深入剖析SolrCloud(四)

         上一篇中介绍了连接Zookeeper集群的方法,这一篇将围绕一个有趣的话题---来展开,这就是Replication(索引复制),关于Solr Replication的详细介绍,可以参考http://wiki.apache.org/solr/SolrReplication

         在开始这个话题之前,先从我最近在应用中引入solrmaster/slave架构时,遇到的一个让我困扰的实际问题。

应用场景简单描述如下:

1)首先master节点下载索引分片,然后创建配置文件,加入master节点的replication配置片段,再对索引分片进行合并(关于mergeIndex,可以参考http://wiki.apache.org/solr/MergingSolrIndexes),然后利用上述配置文件和索引数据去创建一个solr核。

2slave节点创建配置文件,加入slave节点的replication配置片段,创建一个空的solr核,等待从master节点进行索引数据同步

出现的问题:slave节点没有从master节点同步到数据。

问题分析:

1)首先检查master节点,获取最新的可复制索引的版本号,

http://master_host:port/solr/replication?command=indexversion

发现返回的索引版本号是0,这说明mater节点根本没有触发replication动作,

2)为了确认上述判断,在slave节点上进一步查看replication的详细信息

http://slave_host:port/solr/replication?command=details

发现确实如此,尽管master节点的索引版本号和slave节点的索引版本号不一致,但索引却没有同步过来,再分别查看master节点和slave节点的日志,发现索引复制动作确实没有开始。

综上所述,确实是master节点没有触发索引复制动作,那究竟是为何呢?先将原因摆出来,后面会通过源码的分析来加以说明。

原因:solr合并索引时,不管你是通过mergeindexeshttp命令,还是调用底层luceneIndexWriter,记得最后一定要提交一个commit,否则,不仅索引不仅不会对查询可见,更是对于master/slave架构的solr集群来说,master节点的replication动作不会触发,因为indexversion没有感知到变化。

         好了,下面开始对SolrReplication的分析

         Solr容器在加载solr核的时候,会对已经注册的各个实现SolrCoreAware接口的Handler进行回调,调用其inform方法。

         对于ReplicationHandler来说,就是在这里对自己是属于master节点还是slave节点进行判断,若是slave节点,则创建一个SnapPuller对象,定时负责从master节点主动拉索引数据下来;若是master节点,则只设置相应的参数。

  public void inform(SolrCore core) {
    this.core = core;
    registerFileStreamResponseWriter();
    registerCloseHook();
    NamedList slave = (NamedList) initArgs.get("slave");
    boolean enableSlave = isEnabled( slave );
    if (enableSlave) {
      tempSnapPuller = snapPuller = new SnapPuller(slave, this, core);
      isSlave = true;
    }
    NamedList master = (NamedList) initArgs.get("master");
    boolean enableMaster = isEnabled( master );
    
    if (!enableSlave && !enableMaster) {
      enableMaster = true;
      master = new NamedList<Object>();
    }
    
    if (enableMaster) {
      includeConfFiles = (String) master.get(CONF_FILES);
      if (includeConfFiles != null && includeConfFiles.trim().length() > 0) {
        List<String> files = Arrays.asList(includeConfFiles.split(","));
        for (String file : files) {
          if (file.trim().length() == 0) continue;
          String[] strs = file.split(":");
          // if there is an alias add it or it is null
          confFileNameAlias.add(strs[0], strs.length > 1 ? strs[1] : null);
        }
        LOG.info("Replication enabled for following config files: " + includeConfFiles);
      }
      List backup = master.getAll("backupAfter");
      boolean backupOnCommit = backup.contains("commit");
      boolean backupOnOptimize = !backupOnCommit && backup.contains("optimize");
      List replicateAfter = master.getAll(REPLICATE_AFTER);
      replicateOnCommit = replicateAfter.contains("commit");
      replicateOnOptimize = !replicateOnCommit && replicateAfter.contains("optimize");

      if (!replicateOnCommit && ! replicateOnOptimize) {
        replicateOnCommit = true;
      }
      
      // if we only want to replicate on optimize, we need the deletion policy to
      
// save the last optimized commit point.
      if (replicateOnOptimize) {
        IndexDeletionPolicyWrapper wrapper = core.getDeletionPolicy();
        IndexDeletionPolicy policy = wrapper == null ? null : wrapper.getWrappedDeletionPolicy();
        if (policy instanceof SolrDeletionPolicy) {
          SolrDeletionPolicy solrPolicy = (SolrDeletionPolicy)policy;
          if (solrPolicy.getMaxOptimizedCommitsToKeep() < 1) {
            solrPolicy.setMaxOptimizedCommitsToKeep(1);
          }
        } else {
          LOG.warn("Replication can't call setMaxOptimizedCommitsToKeep on " + policy);
        }
      }

      if (replicateOnOptimize || backupOnOptimize) {
        core.getUpdateHandler().registerOptimizeCallback(getEventListener(backupOnOptimize, replicateOnOptimize));
      }
      if (replicateOnCommit || backupOnCommit) {
        replicateOnCommit = true;
        core.getUpdateHandler().registerCommitCallback(getEventListener(backupOnCommit, replicateOnCommit));
      }
      if (replicateAfter.contains("startup")) {
        replicateOnStart = true;
        RefCounted<SolrIndexSearcher> s = core.getNewestSearcher(false);
        try {
          DirectoryReader reader = s==null ? null : s.get().getIndexReader();
          if (reader!=null && reader.getIndexCommit() != null && reader.getIndexCommit().getGeneration() != 1L) {
            try {
              if(replicateOnOptimize){
                Collection<IndexCommit> commits = DirectoryReader.listCommits(reader.directory());
                for (IndexCommit ic : commits) {
                  if(ic.getSegmentCount() == 1){
                    if(indexCommitPoint == null || indexCommitPoint.getGeneration() < ic.getGeneration()) indexCommitPoint = ic;
                  }
                }
              } else{
                indexCommitPoint = reader.getIndexCommit();
              }
            } finally {
              // We don't need to save commit points for replication, the SolrDeletionPolicy
              
// always saves the last commit point (and the last optimized commit point, if needed)
              /***
              if(indexCommitPoint != null){
                core.getDeletionPolicy().saveCommitPoint(indexCommitPoint.getGeneration());
              }
              **
*/
            }
          }

          // reboot the writer on the new index
          core.getUpdateHandler().newIndexWriter();

        } catch (IOException e) {
          LOG.warn("Unable to get IndexCommit on startup", e);
        } finally {
          if (s!=null) s.decref();
        }
      }
      String reserve = (String) master.get(RESERVE);
      if (reserve != null && !reserve.trim().equals("")) {
        reserveCommitDuration = SnapPuller.readInterval(reserve);
      }
      LOG.info("Commits will be reserved for  " + reserveCommitDuration);
      isMaster = true;
    }

  ReplicationHandler可以响应多种命令:

1)       indexversion

这里需要了解的第一个概念是索引提交点(IndexCommit),这是底层lucene的东西,可以自行查阅资料。首先获取最新的索引提交点,然后从其中获取索引版本号和索引所属代。

      IndexCommit commitPoint = indexCommitPoint;  // make a copy so it won't change
      if (commitPoint != null && replicationEnabled.get()) {
        core.getDeletionPolicy().setReserveDuration(commitPoint.getVersion(), reserveCommitDuration);
        rsp.add(CMD_INDEX_VERSION, commitPoint.getVersion());

  rsp.add(GENERATION, commitPoint.getGeneration());  

    2backup。这个命令用来对索引做快照。首先获取最新的索引提交点,然后创建做一个SnapShooter,具体的快照动作由这个对象完成,

   private void doSnapShoot(SolrParams params, SolrQueryResponse rsp, SolrQueryRequest req) { 

    try {
      int numberToKeep = params.getInt(NUMBER_BACKUPS_TO_KEEP, Integer.MAX_VALUE);
      IndexDeletionPolicyWrapper delPolicy = core.getDeletionPolicy();
      IndexCommit indexCommit = delPolicy.getLatestCommit();
      
      if(indexCommit == null) {
        indexCommit = req.getSearcher().getReader().getIndexCommit();
      }
      
      // small race here before the commit point is saved
      new SnapShooter(core, params.get("location")).createSnapAsync(indexCommit, numberToKeep, this);
      
    } catch (Exception e) {
      LOG.warn("Exception during creating a snapshot", e);
      rsp.add("exception", e);
    }
  }

 

快照对象会启动一个线程去异步地做一个索引备份。

void createSnapAsync(final IndexCommit indexCommit, final int numberToKeep, final ReplicationHandler replicationHandler) {

    replicationHandler.core.getDeletionPolicy().saveCommitPoint(indexCommit.getVersion());

 

    new Thread() {

      @Override

      public void run() {

        createSnapshot(indexCommit, numberToKeep, replicationHandler);

      }

    }.start();

 }

 

 void createSnapshot(final IndexCommit indexCommit, int numberToKeep, ReplicationHandler replicationHandler) {

    NamedList details = new NamedList();

    details.add("startTime", new Date().toString());

    File snapShotDir = null;

    String directoryName = null;

    Lock lock = null;

    try {

      if(numberToKeep<Integer.MAX_VALUE) {

        deleteOldBackups(numberToKeep);

      }

      SimpleDateFormat fmt = new SimpleDateFormat(DATE_FMT, Locale.US);

      directoryName = "snapshot." + fmt.format(new Date());

      lock = lockFactory.makeLock(directoryName + ".lock");

      if (lock.isLocked()) return;

      snapShotDir = new File(snapDir, directoryName);

      if (!snapShotDir.mkdir()) {

        LOG.warn("Unable to create snapshot directory: " + snapShotDir.getAbsolutePath());

        return;

      }

      Collection<String> files = indexCommit.getFileNames();

      FileCopier fileCopier = new FileCopier(solrCore.getDeletionPolicy(), indexCommit);

      fileCopier.copyFiles(files, snapShotDir);

 

      details.add("fileCount", files.size());

      details.add("status", "success");

      details.add("snapshotCompletedAt", new Date().toString());

    } catch (Exception e) {

      SnapPuller.delTree(snapShotDir);

      LOG.error("Exception while creating snapshot", e);

      details.add("snapShootException", e.getMessage());

    } finally {

      replicationHandler.core.getDeletionPolicy().releaseCommitPoint(indexCommit.getVersion());  

      replicationHandler.snapShootDetails = details;

      if (lock != null) {

        try {

          lock.release();

        } catch (IOException e) {

          LOG.error("Unable to release snapshoot lock: " + directoryName + ".lock");

        }

      }

    }

  }

3fetchindex。响应来自slave节点的取索引文件的请求,会启动一个线程来实现索引文件的获取。

      String masterUrl = solrParams.get(MASTER_URL);

      if (!isSlave && masterUrl == null) {

        rsp.add(STATUS,ERR_STATUS);

        rsp.add("message","No slave configured or no 'masterUrl' Specified");

        return;

      }

      final SolrParams paramsCopy = new ModifiableSolrParams(solrParams);

      new Thread() {

        @Override

        public void run() {

          doFetch(paramsCopy);

        }

      }.start();

      rsp.add(STATUS, OK_STATUS);

具体的获取动作是通过SnapPuller对象来实现的,首先尝试获取pull对象锁,如果请求锁失败,则说明还有取索引数据动作未结束,如果请求锁成功,就调用SnapPuller对象的fetchLatestIndex方法来取最新的索引数据。

 void doFetch(SolrParams solrParams) {

    String masterUrl = solrParams == null ? null : solrParams.get(MASTER_URL);

    if (!snapPullLock.tryLock())

      return;

    try {

      tempSnapPuller = snapPuller;

      if (masterUrl != null) {

        NamedList<Object> nl = solrParams.toNamedList();

        nl.remove(SnapPuller.POLL_INTERVAL);

        tempSnapPuller = new SnapPuller(nl, this, core);

      }

      tempSnapPuller.fetchLatestIndex(core);

    } catch (Exception e) {

      LOG.error("SnapPull failed ", e);

    } finally {

      tempSnapPuller = snapPuller;

      snapPullLock.unlock();

    }

 }

最后真正的取索引数据过程,首先,若mastet节点的indexversion0,则说明master节点根本没有提供可供复制的索引数据,若master节点和slave节点的indexversion相同,则说明slave节点目前与master节点索引数据状态保持一致,无需同步。若两者的indexversion不同,则开始索引复制过程,首先从master节点上下载指定索引版本号的索引文件列表,然后创建一个索引文件同步服务线程来完成同并工作。

这里需要区分的是,如果master节点的年代比slave节点要老,那就说明两者已经不相容,此时slave节点需要新建一个索引目录,再从master节点做一次全量索引复制。还需要注意的一点是,索引同步也是可以同步配置文件的,若配置文件发生变化,则需要对solr核进行一次reload操作。最对了,还有,和文章开头一样, slave节点同步完数据后,别忘了做一次commit操作,以便刷新自己的索引提交点到最新的状态。最后,关闭并等待同步服务线程结束。此外,具体的取索引文件是通过FileFetcher对象来完成。

 boolean fetchLatestIndex(SolrCore core) throws IOException {

    replicationStartTime = System.currentTimeMillis();

    try {

      //get the current 'replicateable' index version in the master

      NamedList response = null;

      try {

        response = getLatestVersion();

      } catch (Exception e) {

        LOG.error("Master at: " + masterUrl + " is not available. Index fetch failed. Exception: " + e.getMessage());

        return false;

