`
bupt04406
  • 浏览: 347363 次
  • 性别: Icon_minigender_1
  • 来自: 杭州
社区版块
存档分类
最新评论

hive insert overwrite into

    博客分类:
  • Hive
阅读更多
CREATE TABLE records (year STRING, temperature INT, quality INT)
ROW FORMAT DELIMITED
  FIELDS TERMINATED BY '\t';

LOAD DATA LOCAL INPATH '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt' OVERWRITE INTO TABLE records;


hive> explain
    > LOAD DATA LOCAL INPATH '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt' OVERWRITE INTO TABLE records;
OK
ABSTRACT SYNTAX TREE:
  (TOK_LOAD '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt' (TOK_TAB records) LOCAL OVERWRITE)

STAGE DEPENDENCIES:
  Stage-0 is a root stage
  Stage-1 depends on stages: Stage-0

STAGE PLANS:
  Stage: Stage-0
    Copy
      source: file:/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt
      destination: hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_00-10-28_984_2803781885868135028/-ext-10000

  Stage: Stage-1
    Move Operator
      tables:
          replace: true
          table:
              input format: org.apache.hadoop.mapred.TextInputFormat
              output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
              serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe
              name: records


Time taken: 0.333 seconds



(TOK_LOAD
  '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt'
  (TOK_TAB records)
          LOCAL
          OVERWRITE
)


hive> explain extended
    > LOAD DATA LOCAL INPATH '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt' OVERWRITE INTO TABLE records;
OK
ABSTRACT SYNTAX TREE:
  (TOK_LOAD '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt' (TOK_TAB records) LOCAL OVERWRITE)

STAGE DEPENDENCIES:
  Stage-0 is a root stage
  Stage-1 depends on stages: Stage-0

STAGE PLANS:
  Stage: Stage-0
    Copy
      source: file:/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt
      destination: hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-23-29_636_1841974897600272533/-ext-10000

  Stage: Stage-1
    Move Operator
      tables:
          replace: true
          source: hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-23-29_636_1841974897600272533/-ext-10000
          table:
              input format: org.apache.hadoop.mapred.TextInputFormat
              output format: org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
              properties:
                bucket_count -1
                columns year,temperature,quality
                columns.types string:int:int
                field.delim
                file.inputformat org.apache.hadoop.mapred.TextInputFormat
                file.outputformat org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
                location hdfs://localhost:54310/user/hive/warehouse/records
                name records
                serialization.ddl struct records { string year, i32 temperature, i32 quality}
                serialization.format
                serialization.lib org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe
                transient_lastDdlTime 1313975965
              serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe
              name: records
          tmp directory: hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-23-29_636_1841974897600272533/-ext-10001





CliDriver:
CliDriver.main() {
   ret = cli.processLine(line);
}
CliDriver:
public int processLine(String line) {
   ret = processCmd(command);
}
CliDriver:
public int processCmd(String cmd) {
   CommandProcessor proc = CommandProcessorFactory.get(tokens[0]);
   Driver qp = (Driver) proc;
   ret = qp.run(cmd).getResponseCode();
}
Driver:
public CommandProcessorResponse run(String command) {
   // command = LOAD DATA LOCAL INPATH '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt' OVERWRITE INTO TABLE records

   int ret = compile(command);
   ret = execute();
}

Driver:
  public int compile(String command) {
        ctx = new Context(conf); //
       
         SemanticAnalyzerFactory sfactory = new SemanticAnalyzerFactory(conf);
         BaseSemanticAnalyzer sem = sfactory.get(tree); // return new LoadSemanticAnalyzer(conf);
         sem.analyze(tree, ctx);
        
  }


  public Context(Configuration conf) throws IOException {
    this(conf, generateExecutionId());
  }

  /**
   * Create a Context with a given executionId.  ExecutionId, together with
   * user name and conf, will determine the temporary directory locations.
   */
  public Context(Configuration conf, String executionId) throws IOException {
    this.conf = conf;
    this.executionId = executionId; //hive_2011-08-21_00-02-22_445_7799135143086468923

    // non-local tmp location is configurable. however it is the same across
    // all external file systems
    nonLocalScratchPath =
        new Path(HiveConf.getVar(conf, HiveConf.ConfVars.SCRATCHDIR),
                executionId);  // /tmp/hive-tianzhao/hive_2011-08-21_00-02-22_445_7799135143086468923

    // local tmp location is not configurable for now
    localScratchDir = System.getProperty("java.io.tmpdir")
           + Path.SEPARATOR + System.getProperty("user.name") + Path.SEPARATOR
           + executionId;  // /tmp/tianzhao/hive_2011-08-21_00-02-22_445_7799135143086468923
  }


public class LoadSemanticAnalyzer extends BaseSemanticAnalyzer {
  public void analyzeInternal(ASTNode ast) throws SemanticException {
    isLocal = false;
    isOverWrite = false;
    Tree fromTree = ast.getChild(0); //  '/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt'
    Tree tableTree = ast.getChild(1); // TOK_TAB

    if (ast.getChildCount() == 4) { // true
      isLocal = true;  //
      isOverWrite = true; //
    }

    if (ast.getChildCount() == 3) {
      if (ast.getChild(2).getText().toLowerCase().equals("local")) {
        isLocal = true;
      } else {
        isOverWrite = true;
      }
    }

    // initialize load path
    URI fromURI;
    try {
      String fromPath = stripQuotes(fromTree.getText()); ///home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt
      fromURI = initializeFromURI(fromPath);  //  file:/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt
    } catch (IOException e) {
      throw new SemanticException(ErrorMsg.INVALID_PATH.getMsg(fromTree, e
          .getMessage()), e);
    } catch (URISyntaxException e) {
      throw new SemanticException(ErrorMsg.INVALID_PATH.getMsg(fromTree, e
          .getMessage()), e);
    }

    // initialize destination table/partition
    tableSpec ts = new tableSpec(db, conf, (ASTNode) tableTree); //
    Table tbl = ts.tableHandle;
    if (tbl.isView()) {
      throw new SemanticException(ErrorMsg.DML_AGAINST_VIEW.getMsg());
    }
    if (tbl.isNonNative()) {
      throw new SemanticException(ErrorMsg.LOAD_INTO_NON_NATIVE.getMsg());
    }

    genAuthorizeEntry(db.getCurrentDatabase(), tbl.getTableName(), null, Privilege.INSERT_PRIV);

    URI toURI = (ts.partHandle != null) ? ts.partHandle.getDataLocation()
        : ts.tableHandle.getDataLocation(); // hdfs://localhost:54310/user/hive/warehouse/records

    List<FieldSchema> parts = ts.tableHandle.getPartitionKeys();
    if (isOverWrite && (parts != null && parts.size() > 0)
        && (ts.partSpec == null || ts.partSpec.size() == 0)) {
      throw new SemanticException(ErrorMsg.NEED_PARTITION_ERROR.getMsg());
    }

    // make sure the arguments make sense
    applyConstraints(fromURI, toURI, fromTree, isLocal);

    Task<? extends Serializable> rTask = null;

    // create copy work
    if (isLocal) {  //true
      // if the local keyword is specified - we will always make a copy. this
      // might seem redundant in the case
      // that the hive warehouse is also located in the local file system - but
      // that's just a test case.
      String copyURIStr = ctx.getExternalTmpFileURI(toURI); //hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10000
      URI copyURI = URI.create(copyURIStr);
      rTask = TaskFactory.get(new CopyWork(fromURI.toString(), copyURIStr),
          conf);
      fromURI = copyURI;
    }

    // create final load/move work

    String loadTmpPath = ctx.getExternalTmpFileURI(toURI);  // hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10001
    Map<String, String> partSpec = ts.getPartSpec();
    if (partSpec == null) {
      partSpec = new LinkedHashMap<String, String>();
    }
    LoadTableDesc loadTableWork = new LoadTableDesc(fromURI.toString(),
        loadTmpPath, Utilities.getTableDesc(ts.tableHandle), partSpec, isOverWrite);

    if (rTask != null) {
      rTask.addDependentTask(TaskFactory.get(new MoveWork(getInputs(),
          getOutputs(), loadTableWork, null, true), conf));
    } else {
      rTask = TaskFactory.get(new MoveWork(getInputs(), getOutputs(),
          loadTableWork, null, true), conf);
    }

    rootTasks.add(rTask);
}



  private URI initializeFromURI(String fromPath) throws IOException,
      URISyntaxException {
    URI fromURI = new Path(fromPath).toUri();  //  /home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt

    String fromScheme = fromURI.getScheme(); // null
    String fromAuthority = fromURI.getAuthority(); // null
    String path = fromURI.getPath(); // /home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt

    // generate absolute path relative to current directory or hdfs home
    // directory
    if (!path.startsWith("/")) {  //不是绝对路径,执行下面的操作
      if (isLocal) {
        path = new Path(System.getProperty("user.dir"), path).toString();
      } else {
        path = new Path(new Path("/user/" + System.getProperty("user.name")),
            path).toString();
      }
    }

    // set correct scheme and authority
    if (StringUtils.isEmpty(fromScheme)) { // true
      if (isLocal) { //true
        // file for local
        fromScheme = "file"; //
      } else {
        // use default values from fs.default.name
        URI defaultURI = FileSystem.get(conf).getUri();
        fromScheme = defaultURI.getScheme();
        fromAuthority = defaultURI.getAuthority();
      }
    }

    // if scheme is specified but not authority then use the default authority
    if (fromScheme.equals("hdfs") && StringUtils.isEmpty(fromAuthority)) {
      URI defaultURI = FileSystem.get(conf).getUri();
      fromAuthority = defaultURI.getAuthority();
    }

    LOG.debug(fromScheme + "@" + fromAuthority + "@" + path);
    return new URI(fromScheme, fromAuthority, path, null, null); //file:/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt
  }



    public tableSpec(Hive db, HiveConf conf, ASTNode ast)
        throws SemanticException {

      assert (ast.getToken().getType() == HiveParser.TOK_TAB);
      int childIndex = 0;
      numDynParts = 0;

      try {
        // get table metadata
        tableName = unescapeIdentifier(ast.getChild(0).getText());  //  records
        boolean testMode = conf.getBoolVar(HiveConf.ConfVars.HIVETESTMODE);  // false
        if (testMode) {
          tableName = conf.getVar(HiveConf.ConfVars.HIVETESTMODEPREFIX)
              + tableName;
        }

        tableHandle = db.getTable(tableName);
      } catch (InvalidTableException ite) {
        throw new SemanticException(ErrorMsg.INVALID_TABLE.getMsg(ast
            .getChild(0)), ite);
      } catch (HiveException e) {
        throw new SemanticException(ErrorMsg.GENERIC_ERROR.getMsg(ast
            .getChild(childIndex), e.getMessage()), e);
      }
      // get partition metadata if partition specified
      if (ast.getChildCount() == 2) {  // false
        childIndex = 1;
        ASTNode partspec = (ASTNode) ast.getChild(1);
        // partSpec is a mapping from partition column name to its value.
        partSpec = new LinkedHashMap<String, String>(partspec.getChildCount());
        for (int i = 0; i < partspec.getChildCount(); ++i) {
          ASTNode partspec_val = (ASTNode) partspec.getChild(i);
          String val = null;
          if (partspec_val.getChildCount() < 2) { // DP in the form of T partition (ds, hr)
            ++numDynParts;
          } else { // in the form of T partition (ds="2010-03-03")
            val = stripQuotes(partspec_val.getChild(1).getText());
          }
          partSpec.put(unescapeIdentifier(partspec_val.getChild(0).getText().toLowerCase()), val);
        }
        // check if the partition spec is valid
        if (numDynParts > 0) {
          List<FieldSchema> parts = tableHandle.getPartitionKeys();
          int numStaPart = parts.size() - numDynParts;
          if (numStaPart == 0 &&
              conf.getVar(HiveConf.ConfVars.DYNAMICPARTITIONINGMODE).equalsIgnoreCase("strict")) {
            throw new SemanticException(ErrorMsg.DYNAMIC_PARTITION_STRICT_MODE.getMsg());
          }
        for (FieldSchema fs: parts) {
          if (partSpec.get(fs.getName().toLowerCase()) == null) {
            if (numStaPart > 0) { // found a DP, but there exists ST as subpartition
              throw new SemanticException(
                  ErrorMsg.PARTITION_DYN_STA_ORDER.getMsg(ast.getChild(childIndex)));
            }
            break;
          } else {
            --numStaPart;
          }
        }
          partHandle = null;
        } else {
          try {
            // this doesn't create partition. partition is created in MoveTask
            partHandle = new Partition(tableHandle, partSpec, null);
        } catch (HiveException e) {
         throw new SemanticException(
             ErrorMsg.INVALID_PARTITION.getMsg(ast.getChild(childIndex)));
        }
        }
      }
    }




Driver.execute() {
  
}

CopyTask.execute()
  public int execute(DriverContext driverContext) {
    FileSystem dstFs = null;
    Path toPath = null;
    try {
      Path fromPath = new Path(work.getFromPath()); //file:/home/tianzhao/book/hadoop-book/input/ncdc/micro-tab/sample.txt
      toPath = new Path(work.getToPath()); //hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10000

      console.printInfo("Copying data from " + fromPath.toString(), " to "
          + toPath.toString());

      FileSystem srcFs = fromPath.getFileSystem(conf); //org.apache.hadoop.fs.LocalFileSystem@1c9e4d2
      dstFs = toPath.getFileSystem(conf);

      FileStatus[] srcs = LoadSemanticAnalyzer.matchFilesOrDir(srcFs, fromPath);

      if (srcs == null || srcs.length == 0) {
        console.printError("No files matching path: " + fromPath.toString());
        errorMessage = "No files matching path: " + fromPath.toString();
        return 3;
      }

      if (!dstFs.mkdirs(toPath)) {
        console
            .printError("Cannot make target directory: " + toPath.toString());
        errorMessage = "Cannot make target directory: " + toPath.toString();
        return 2;
      }

      for (FileStatus oneSrc : srcs) {
        LOG.debug("Copying file: " + oneSrc.getPath().toString());
        if (!FileUtil.copy(srcFs, oneSrc.getPath(), dstFs, toPath, false, // delete
            // source
            true, // overwrite destination
            conf)) {
          console.printError("Failed to copy: '" + oneSrc.getPath().toString()
              + "to: '" + toPath.toString() + "'");
          errorMessage = "Failed to copy: '" + oneSrc.getPath().toString()
              + "to: '" + toPath.toString() + "'";
          return 1;
        }
      }
      return 0;

    } catch (Exception e) {
      console.printError("Failed with exception " + e.getMessage(), "\n"
          + StringUtils.stringifyException(e));
      errorMessage = "Failed with exception " + e.getMessage()+ "\n"
          + StringUtils.stringifyException(e);
      return (1);
    }
  }




MoveTask:
  public int execute(DriverContext driverContext) {

    try {
      // Do any hive related operations like moving tables and files
      // to appropriate locations
      LoadFileDesc lfd = work.getLoadFileWork();  // null
      if (lfd != null) {
        Path targetPath = new Path(lfd.getTargetDir());
        Path sourcePath = new Path(lfd.getSourceDir());
        FileSystem fs = sourcePath.getFileSystem(conf);
        if (lfd.getIsDfsDir()) {
          // Just do a rename on the URIs, they belong to the same FS
          String mesg = "Moving data to: " + lfd.getTargetDir();
          String mesg_detail = " from " + lfd.getSourceDir();
          console.printInfo(mesg, mesg_detail);

          // delete the output directory if it already exists
          fs.delete(targetPath, true);
          // if source exists, rename. Otherwise, create a empty directory
          if (fs.exists(sourcePath)) {
            if (!fs.rename(sourcePath, targetPath)) {
              throw new HiveException("Unable to rename: " + sourcePath
                  + " to: " + targetPath);
            }
          } else if (!fs.mkdirs(targetPath)) {
            throw new HiveException("Unable to make directory: " + targetPath);
          }
        } else {
          // This is a local file
          String mesg = "Copying data to local directory " + lfd.getTargetDir();
          String mesg_detail = " from " + lfd.getSourceDir();
          console.printInfo(mesg, mesg_detail);

          // delete the existing dest directory
          LocalFileSystem dstFs = FileSystem.getLocal(conf);

          if (dstFs.delete(targetPath, true) || !dstFs.exists(targetPath)) {
            console.printInfo(mesg, mesg_detail);
            // if source exists, rename. Otherwise, create a empty directory
            if (fs.exists(sourcePath)) {
              fs.copyToLocalFile(sourcePath, targetPath);
            } else {
              if (!dstFs.mkdirs(targetPath)) {
                throw new HiveException("Unable to make local directory: "
                    + targetPath);
              }
            }
          } else {
            throw new AccessControlException(
                "Unable to delete the existing destination directory: "
                + targetPath);
          }
        }
      }

      // Next we do this for tables and partitions
      LoadTableDesc tbd = work.getLoadTableWork();
      if (tbd != null) {
        StringBuilder mesg = new StringBuilder("Loading data to table ")
            .append( tbd.getTable().getTableName());
        if (tbd.getPartitionSpec().size() > 0) {
          mesg.append(" partition (");
          Map<String, String> partSpec = tbd.getPartitionSpec();
          for (String key: partSpec.keySet()) {
            mesg.append(key).append('=').append(partSpec.get(key)).append(", ");
          }
          mesg.setLength(mesg.length()-2);
          mesg.append(')');
        }
        String mesg_detail = " from " + tbd.getSourceDir();
        console.printInfo(mesg.toString(), mesg_detail);  //11/08/21 21:58:44 INFO exec.MoveTask: Loading data to table records from hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10000

        Table table = db.getTable(db.getCurrentDatabase(), tbd
            .getTable().getTableName());

        if (work.getCheckFileFormat()) {
          // Get all files from the src directory
          FileStatus[] dirs;
          ArrayList<FileStatus> files;
          FileSystem fs;
          try {
            fs = FileSystem.get(table.getDataLocation(), conf);
            dirs = fs.globStatus(new Path(tbd.getSourceDir()));
            files = new ArrayList<FileStatus>();
            for (int i = 0; (dirs != null && i < dirs.length); i++) {
              files.addAll(Arrays.asList(fs.listStatus(dirs[i].getPath())));
              // We only check one file, so exit the loop when we have at least
              // one.
              if (files.size() > 0) {
                break;
              }
            }
          } catch (IOException e) {
            throw new HiveException(
                "addFiles: filesystem error in check phase", e);
          }
          if (HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVECHECKFILEFORMAT)) {
            // Check if the file format of the file matches that of the table.
            boolean flag = HiveFileFormatUtils.checkInputFormat(
                fs, conf, tbd.getTable().getInputFileFormatClass(), files);
            if (!flag) {
              throw new HiveException(
                  "Wrong file format. Please check the file's format.");
            }
          }
        }

        // Create a data container
        DataContainer dc = null;
        if (tbd.getPartitionSpec().size() == 0) {
          dc = new DataContainer(table.getTTable());
          db.loadTable(new Path(tbd.getSourceDir()), tbd.getTable()
              .getTableName(), tbd.getReplace(), new Path(tbd.getTmpDir()));  // 替换。
          if (work.getOutputs() != null) {
            work.getOutputs().add(new WriteEntity(table));
          }
        } else {
          LOG.info("Partition is: " + tbd.getPartitionSpec().toString());
          // deal with dynamic partitions
          DynamicPartitionCtx dpCtx = tbd.getDPCtx();
          if (dpCtx != null && dpCtx.getNumDPCols() > 0) { // dynamic partitions
            // load the list of DP partitions and return the list of partition specs
            ArrayList<LinkedHashMap<String, String>> dp =
              db.loadDynamicPartitions(
                  new Path(tbd.getSourceDir()),
                  tbd.getTable().getTableName(),
                tbd.getPartitionSpec(),
                tbd.getReplace(),
                new Path(tbd.getTmpDir()),
                dpCtx.getNumDPCols());
            // for each partition spec, get the partition
            // and put it to WriteEntity for post-exec hook
            for (LinkedHashMap<String, String> partSpec: dp) {
              Partition partn = db.getPartition(table, partSpec, false);

              WriteEntity enty = new WriteEntity(partn);
              if (work.getOutputs() != null) {
                work.getOutputs().add(enty);
              }
              // Need to update the queryPlan's output as well so that post-exec hook get executed.
              // This is only needed for dynamic partitioning since for SP the the WriteEntity is
              // constructed at compile time and the queryPlan already contains that.
              // For DP, WriteEntity creation is deferred at this stage so we need to update
              // queryPlan here.
              if (queryPlan.getOutputs() == null) {
                queryPlan.setOutputs(new HashSet<WriteEntity>());
              }
              queryPlan.getOutputs().add(enty);

              // update columnar lineage for each partition
              dc = new DataContainer(table.getTTable(), partn.getTPartition());

              if (SessionState.get() != null) {
                SessionState.get().getLineageState().setLineage(tbd.getSourceDir(), dc,
                    table.getCols());
              }

              console.printInfo("\tLoading partition " + partSpec);
            }
            dc = null; // reset data container to prevent it being added again.
          } else { // static partitions
            db.loadPartition(new Path(tbd.getSourceDir()), tbd.getTable().getTableName(),
                tbd.getPartitionSpec(), tbd.getReplace(), new Path(tbd.getTmpDir()));
          Partition partn = db.getPartition(table, tbd.getPartitionSpec(), false);
          dc = new DataContainer(table.getTTable(), partn.getTPartition());
          // add this partition to post-execution hook
          if (work.getOutputs() != null) {
            work.getOutputs().add(new WriteEntity(partn));
          }
         }
        }
        if (SessionState.get() != null && dc != null) {
          SessionState.get().getLineageState().setLineage(tbd.getSourceDir(), dc,
              table.getCols());
        }
      }

      return 0;
    } catch (Exception e) {
      console.printError("Failed with exception " + e.getMessage(), "\n"
          + StringUtils.stringifyException(e));
      errorMessage = "Failed with exception " + e.getMessage() + "\n"
          + StringUtils.stringifyException(e);
      return (1);
    }
  }


Hive.java:
  public void loadTable(Path loadPath, String tableName, boolean replace,
      Path tmpDirPath) throws HiveException {
    Table tbl = getTable(tableName); //records

    if (replace) {  //true
      tbl.replaceFiles(loadPath, tmpDirPath); 
// loadPath=hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10000
// tmpDirPath=hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10001
    } else {
      tbl.copyFiles(loadPath);
    }
  }

Table:
  protected void replaceFiles(Path srcf, Path tmpd) throws HiveException {
    FileSystem fs;
    try {
      fs = FileSystem.get(getDataLocation(), Hive.get().getConf());
      Hive.replaceFiles(srcf, new Path(getDataLocation().getPath()), fs, tmpd);
    } catch (IOException e) {
      throw new HiveException("addFiles: filesystem error in check phase", e);
    }
  }
getDataLocation() // hdfs://localhost:54310/user/hive/warehouse/records



  static protected void replaceFiles(Path srcf, Path destf, FileSystem fs,
      Path tmppath) throws HiveException {
    FileStatus[] srcs;
    try {
      srcs = fs.globStatus(srcf);  // srcf = hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10000
    } catch (IOException e) {
      throw new HiveException("addFiles: filesystem error in check phase", e);
    }
    if (srcs == null) {
      LOG.info("No sources specified to move: " + srcf);
      return;
      // srcs = new FileStatus[0]; Why is this needed?
    }
    checkPaths(fs, srcs, destf, true);   //   destf = /user/hive/warehouse/records

    try {
      fs.mkdirs(tmppath);  //  tmppath = hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10001
      for (FileStatus src : srcs) {
        FileStatus[] items = fs.listStatus(src.getPath());
        for (int j = 0; j < items.length; j++) {
          if (!fs.rename(items[j].getPath(), new Path(tmppath, items[j]
              .getPath().getName()))) {
//
//   public boolean rename(Path src, Path dst) throws IOException {
//    return dfs.rename(getPathName(src), getPathName(dst));
//  }
// src = hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10000/sample.txt
// dst = hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10001/sample.txt

            throw new HiveException("Error moving: " + items[j].getPath()
                + " into: " + tmppath);
          }
        }
      }

      // point of no return
      boolean b = fs.delete(destf, true);  // destf = /user/hive/warehouse/records
      LOG.debug("Deleting:" + destf.toString() + ",Status:" + b);

