorInfo.setClusterId(clusterId);
monitorInfo.setContainerId(cid);
containerMonitorInfos.add(monitorInfo);
});
}
}
});
......
bishe-edge-center/src/main/java/top/ninwoo/edgecenter/controller/IndexController.java
View file @ 3b84df6c
......@@ -85,12 +85,15 @@ public class IndexController {
return false;
}
cids.forEach(cid -> {
// 为每一个cid容器创建一个定时任务
NetworkMonitorThread networkMonitorThread = new NetworkMonitorThread(cid, containerInfoMap, clusterService);
// 放入线程池执行任务
ScheduledFuture<?> future = scheduledExecutorService.scheduleAtFixedRate(networkMonitorThread, 0, 1, TimeUnit.SECONDS);
// 保存任务，以方便进行定时任务的关闭
scheduledFutureMap.put(cid, future);
// 为每一个cid容器创建一个定时任务 TODO 应该加一个判断，防止重复提交
if (!scheduledFutureMap.containsKey(cid)) {
containerInfoMap.put(cid, new ContainerInfo());
NetworkMonitorThread networkMonitorThread = new NetworkMonitorThread(cid, containerInfoMap, clusterService);
// 放入线程池执行任务
ScheduledFuture<?> future = scheduledExecutorService.scheduleAtFixedRate(networkMonitorThread, 0, 1, TimeUnit.SECONDS);
// 保存任务，以方便进行定时任务的关闭
scheduledFutureMap.put(cid, future);
}
});
return true;
......@@ -105,9 +108,16 @@ public class IndexController {
}
cids.forEach(cid -> {
scheduledFutureMap.get(cid).cancel(true);
scheduledFutureMap.remove(cid);
containerInfoMap.remove(cid);
// TODO: 不安全的操作
// cancel不安全，get(cid)不安全
if(scheduledFutureMap.containsKey(cid)) {
scheduledFutureMap.get(cid).cancel(false);
while (!scheduledFutureMap.get(cid).isCancelled()) {
System.out.println("等待线程停止" + cid);
}
scheduledFutureMap.remove(cid);
containerInfoMap.remove(cid);
}
});
return true;
......@@ -115,13 +125,13 @@ public class IndexController {
@RequestMapping(value = "/createCluster", method = RequestMethod.POST)
public long createCluster(@RequestBody ClusterConfig clusterConfig) {
public Map<String, Set<String>> createCluster(@RequestBody ClusterConfig clusterConfig) {
// 创建一个解析器
long l = clusterService.initContainers(clusterConfig);
Map<String, Set<String>> resMap = clusterService.initContainers(clusterConfig);
// 创建网络
// 使用topo创建工具
topologyService.createTopo(l, clusterConfig.getTopology());
return l;
topologyService.createTopo(clusterConfig.getId(), clusterConfig.getTopology());
return resMap;
}
@RequestMapping(value = "/getContainerIds")
......@@ -150,7 +160,10 @@ public class IndexController {
containerInfo.setId(containerId);
containerInfo.setNetworkInfo(networkInfo);
containerInfo.setMem(mem);
map.put(containerId, containerInfo);
if(map.containsKey(containerId)) {
map.put(containerId, containerInfo); // TODO: 这里有又重新把容器ｉｄ注册回去了
}
System.out.println("监控运行中" + containerId);
}
public void setContainerId(String containerId) {
......
bishe-edge-center/src/main/java/top/ninwoo/edgecenter/service/ClusterService.java
View file @ 3b84df6c
......@@ -6,6 +6,7 @@ import top.ninwoo.utils.entity.NetworkInfo;
import top.ninwoo.utils.entity.OvsBridge;
import java.util.List;
import java.util.Map;
import java.util.Set;
/**
......@@ -19,7 +20,7 @@ public interface ClusterService {
List<DockerContainer> getContainerIds(boolean isAll);
// 创建容器组
long initContainers(ClusterConfig clusterConfig);
Map<String, Set<String>> initContainers(ClusterConfig clusterConfig);
// 动态调整容器规模，涉及到网络的一个调整
String adjustCluster(ClusterConfig clusterConfig);
......
bishe-edge-center/src/main/java/top/ninwoo/edgecenter/service/impl/ClusterServiceImpl.java
View file @ 3b84df6c
......@@ -87,7 +87,7 @@ public class ClusterServiceImpl implements ClusterService {
* @return
*/
@Override
public long initContainers(ClusterConfig clusterConfig) {
public Map<String, Set<String>> initContainers(ClusterConfig clusterConfig) {
// 校验ClusterConfig是否符合要求
if(clusterConfig.getId() == 0L) {
// 需要保存到数据库中，获取全局唯一的集群id
......@@ -125,7 +125,7 @@ public class ClusterServiceImpl implements ClusterService {
}
});
clusterConfig.setCreateTime(new Date());
return clusterConfig.getId();
return clustersInfo.get(clusterConfig.getId());
}
// 动态调整容器规模，涉及到网络的一个调整
......
bishe-edge-center/src/test/java/top/ninwoo/edgecenter/ClusterServiceTest.java
View file @ 3b84df6c
......@@ -58,8 +58,8 @@ public class ClusterServiceTest {
clusterConfig.setDockers(cds);
long id = clusterService.initContainers(clusterConfig);
Assert.assertEquals(id, clusterId);
//long id = clusterService.initContainers(clusterConfig).getId();
//Assert.assertEquals(id, clusterId);
Assert.assertTrue(clusterService.getClusterIds().contains(clusterId));
System.out.println(clusterService.getContainerIdsByClusterId(clusterId, "APP"));
......
bishe-test/src/test/java/top/ninwoo/BisheTests.java
View file @ 3b84df6c
......@@ -5,6 +5,7 @@ import org.junit.runner.RunWith;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.boot.test.context.SpringBootTest;
import org.springframework.test.context.junit4.SpringRunner;
import top.ninwoo.bishe.starter.entity.ContainerMonitor;
import top.ninwoo.bishe.starter.service.ClusterService;
import top.ninwoo.bishe.starter.service.NetworkService;
import top.ninwoo.common.entity.*;
......@@ -85,4 +86,29 @@ public class BisheTests {
List<String> ipList = networkService.getIpListByAppName(11111L, "Run");
System.out.println(ipList);
}
@Test
public void enableNetworkMonitorTest() throws InterruptedException {
String res = networkService.enableNetworkMonitor(11111L, "Run");
System.out.println(res);
}
@Test
public void cancelNetworkMonitorTest() {
String run = networkService.cancelNetworkMonitor(11111L, "Run");
System.out.println(run);
}
@Test
public void getNetworkMonitorTest() throws InterruptedException {
List<ContainerMonitor> run = networkService.getContainerMonitors(11111L, "Run");
int i = 0;
while(i < 10) {
run.forEach(c -> {
System.out.println(c.getContainerInfo());
});
Thread.sleep(500);
i++;
}
}
}
bishe-utils/src/main/java/top/ninwoo/utils/service/impl/DockerServiceImpl.java
View file @ 3b84df6c
......@@ -98,7 +98,8 @@ public class DockerServiceImpl implements DockerService, InitializingBean {
if("".equals(cid)) {
throw new RuntimeException("不存在这个容器名称[" + name + "]");
}
dockerUtils.deleteDockerById(name);
// TODO: 待测试
dockerUtils.deleteDockerById(cid);
return cid;
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment