.net core microservices 架构之 分布式
一:简介
自动计算都是常驻内存的,没有人机交互。我们经常用到的就是console job和sql job了。sqljob有自己的宿主,与数据库产品有很关联,暂时不提。console job使用quartz.net框架,目前3.x已经支持netcore。
如果单台服务器运行计算,那就很简单了,quartz很强大了,而且支持故障灾难转移集群,docker做宿主,很容易实现。但是分布式就不可同日而语了。如果你的一个数据处理太慢,需要多进程多主机处理,那么就需要多进程自动协调处理这一数据,比如如果你的订单太多,而一个进程处理延迟是10秒,那用户体验就会非常不好,虽然异步已经提高了你的吞吐量,但是延迟太久,对后续业务还是造成很大的干扰,时不时的会进行停顿。如果两到三台的机器进行处理,那延迟就会大大的减低,但是这两到三台服务器如果分配处理任务?如何分割这个数据,多进程进行处理?就需要到这一篇讲解的知识了。
在多个job应用之间进行协调的工具,就是zookeeper了,zookeeper官方介绍:一个分布式应用协调服务。其实他也是一个类似文件系统的写一致的数据存储软件,我们可以使用它做分布式锁,对应用进行协调控制。
目前流行的这一类产品也比较多,但是我是熟悉quartz,至于切片的功能,在quartz之上可以进行封装,因为所要封装的功能不多,所以我还是选择了quartz。
二 zookeeper 服务
首先就是zookeeper服务,和前面log日志一样,首先创建构建和配置文件类。
首先看看配置类:
ZookeeperConfiguration.cs
using Microsoft.Extensions.Configuration; namespace Walt.Framework.Configuration { public class ZookeeperConfiguration { public IConfiguration Configuration { get; } public ZookeeperConfiguration(IConfiguration configuration) { Configuration = configuration; } } }
using Microsoft.Extensions.Configuration; using Microsoft.Extensions.DependencyInjection; using Microsoft.Extensions.DependencyInjection.Extensions; using Microsoft.Extensions.Options; using Walt.Framework.Service.Zookeeper; namespace Walt.Framework.Configuration { public static class ZookeeperConfigurationExtensioncs { public static IZookeeperBuilder AddConfiguration(this IZookeeperBuilder builder ,IConfiguration configuration) { InitService( builder,configuration); return builder; } public static void InitService(IZookeeperBuilder builder,IConfiguration configuration) { builder.Services.TryAddSingleton<IConfigureOptions<ZookeeperOptions>>( new ZookeeperConfigurationOptions(configuration)); builder.Services.TryAddSingleton (ServiceDescriptor.Singleton<IOptionsChangeTokenSource<ZookeeperOptions>>( new ConfigurationChangeTokenSource<ZookeeperOptions>(configuration)) ); builder.Services .TryAddEnumerable(ServiceDescriptor.Singleton<IConfigureOptions<ZookeeperOptions>> (new ConfigureFromConfigurationOptions<ZookeeperOptions>(configuration))); builder.Services.AddSingleton(new ZookeeperConfiguration(configuration)); } } }
using Microsoft.Extensions.Configuration; using Microsoft.Extensions.Options; using Walt.Framework.Service.Zookeeper; namespace Walt.Framework.Configuration { public class ZookeeperConfigurationOptions : IConfigureOptions<ZookeeperOptions> { private readonly IConfiguration _configuration; public ZookeeperConfigurationOptions(IConfiguration configuration) { _configuration=configuration; } public void Configure(ZookeeperOptions options) { System.Diagnostics.Debug.WriteLine("zookeeper配置类,适配方法。" +Newtonsoft.Json.JsonConvert.SerializeObject(options)); } } }
以上这三个类就是配置类,下面是构建类和配置信息类:
using Microsoft.Extensions.DependencyInjection; namespace Walt.Framework.Service.Zookeeper { public class ZookeeperBuilder : IZookeeperBuilder { public IServiceCollection Services {get;} public ZookeeperBuilder(IServiceCollection services) { Services=services; } } }
using System; using System.Collections.Generic; namespace Walt.Framework.Service.Zookeeper { public class ZookeeperOptions { /// <param name="connectstring">comma separated host:port pairs, each corresponding to a zk server. /// e.g. "127.0.0.1:3000,127.0.0.1:3001,127.0.0.1:3002" If the optional chroot suffix is used the example would look like: /// "127.0.0.1:3000,127.0.0.1:3001,127.0.0.1:3002/app/a" where the client would be rooted at "/app/a" and all paths would /// be relative to this root - ie getting/setting/etc... "/foo/bar" would result in operations being run on "/app/a/foo/bar" /// (from the server perspective).</param> public string Connectstring{get;set;} public int SessionTimeout{get;set;} public int SessionId{get;set;} public string SessionPwd{get;set;} public bool IsReadOnly{get;set;} } }
using System; using System.Collections.Generic; using System.Threading.Tasks; using Microsoft.Extensions.Logging; using Microsoft.Extensions.Options; using org.apache.zookeeper; using org.apache.zookeeper.data; using static org.apache.zookeeper.ZooKeeper; using System.Linq; using System.Linq.Expressions; using System.Threading; using static org.apache.zookeeper.Watcher.Event; using Newtonsoft.Json; using System.Collections.Concurrent; namespace Walt.Framework.Service.Zookeeper { internal class WaitLockWatch:Watcher { private AutoResetEvent _autoResetEvent; private ILogger _logger; private string _path; private ZookeeperService _zookeeperService; public string _tempNode; public WaitLockWatch(AutoResetEvent autoResetEvent ,ZookeeperService zookeeperService ,ILogger logger,string path ,string tempNode) { _autoResetEvent=autoResetEvent; _zookeeperService=zookeeperService; _logger=logger; _path=path; _tempNode=tempNode; } public override Task process(WatchedEvent @event) { _logger.LogDebug("{0}节点下子节点发生改变,激发监视方法。",_path); var childList=_zookeeperService.GetChildrenAsync(_path,null,true).Result; if(childList==null||childList.Children==null||childList.Children.Count<1) { _logger.LogDebug("获取子序列失败,计数为零.path:{0}",_path); return Task.FromResult(0); } var top=childList.Children.OrderBy(or=>or).First(); if(_path+"/"+top==_tempNode) { _logger.LogDebug("释放阻塞"); _autoResetEvent.Set(); } return Task.FromResult(0); } } internal class WaitConnWatch : Watcher { private AutoResetEvent _autoResetEvent; private ILogger _logger; public WaitConnWatch(AutoResetEvent autoResetEvent ,ILogger logger) { _autoResetEvent=autoResetEvent; _logger=logger; } public override Task process(WatchedEvent @event) { _logger.LogDebug("watch激发,回掉状态:{0}",@event.getState().ToString()); if(@event.getState()== KeeperState.SyncConnected ||@event.getState()== KeeperState.ConnectedReadOnly) { _logger.LogDebug("释放阻塞"); _autoResetEvent.Set(); } return Task.FromResult(0); } } public class ZookeeperService : IZookeeperService { public List<string> requestLockSequence=new List<string>(); private object _lock=new object(); private ZookeeperOptions _zookeeperOptions; private ZooKeeper _zookeeper; private static readonly byte[] NO_PASSWORD = new byte[0]; public Watcher Wathcer {get;set;} public ILoggerFactory LoggerFac { get; set; } private ILogger _logger; AutoResetEvent[] autoResetEvent=new AutoResetEvent[2] {new AutoResetEvent(false),new AutoResetEvent(false)}; public ZookeeperService(IOptionsMonitor<ZookeeperOptions> zookeeperOptions ,ILoggerFactory loggerFac) { LoggerFac=loggerFac; _logger=LoggerFac.CreateLogger<ZookeeperService>(); _zookeeperOptions=zookeeperOptions.CurrentValue; _logger.LogDebug("配置参数:{0}",JsonConvert.SerializeObject(_zookeeperOptions)); zookeeperOptions.OnChange((zookopt,s)=>{ _zookeeperOptions=zookopt; }); _logger.LogDebug("开始连接"); Conn(_zookeeperOptions); } private void Conn(ZookeeperOptions zookeeperOptions) { bool isReadOnly=default(Boolean); Wathcer=new WaitConnWatch(autoResetEvent[0],_logger); if(isReadOnly!=zookeeperOptions.IsReadOnly) { isReadOnly=zookeeperOptions.IsReadOnly; } byte[] pwd=new byte[0]; //如果没有密码和sessionId if(string.IsNullOrEmpty(zookeeperOptions.SessionPwd) &&_zookeeperOptions.SessionId==default(int)) { _zookeeper=new ZooKeeper(zookeeperOptions.Connectstring,zookeeperOptions.SessionTimeout,Wathcer,isReadOnly); } else if (!string.IsNullOrEmpty(zookeeperOptions.SessionPwd)) { pwd=System.Text.Encoding.Default.GetBytes(zookeeperOptions.SessionPwd); _zookeeper=new ZooKeeper(zookeeperOptions.Connectstring,zookeeperOptions.SessionTimeout,Wathcer,0,pwd,isReadOnly); } else { _zookeeper=new ZooKeeper(zookeeperOptions.Connectstring ,zookeeperOptions.SessionTimeout,Wathcer,zookeeperOptions.SessionId,pwd,isReadOnly); } if(_zookeeper.getState()==States.CONNECTING) { _logger.LogDebug("当前状态:CONNECTING。阻塞等待"); autoResetEvent[0].WaitOne(); } } public Task<string> CreateZNode(string path,string data,CreateMode createMode,List<ACL> aclList) { ReConn(); if(string.IsNullOrEmpty(path)||!path.StartsWith('/')) { _logger.LogDebug("path路径非法,参数:path:{0}",path); return null; } byte[] dat=new byte[0]; if(string.IsNullOrEmpty(data)) { dat=System.Text.Encoding.Default.GetBytes(data); } if(createMode==null) { _logger.LogDebug("createMode为null,默认使用CreateMode.PERSISTENT"); createMode=CreateMode.PERSISTENT; } return _zookeeper.createAsync(path,dat,aclList,createMode); } public Task<DataResult> GetDataAsync(string path,Watcher watcher,bool isSync) { ReConn(); if(_zookeeper.existsAsync(path).Result==null ) { _logger.LogDebug("path不存在"); return null; } if(isSync) { _logger.LogDebug("即将进行同步。"); var task=Task.Run(async ()=>{ await _zookeeper.sync(path); }); task.Wait(); } return _zookeeper.getDataAsync(path,watcher); } public Task<Stat> SetDataAsync(string path,string data,bool isSync) { ReConn(); if(_zookeeper.existsAsync(path).Result==null ) { _logger.LogDebug("path不存在"); return null; } byte[] dat=new byte[0]; if(!string.IsNullOrEmpty(data)) { dat=System.Text.Encoding.Default.GetBytes(data); } return _zookeeper.setDataAsync(path,dat); } public async Task<ChildrenResult> GetChildrenAsync(string path,Watcher watcher,bool isSync) { ReConn(); if(_zookeeper.existsAsync(path).Result==null ) { _logger.LogDebug("path不存在"); return null; } if(isSync) { _logger.LogDebug("即将进行同步。"); var task=Task.Run(async ()=>{ _logger.LogDebug("开始同步"); await _zookeeper.sync(path); }); task.Wait(); } return await _zookeeper.getChildrenAsync(path,watcher); } public void DeleteNode(string path,String tempNode) { if(!string.IsNullOrEmpty(tempNode)) { requestLockSequence.Remove(tempNode); } ReConn(); if(_zookeeper.existsAsync(path).Result==null ) { _logger.LogDebug("path不存在"); return; } var task=Task.Run(async ()=>{ _logger.LogDebug("删除node:{0}",path); await _zookeeper.deleteAsync(path); }); task.Wait(); var sequencePath=requestLockSequence.Where(w=>path==w).FirstOrDefault(); if(sequencePath!=null) { requestLockSequence.Remove(sequencePath); } } public string GetDataByLockNode(string path,string sequenceName,List<ACL> aclList,out string tempNodeOut) { _logger.LogInformation("获取分布式锁开始。"); ReConn(); string tempNode=string.Empty; tempNodeOut=string.Empty; if(_zookeeper.existsAsync(path).Result==null ) { _logger.LogDebug("path不存在"); return null; } try { _logger.LogDebug("开始锁定语句块"); lock(_lock) { _logger.LogDebug("锁定,访问requestLockSequence的代码应该同步。"); tempNode=requestLockSequence .Where(w=>w.StartsWith(path+"/"+sequenceName)).FirstOrDefault(); if(tempNode==null) { tempNode=CreateZNode(path+"/"+sequenceName,"",CreateMode.EPHEMERAL_SEQUENTIAL,aclList).Result; _logger.LogDebug("创建节点:{0}",tempNode); if(tempNode==null) { _logger.LogDebug("创建临时序列节点失败。详细参数:path:{0},data:{1},CreateMode:{2}" ,path+"/squence","",CreateMode.EPHEMERAL_SEQUENTIAL); return null; } _logger.LogInformation("创建成功,加入requestLockSequence列表。"); requestLockSequence.Add(tempNode); } else { _logger.LogDebug("已经存在的锁节点,返回null"); return null; } } var childList= GetChildrenAsync(path,null,true).Result; if(childList==null||childList.Children==null||childList.Children.Count<1) { _logger.LogDebug("获取子序列失败,计数为零.path:{0}",path); return null; } _logger.LogDebug("获取path:{0}的子节点:{1}",path,Newtonsoft.Json.JsonConvert.SerializeObject(childList.Children)); var top=childList.Children.OrderBy(or=>or).First(); byte[] da=null; if(path+"/"+top==tempNode) { tempNodeOut =tempNode; da= GetDataAsync(path,null,true).Result.Data; if(da==null||da.Length<1) { return string.Empty; } return System.Text.Encoding.Default.GetString(da); } else { childList= GetChildrenAsync(path,new WaitLockWatch(autoResetEvent[1],this,_logger,path,tempNode),true).Result; autoResetEvent[1].WaitOne(); } _logger.LogDebug("继续执行。"); tempNodeOut =tempNode; da= GetDataAsync(path,null,true).Result.Data; if(da==null||da.Length<1) { return string.Empty; } return System.Text.Encoding.Default.GetString(da); } catch(Exception ep) { _logger.LogError(ep,"获取同步锁出现错误。"); if(!string.IsNullOrEmpty(tempNode)) { DeleteNode(tempNode,tempNode); } } return null; } private void ReConn() { _logger.LogInformation("检查连接状态"); if(_zookeeper.getState()==States.CLOSED ||_zookeeper.getState()== States.NOT_CONNECTED) { _logger.LogInformation("连接为关闭,开始重新连接。"); Conn(_zookeeperOptions); } } public void Close(string tempNode) { var task=Task.Run(async ()=>{ requestLockSequence.Remove(tempNode); await _zookeeper.closeAsync(); }); task.Wait(); } } }
前面的类如果了解了netcore的扩展服务和配置机制,就很简单理解了,我们主要是讲解这个服务类的功能。
首先看服务类其中的一段代码:
private void Conn(ZookeeperOptions zookeeperOptions) { bool isReadOnly=default(Boolean); Wathcer=new WaitConnWatch(autoResetEvent[0],_logger); //监控连接是否连接成功 if(isReadOnly!=zookeeperOptions.IsReadOnly) { isReadOnly=zookeeperOptions.IsReadOnly; } byte[] pwd=new byte[0]; //如果没有密码和sessionId if(string.IsNullOrEmpty(zookeeperOptions.SessionPwd) &&_zookeeperOptions.SessionId==default(int)) { _zookeeper=new ZooKeeper(zookeeperOptions.Connectstring,zookeeperOptions.SessionTimeout,Wathcer,isReadOnly); } else if (!string.IsNullOrEmpty(zookeeperOptions.SessionPwd)) { pwd=System.Text.Encoding.Default.GetBytes(zookeeperOptions.SessionPwd); _zookeeper=new ZooKeeper(zookeeperOptions.Connectstring,zookeeperOptions.SessionTimeout,Wathcer,0,pwd,isReadOnly); } else { _zookeeper=new ZooKeeper(zookeeperOptions.Connectstring ,zookeeperOptions.SessionTimeout,Wathcer,zookeeperOptions.SessionId,pwd,isReadOnly); } if(_zookeeper.getState()==States.CONNECTING) { _logger.LogDebug("当前状态:CONNECTING。阻塞等待"); autoResetEvent[0].WaitOne(); } }
这个方法是连接zookeeper,我们在构造函数中调用它,注意,zookeeper是异步,所以需要watcher类监控和AutoResetEvent阻塞当前线程,因为如果不阻塞,在连接还没建立的时候,后面调用,
会出现错误。在监控类被触发的时候,执行取消阻塞。
internal class WaitConnWatch : Watcher { private AutoResetEvent _autoResetEvent; private ILogger _logger; public WaitConnWatch(AutoResetEvent autoResetEvent ,ILogger logger) { _autoResetEvent=autoResetEvent; _logger=logger; } public override Task process(WatchedEvent @event) { _logger.LogDebug("watch激发,回掉状态:{0}",@event.getState().ToString()); if(@event.getState()== KeeperState.SyncConnected ||@event.getState()== KeeperState.ConnectedReadOnly) { _logger.LogDebug("释放阻塞"); _autoResetEvent.Set(); //取消阻塞当前线程。 } return Task.FromResult(0); } }
大家看zookeeper服务类,里面很多方法在执行前需要执行conn方法,这只是一种失败重连的机制,因为一般没有线程池的,我都会给这个服务单例。哪有人会问,如果失败重连,会不会阻塞当前应用,
这个不会,因为netcore是多线程的,但是会降低这个应用的生产力。我前面翻译过一篇net的关于线程的知识,后面也会单独一篇讲解netcore的线程模型。还有代码中我很少用异常去处理容错性,尽量抛出原生的异常,使用日志去记录,然后为这个类返回个null,异常对性能也有一定的消耗,当然看自己习惯了,目的都是为了应用的健壮性。
其他方法是操作zookeeper的类,大家可以看我贴出来的代码。因为zookeeper最出名的估计就是分布式锁了,所以就把这个功能加进来。
public string GetDataByLockNode(string path,string sequenceName,List<ACL> aclList,out string tempNodeOut) { _logger.LogInformation("获取分布式锁开始。"); ReConn(); string tempNode=string.Empty; tempNodeOut=string.Empty; if(_zookeeper.existsAsync(path).Result==null ) { _logger.LogDebug("path不存在"); return null; } try { _logger.LogDebug("开始锁定语句块"); lock(_lock) //这是我为了防止重复提交做的防止并发,当然实际用的地方是协调应用之间的功能,而肯定不会用人机交互。你可以把这个看作多此一举。 { _logger.LogDebug("锁定,访问requestLockSequence的代码应该同步。"); tempNode=requestLockSequence .Where(w=>w.StartsWith(path+"/"+sequenceName)).FirstOrDefault(); if(tempNode==null) { tempNode=CreateZNode(path+"/"+sequenceName,"",CreateMode.EPHEMERAL_SEQUENTIAL,aclList).Result; _logger.LogDebug("创建节点:{0}",tempNode); if(tempNode==null) { _logger.LogDebug("创建临时序列节点失败。详细参数:path:{0},data:{1},CreateMode:{2}" ,path+"/squence","",CreateMode.EPHEMERAL_SEQUENTIAL); return null; } _logger.LogInformation("创建成功,加入requestLockSequence列表。"); requestLockSequence.Add(tempNode); } else { _logger.LogDebug("已经存在的锁节点,返回null"); return null; } } var childList= GetChildrenAsync(path,null,true).Result; //首先获取lock子节点。 