      }

      long latestVersion = (Long) response.get(CMD_INDEX_VERSION);

      long latestGeneration = (Long) response.get(GENERATION);

      if (latestVersion == 0L) {

        //there is nothing to be replicated

        return false;

      }

      IndexCommit commit;

      RefCounted<SolrIndexSearcher> searcherRefCounted = null;

      try {

        searcherRefCounted = core.getNewestSearcher(false);

        commit = searcherRefCounted.get().getReader().getIndexCommit();

      } finally {

        if (searcherRefCounted != null)

          searcherRefCounted.decref();

      }

      if (commit.getVersion() == latestVersion && commit.getGeneration() == latestGeneration) {

        //master and slave are alsready in sync just return

        LOG.info("Slave in sync with master.");

        return false;

      }

      LOG.info("Master's version: " + latestVersion + ", generation: " + latestGeneration);

      LOG.info("Slave's version: " + commit.getVersion() + ", generation: " + commit.getGeneration());

      LOG.info("Starting replication process");

      // get the list of files first

      fetchFileList(latestVersion);

      // this can happen if the commit point is deleted before we fetch the file list.

      if(filesToDownload.isEmpty()) return false;

      LOG.info("Number of files in latest index in master: " + filesToDownload.size());

 

      // Create the sync service

      fsyncService = Executors.newSingleThreadExecutor();

      // use a synchronized list because the list is read by other threads (to show details)

      filesDownloaded = Collections.synchronizedList(new ArrayList<Map<String, Object>>());

      // if the generateion of master is older than that of the slave , it means they are not compatible to be copied

      // then a new index direcory to be created and all the files need to be copied

      boolean isFullCopyNeeded = commit.getGeneration() >= latestGeneration;

      File tmpIndexDir = createTempindexDir(core);

      if (isIndexStale())

        isFullCopyNeeded = true;

      successfulInstall = false;

      boolean deleteTmpIdxDir = true;

      File indexDir = null ;

      try {

        indexDir = new File(core.getIndexDir());

        downloadIndexFiles(isFullCopyNeeded, tmpIndexDir, latestVersion);

        LOG.info("Total time taken for download : " + ((System.currentTimeMillis() - replicationStartTime) / 1000) + " secs");

        Collection<Map<String, Object>> modifiedConfFiles = getModifiedConfFiles(confFilesToDownload);

        if (!modifiedConfFiles.isEmpty()) {

          downloadConfFiles(confFilesToDownload, latestVersion);

          if (isFullCopyNeeded) {

            successfulInstall = modifyIndexProps(tmpIndexDir.getName());

            deleteTmpIdxDir = false;

          } else {

            successfulInstall = copyIndexFiles(tmpIndexDir, indexDir);

          }

          if (successfulInstall) {

            LOG.info("Configuration files are modified, core will be reloaded");

            logReplicationTimeAndConfFiles(modifiedConfFiles, successfulInstall);//write to a file time of replication and conf files.

            reloadCore();

          }

        } else {

          terminateAndWaitFsyncService();

          if (isFullCopyNeeded) {

            successfulInstall = modifyIndexProps(tmpIndexDir.getName());

            deleteTmpIdxDir = false;

          } else {

            successfulInstall = copyIndexFiles(tmpIndexDir, indexDir);

          }

          if (successfulInstall) {

            logReplicationTimeAndConfFiles(modifiedConfFiles, successfulInstall);

            doCommit();

          }

        }

        replicationStartTime = 0;

        return successfulInstall;

      } catch (ReplicationHandlerException e) {

        LOG.error("User aborted Replication");

      } catch (SolrException e) {

        throw e;

      } catch (Exception e) {

        throw new SolrException(SolrException.ErrorCode.SERVER_ERROR, "Index fetch failed : ", e);

      } finally {

        if (deleteTmpIdxDir) delTree(tmpIndexDir);

        else delTree(indexDir);

      }

      return successfulInstall;

    } finally {

      if (!successfulInstall) {

        logReplicationTimeAndConfFiles(null, successfulInstall);

      }

      filesToDownload = filesDownloaded = confFilesDownloaded = confFilesToDownload = null;

      replicationStartTime = 0;

      fileFetcher = null;

      if (fsyncService != null && !fsyncService.isShutdown()) fsyncService.shutdownNow();

      fsyncService = null;

      stop = false;

      fsyncException = null;

    }

 } 

posted on 2012-02-29 00:07 Phinecos(洞庭散人) 阅读(...) 评论(...) 编辑 收藏

导航

统计

公告