      // create the parent directory otherwise rename can fail if the parent
      // doesn't exist
      if (!fs.mkdirs(destf.getParent())) {
        throw new HiveException("Unable to create destination directory: "
            + destf.getParent().toString());
      }

      b = fs.rename(tmppath, destf); 
// tmppath = hdfs://localhost:54310/tmp/hive-tianzhao/hive_2011-08-21_18-32-25_911_8382925475653721697/-ext-10001
// destf = /user/hive/warehouse/records
      if (!b) {
        throw new HiveException("Unable to move results from " + tmppath
            + " to destination directory: " + destf.getParent().toString());
      }
      LOG.debug("Renaming:" + tmppath.toString() + ",Status:" + b);

    } catch (IOException e) {
      throw new HiveException("replaceFiles: error while moving files from "
          + tmppath + " to " + destf + "!!!", e);
    }
    // In case of error, we should leave the temporary data there, so
    // that user can recover the data if necessary.
  }


  static private void checkPaths(FileSystem fs, FileStatus[] srcs, Path destf,
      boolean replace) throws HiveException {
    try {
      for (FileStatus src : srcs) {
        FileStatus[] items = fs.listStatus(src.getPath());
        for (FileStatus item : items) {

          if (Utilities.isTempPath(item)) {
            // This check is redundant because temp files are removed by
            // execution layer before
            // calling loadTable/Partition. But leaving it in just in case.
            fs.delete(item.getPath(), true);
            continue;
          }
          if (item.isDir()) {
            throw new HiveException("checkPaths: " + src.getPath()
                + " has nested directory" + item.getPath());
          }
          Path tmpDest = new Path(destf, item.getPath().getName()); ///user/hive/warehouse/records/sample.txt
          if (!replace && fs.exists(tmpDest)) {   //  replace = true
            throw new HiveException("checkPaths: " + tmpDest
                + " already exists");
          }
        }
      }
    } catch (IOException e) {
      throw new HiveException("checkPaths: filesystem error in check phase", e);
    }
  }


分享到:
评论
1 楼 lvyuan1234 2017-02-24  
你好,你那个sample.txt文件可以分享给我吗

相关推荐

    Hive那些事儿之八-大数据踩过的坑——Hive insert

    insert into db_name.table_name_1 ( col_1,col2,col3 ) with temp_table_1 as ( select id,col_2 from db_name.table_name_2 where id = condatition ), temp_table_2 as ( select id,col_3 from db_name....

    HIVE-SQL操作语句

    hive&gt; LOAD DATA LOCAL INPATH './examples/files/kv2.txt' OVERWRITE INTO TABLE invites PARTITION (ds='2008-08-15'); ``` ##### 加载 HDFS 数据 ```sql hive&gt; LOAD DATA INPATH '/user/myname/kv2.txt' ...

    Hive开发规范及要点

    1. 从文件加载数据:使用`load data local inpath '/xxx/xxx/dim_cube.txt' overwrite into table dim.dim_cube_config;`语句可以从文件加载数据到表中。 2. 从查询语句给表插入数据:使用`insert overwrite table ...

    Hive 基本命令操作1

    `LOAD DATA`用于将文件直接加载到指定的分区,而`INSERT OVERWRITE`则可以用于插入单行或多行数据。 ```sql -- 使用LOAD DATA命令 LOAD DATA LOCAL INPATH '/path/to/file' INTO TABLE employee PARTITION (year=...

    hive数据加载导出查询

    - **INSERT OVERWRITE**: 当需要将 Hive 表中的数据导出到指定目录时,可以使用 `INSERT OVERWRITE DIRECTORY` 语法。 ```sql INSERT OVERWRITE DIRECTORY '/user/hive/output' SELECT * FROM db_0309.emp; ``` ...

    Hive常用的SQL命令操作[定义].pdf

    LOAD DATA LOCAL INPATH './examples/files/kv2.txt' OVERWRITE INTO TABLE invites PARTITION (ds='2008-08-15'); ``` 还可以加载 DFS 数据,同时给定分区信息,例如: ``` LOAD DATA INPATH '/user/myname/kv2.txt...

    hive 表 dml 操作.zip

    Hive 支持 INSERT INTO 和 INSERT OVERWRITE 两种方式。INSERT INTO 向表中追加数据,如果表中已有数据,新数据将添加到原有数据之后。而 INSERT OVERWRITE 则会覆盖原有数据,当指定某个分区时,只替换指定分区的...