if(childList==null||childList.Children==null||childList.Children.Count<1) { _logger.LogDebug("获取子序列失败,计数为零.path:{0}",path); return null; } _logger.LogDebug("获取path:{0}的子节点:{1}",path,Newtonsoft.Json.JsonConvert.SerializeObject(childList.Children)); var top=childList.Children.OrderBy(or=>or).First(); byte[] da=null; if(path+"/"+top==tempNode) //判断是否是当前自己的节点在队列顶端。 { tempNodeOut =tempNode; da= GetDataAsync(path,null,true).Result.Data; if(da==null||da.Length<1) { return string.Empty; } return System.Text.Encoding.Default.GetString(da); } else { childList= GetChildrenAsync(path,new WaitLockWatch(autoResetEvent[1],this,_logger,path,tempNode),true).Result; //如果自己不再队列顶端,则加监听等待这个节点有更改。 autoResetEvent[1].WaitOne(); } _logger.LogDebug("继续执行。"); tempNodeOut =tempNode; da= GetDataAsync(path,null,true).Result.Data; if(da==null||da.Length<1) { return string.Empty; } return System.Text.Encoding.Default.GetString(da); } catch(Exception ep) { _logger.LogError(ep,"获取同步锁出现错误。"); if(!string.IsNullOrEmpty(tempNode)) { DeleteNode(tempNode,tempNode); } } return null; }
接下来看监听类:
internal class WaitLockWatch:Watcher { private AutoResetEvent _autoResetEvent; private ILogger _logger; private string _path; private ZookeeperService _zookeeperService; public string _tempNode; public WaitLockWatch(AutoResetEvent autoResetEvent ,ZookeeperService zookeeperService ,ILogger logger,string path ,string tempNode) { _autoResetEvent=autoResetEvent; _zookeeperService=zookeeperService; _logger=logger; _path=path; _tempNode=tempNode; } public override Task process(WatchedEvent @event) { _logger.LogDebug("{0}节点下子节点发生改变,激发监视方法。",_path); var childList=_zookeeperService.GetChildrenAsync(_path,null,true).Result; if(childList==null||childList.Children==null||childList.Children.Count<1) { _logger.LogDebug("获取子序列失败,计数为零.path:{0}",_path); return Task.FromResult(0); } var top=childList.Children.OrderBy(or=>or).First(); if(_path+"/"+top==_tempNode) //判断当前节点是否队列顶端 { _logger.LogDebug("释放阻塞"); _autoResetEvent.Set(); } return Task.FromResult(0); } }
这个类,在path节点每次更改或者子节点更改的时候都会激发,仅仅是判断当前节点是不是列表的顶端,再执行 _autoResetEvent.Set();释放阻塞让继续执行。
编译提交nuget,然后集成到测试程序看效果。
首先看zookeeper:
目前节点下没有一个子节点,再看看这个节点的值:
首先新建一个看看:
create -s -e /testzookeeper/sequence ""
然后启动程序,当程序提交一个子节点,他是排序的:
我们删除我们刚刚用命令行创建的节点:
delete /testzookeeper/sequence0000000072
然后再看值:
我们的客户端就是获取到锁后更改这个值,增加了一段字符串,客户端调用如下:
[HttpGet("{id}")] public ActionResult<string> Get(int id) { string sequenceNode=string.Empty; string nodename=_zookeeperService.GetDataByLockNode("/testzookeeper","sequence",ZooDefs.Ids.OPEN_ACL_UNSAFE,out sequenceNode); if(sequenceNode==null) { return "获取分布式锁失败,请查看日志。"; } _zookeeperService.SetDataAsync("/testzookeeper",nodename+"执行了"+DateTime.Now.ToString("yyyyMMddhhmmss"),false); if(!string.IsNullOrEmpty(sequenceNode)) { _zookeeperService.DeleteNode(sequenceNode,sequenceNode); return "取得锁并且成功处理数据,释放锁成功。"; } return "出现错误,请查日志。"; }
当然这个是我做测试用的,实际开发,web就是用来做他的数据库存取的,尽量不要去做一些额外的功能,因为i守护进程或者后台服务和job有它自己的任务。