hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (Jira)" <j...@apache.org>
Subject [jira] [Work logged] (HIVE-23955) Classification of Error Codes in Replication
Date Thu, 06 Aug 2020 15:04:00 GMT

     [ https://issues.apache.org/jira/browse/HIVE-23955?focusedWorklogId=467346&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-467346
]

ASF GitHub Bot logged work on HIVE-23955:
-----------------------------------------

                Author: ASF GitHub Bot
            Created on: 06/Aug/20 15:03
            Start Date: 06/Aug/20 15:03
    Worklog Time Spent: 10m 
      Work Description: pkumarsinha commented on a change in pull request #1358:
URL: https://github.com/apache/hive/pull/1358#discussion_r466420110



##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/AtlasDumpTask.java
##########
@@ -132,31 +130,40 @@ private AtlasReplInfo createAtlasReplInfo() throws SemanticException,
MalformedU
 
   private long lastStoredTimeStamp() throws SemanticException {
     Path prevMetadataPath = new Path(work.getPrevAtlasDumpDir(), EximUtil.METADATA_NAME);
-    BufferedReader br = null;
+    Retryable retryable = Retryable.builder()
+      .withHiveConf(conf)
+      .withRetryOnException(IOException.class)
+      .withFailOnException(FileNotFoundException.class).build();
     try {
-      FileSystem fs = prevMetadataPath.getFileSystem(conf);
-      br = new BufferedReader(new InputStreamReader(fs.open(prevMetadataPath), Charset.defaultCharset()));
-      String line = br.readLine();
-      if (line == null) {
-        throw new SemanticException("Could not read lastStoredTimeStamp from atlas metadata
file");
-      }
-      String[] lineContents = line.split("\t", 5);
-      return Long.parseLong(lineContents[1]);
-    } catch (Exception ex) {
-      throw new SemanticException(ex);
-    } finally {
-      if (br != null) {
+      return retryable.executeCallable(() -> {
+        BufferedReader br = null;
         try {
-          br.close();
-        } catch (IOException e) {
-          throw new SemanticException(e);
+          FileSystem fs = prevMetadataPath.getFileSystem(conf);
+          br = new BufferedReader(new InputStreamReader(fs.open(prevMetadataPath), Charset.defaultCharset()));
+          String line = br.readLine();
+          if (line == null) {
+            throw new SemanticException(ErrorMsg.REPL_INVALID_CONFIG_FOR_SERVICE

Review comment:
       lastStoredTimeStamp is maintained by hive itself. Should we have better error message
category for this? 

##########
File path: common/src/java/org/apache/hadoop/hive/ql/ErrorMsg.java
##########
@@ -505,18 +505,9 @@
           " queue: {1}. Please fix and try again.", true),
   SPARK_RUNTIME_OOM(20015, "Spark job failed because of out of memory."),
 
-  //if the error message is changed for REPL_EVENTS_MISSING_IN_METASTORE, then need modification
in getNextNotification
-  //method in HiveMetaStoreClient
-  REPL_EVENTS_MISSING_IN_METASTORE(20016, "Notification events are missing in the meta store."),
-  REPL_BOOTSTRAP_LOAD_PATH_NOT_VALID(20017, "Load path {0} not valid as target database is
bootstrapped " +
-          "from some other path : {1}."),
-  REPL_FILE_MISSING_FROM_SRC_AND_CM_PATH(20018, "File is missing from both source and cm
path."),
-  REPL_LOAD_PATH_NOT_FOUND(20019, "Load path does not exist."),
-  REPL_DATABASE_IS_NOT_SOURCE_OF_REPLICATION(20020,
-          "Source of replication (repl.source.for) is not set in the database properties."),
-  REPL_INVALID_DB_OR_TABLE_PATTERN(20021,
-          "Invalid pattern for the DB or table name in the replication policy. "
-                  + "It should be a valid regex enclosed within single or double quotes."),
+  REPL_FILE_MISSING_FROM_SRC_AND_CM_PATH(20016, "File is missing from both source and cm
path."),
+  REPL_EXTERNAL_SERVICE_CONNECTION_ERROR(20017, "Failed to connect to {0} service. Error
code {1}.",
+    true),

Review comment:
       nit: Can accommodate in same line.

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/AtlasDumpTask.java
##########
@@ -42,11 +43,7 @@
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
-import java.io.BufferedReader;
-import java.io.IOException;
-import java.io.InputStream;
-import java.io.InputStreamReader;
-import java.io.Serializable;
+import java.io.*;

Review comment:
       Should we revert this?

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/atlas/AtlasRestClientImpl.java
##########
@@ -125,17 +127,15 @@ private AtlasImportResult getDefaultAtlasImportResult(AtlasImportRequest
request
     return new AtlasImportResult(request, "", "", "", 0L);
   }
 
-  public AtlasServer getServer(String endpoint) throws SemanticException {
+  public AtlasServer getServer(String endpoint, HiveConf conf) throws SemanticException {
+    Retryable retryable = Retryable.builder()
+      .withHiveConf(conf)
+      .withRetryOnException(Exception.class).build();

Review comment:
       Should we not retry on just AtlasServiceException and catch finally only that exception
as that's what getServer says to throw?

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/DirCopyTask.java
##########
@@ -43,15 +44,15 @@
  */
 public class DirCopyTask extends Task<DirCopyWork> implements Serializable {
   private static final Logger LOG = LoggerFactory.getLogger(DirCopyTask.class);
-  private static final int MAX_COPY_RETRY = 5;
 
   private boolean createAndSetPathOwner(Path destPath, Path sourcePath) throws IOException
{
     FileSystem targetFs = destPath.getFileSystem(conf);
     boolean createdDir = false;
     if (!targetFs.exists(destPath)) {
       // target path is created even if the source path is missing, so that ddl task does
not try to create it.
       if (!targetFs.mkdirs(destPath)) {
-        throw new IOException(destPath + " is not a directory or unable to create one");
+        throw new IOException(ErrorMsg.REPL_FILE_SYSTEM_OPERATION_RETRY.format(

Review comment:
       This should be retryable for communication failure and non-retryable for the permission
related issue

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/ReplDumpTask.java
##########
@@ -668,22 +666,25 @@ private void cleanFailedEventDirIfExists(Path dumpDir, long resumeFrom)
throws S
   }
 
   private long getResumeFrom(Path ackFile) throws SemanticException {
-    BufferedReader br = null;
+    Retryable retryable = Retryable.builder()
+      .withHiveConf(conf)
+      .withRetryOnException(Exception.class).build();
     try {
-      FileSystem fs = ackFile.getFileSystem(conf);
-      br = new BufferedReader(new InputStreamReader(fs.open(ackFile), Charset.defaultCharset()));
-      long lastEventID = Long.parseLong(br.readLine());
-      return lastEventID;
-    } catch (Exception ex) {
-      throw new SemanticException(ex);
-    } finally {
-      if (br != null) {
+      return retryable.executeCallable(() -> {
+        BufferedReader br = null;
         try {
-          br.close();
-        } catch (IOException e) {
-          throw new SemanticException(e);
+          FileSystem fs = ackFile.getFileSystem(conf);
+          br = new BufferedReader(new InputStreamReader(fs.open(ackFile), Charset.defaultCharset()));
+          long lastEventID = Long.parseLong(br.readLine());
+          return lastEventID;
+        } finally {
+          if (br != null) {
+            br.close();

Review comment:
       Should we ignore exception during close here as well?

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/AtlasDumpTask.java
##########
@@ -132,31 +130,40 @@ private AtlasReplInfo createAtlasReplInfo() throws SemanticException,
MalformedU
 
   private long lastStoredTimeStamp() throws SemanticException {
     Path prevMetadataPath = new Path(work.getPrevAtlasDumpDir(), EximUtil.METADATA_NAME);
-    BufferedReader br = null;
+    Retryable retryable = Retryable.builder()
+      .withHiveConf(conf)
+      .withRetryOnException(IOException.class)
+      .withFailOnException(FileNotFoundException.class).build();
     try {
-      FileSystem fs = prevMetadataPath.getFileSystem(conf);
-      br = new BufferedReader(new InputStreamReader(fs.open(prevMetadataPath), Charset.defaultCharset()));
-      String line = br.readLine();
-      if (line == null) {
-        throw new SemanticException("Could not read lastStoredTimeStamp from atlas metadata
file");
-      }
-      String[] lineContents = line.split("\t", 5);
-      return Long.parseLong(lineContents[1]);
-    } catch (Exception ex) {
-      throw new SemanticException(ex);
-    } finally {
-      if (br != null) {
+      return retryable.executeCallable(() -> {
+        BufferedReader br = null;
         try {
-          br.close();
-        } catch (IOException e) {
-          throw new SemanticException(e);
+          FileSystem fs = prevMetadataPath.getFileSystem(conf);
+          br = new BufferedReader(new InputStreamReader(fs.open(prevMetadataPath), Charset.defaultCharset()));
+          String line = br.readLine();
+          if (line == null) {
+            throw new SemanticException(ErrorMsg.REPL_INVALID_CONFIG_FOR_SERVICE
+              .format("Could not read lastStoredTimeStamp from atlas metadata file", "atlas"));
+          }
+          String[] lineContents = line.split("\t", 5);
+          return Long.parseLong(lineContents[1]);
+        } finally {
+          if (br != null) {
+            try {
+              br.close();
+            } catch (IOException e) {
+              //Do nothing
+            }
+          }
         }
-      }
+      });
+    } catch (Exception e) {
+      throw new SemanticException(ErrorMsg.REPL_RETRY_EXHAUSTED.format(e.getMessage()), e);

Review comment:
       When the Exception 'e' is of type SemanticException, we  can skip creation of another
SemanticException .