    Flink on Hive构建流批一体实时数仓.pdf

    在Flink SQL中,可以使用`INSERT INTO`或`INSERT OVERWRITE`语句来实现。对于非分区表,可以直接插入数据;对于分区表,可以指定静态分区或动态分区进行写入。 例如,使用批处理模式向非分区表写入数据: ```sql ...

    HIVE文件读写

    5. 数据导出:当需要将Hive中的结果导出到文件时,可以使用INSERT OVERWRITE命令。例如,将查询结果写入一个新的CSV文件: ```sql INSERT OVERWRITE LOCAL DIRECTORY '/path/to/output' SELECT * FROM my_table ...

    hive常用命令

    2. **加载数据同时指定分区信息**: `LOAD DATA LOCAL INPATH './examples/files/kv2.txt' OVERWRITE INTO TABLE invites PARTITION (ds='2008-08-15');` - 加载数据时指定分区值,确保数据被正确地分配到对应的分区...

    Hive基本操作命令大全

    * INSERT OVERWRITE TABLE..SELECT:`FROM records2 INSERT OVERWRITE TABLE records SELECT *;` * EXPLAIN 查询:`EXPLAIN SELECT sales.*, things.* FROM sales JOIN things ON (sales.id = things.id);` Hive...

    大数据实验六实验报告:熟悉Hive的基本操作

    load data local inpath '/usr/local/hive/dividends.csv' overwrite into table dividends_unpartitioned; ``` 4. **利用Hive自动分区插入数据**: 对`dividends_unpartitioned`表执行查询并将其结果插入到`...

    Hive-工具篇_hive_

    - **插入数据**:INSERT INTO/OVERWRITE TABLE用于向表中添加数据,可一次性插入整行或整个文件。 - **查询数据**:SELECT语句用于检索数据,支持各种聚合函数、连接操作和子查询。 - **更新和删除数据**:Hive不...

    13-Hive基本操作1

    使用`INSERT INTO`或`INSERT OVERWRITE`来向表中添加数据。前者追加数据,后者覆盖原有数据。 9. **更新和删除数据**: Hive 0.14版本开始支持`UPDATE`和`DELETE`操作,但这些操作通常在支持ACID特性的表上进行,...

    大数据hive实现原理.zip_hive中怎么更新数据

    因此,在设计Hive数据模型时,应尽可能减少UPDATE和DELETE操作,优先考虑INSERT和REPLACE INTO等方法来维护数据。 在实际应用中,Hive更新数据通常涉及以下步骤: 1. **创建一个临时表**:用于存放更新后的数据。 2...

    深入浅出 Hive

    - 数据加载与删除:除了 `LOAD DATA`,还可以使用 `INSERT INTO` 或 `INSERT OVERWRITE` 语句来加载或替换数据。 - 表操作:ALTER TABLE 改变表结构,DROP TABLE 删除表,TRUNCATE TABLE 清空表。 5. HiveQL 执行...

    hive进阶强化必备

    - 使用`insert overwrite table tablename [partition(partcol1=val1, partcol2=val2)] select_statement from from_statement`命令,其中`select_statement`是从其他表中选取数据的SQL语句,`from_statement`是源...

    大数据技术基础实验报告-Hive安装配置与应用.doc

    INSERT OVERWRITE DIRECTORY '/path/to/export' SELECT * FROM employees; ``` Hive的优势在于其可扩展性、容错性和对大规模数据的处理能力。然而,由于其基于MapReduce,因此对于实时查询和低延迟操作可能不太...

    Hive用户指南 Hive user guide 中文版

    - **将数据插入到Hive表**:可以使用`INSERT INTO TABLE table_name [PARTITION (partition_spec)] SELECT ...`命令将数据从查询结果插入到Hive表中。 - **将查询结果写入文件系统**:通过`INSERT OVERWRITE ...

Global site tag (gtag.js) - Google Analytics