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/AtlasDumpTask.java
##########
@@ -196,12 +203,14 @@ private String checkHiveEntityGuid(AtlasRequestBuilder atlasRequestBuilder,
Stri
     AtlasObjectId objectId = atlasRequestBuilder.getItemToExport(clusterName, srcDb);
     Set<Map.Entry<String, Object>> entries = objectId.getUniqueAttributes().entrySet();
     if (entries == null || entries.isEmpty()) {
-      throw new SemanticException("Could find entries in objectId for:" + clusterName);
+      throw new SemanticException(ErrorMsg.REPL_INVALID_CONFIG_FOR_SERVICE.format("Could
find " +
+        "entries in objectId for:" + clusterName, "atlas"));

Review comment:
       Can we have one constant defined per service and use that during REPL_INVALID_CONFIG_FOR_SERVICE.format()
everywhere?

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/AtlasLoadTask.java
##########
@@ -113,26 +115,30 @@ AtlasReplInfo createAtlasReplInfo() throws SemanticException, MalformedURLExcept
 
   private String getStoredFsUri(Path atlasDumpDir) throws SemanticException {
     Path metadataPath = new Path(atlasDumpDir, EximUtil.METADATA_NAME);
-    BufferedReader br = null;
+    Retryable retryable = Retryable.builder()
+      .withHiveConf(conf)
+      .withRetryOnException(IOException.class).build();
     try {
-      FileSystem fs = metadataPath.getFileSystem(conf);
-      br = new BufferedReader(new InputStreamReader(fs.open(metadataPath), Charset.defaultCharset()));
-      String line = br.readLine();
-      if (line == null) {
-        throw new SemanticException("Could not read stored src FS Uri from atlas metadata
file");
-      }
-      String[] lineContents = line.split("\t", 5);
-      return lineContents[0];
-    } catch (Exception ex) {
-      throw new SemanticException(ex);
-    } finally {
-      if (br != null) {
+      return retryable.executeCallable(() -> {
+        BufferedReader br = null;
         try {
-          br.close();
-        } catch (IOException e) {
-          throw new SemanticException(e);
+          FileSystem fs = metadataPath.getFileSystem(conf);
+          br = new BufferedReader(new InputStreamReader(fs.open(metadataPath), Charset.defaultCharset()));
+          String line = br.readLine();
+          if (line == null) {
+            throw new SemanticException(ErrorMsg.REPL_INVALID_CONFIG_FOR_SERVICE.format("Could
not read stored " +

Review comment:
       Can we have separate Error message type for such runtime error cases cases and use
REPL_INVALID_CONFIG_FOR_SERVICE when it is to do only with CONF related errors?

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/AtlasDumpTask.java
##########
@@ -132,31 +130,40 @@ private AtlasReplInfo createAtlasReplInfo() throws SemanticException,
MalformedU
 
   private long lastStoredTimeStamp() throws SemanticException {
     Path prevMetadataPath = new Path(work.getPrevAtlasDumpDir(), EximUtil.METADATA_NAME);
-    BufferedReader br = null;
+    Retryable retryable = Retryable.builder()

Review comment:
       In createAtlasReplInfo(), we have some cases where we have wrong/no config being passed.
We should classify them at least. They aren't retryable. eg:
   String tgtDB = ReplUtils.getNonEmpty(HiveConf.ConfVars.REPL_ATLAS_REPLICATED_TO_DB.varname,
conf, errorFormat);
       String srcCluster = ReplUtils.getNonEmpty(HiveConf.ConfVars.REPL_SOURCE_CLUSTER_NAME.varname,
conf, errorFormat);
       String tgtCluster = ReplUtils.getNonEmpty(HiveConf.ConfVars.REPL_TARGET_CLUSTER_NAME.varname,
conf, errorFormat);
   

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/DirCopyTask.java
##########
@@ -86,107 +87,59 @@ private boolean checkIfPathExist(Path sourcePath, UserGroupInformation
proxyUser
     return proxyUser.doAs((PrivilegedExceptionAction<Boolean>) () -> sourcePath.getFileSystem(conf).exists(sourcePath));
   }
 
-  private int handleException(Exception e, Path sourcePath, Path targetPath,
-                              int currentRetry, UserGroupInformation proxyUser) {
-    try {
-      LOG.info("Checking if source path " + sourcePath + " is missing for exception ", e);
-      if (!checkIfPathExist(sourcePath, proxyUser)) {
-        LOG.info("Source path is missing. Ignoring exception.");
-        return 0;
-      }
-    } catch (Exception ex) {
-      LOG.warn("Source path missing check failed. ", ex);
-    }
-    // retry logic only for i/o exception
-    if (!(e instanceof IOException)) {
-      LOG.error("Unable to copy {} to {}", sourcePath, targetPath, e);
-      setException(e);
-      return ErrorMsg.getErrorMsg(e.getMessage()).getErrorCode();
-    }
-
-    if (currentRetry <= MAX_COPY_RETRY) {
-      LOG.warn("Unable to copy {} to {}", sourcePath, targetPath, e);
-    } else {
-      LOG.error("Unable to copy {} to {} even after retrying for {} time", sourcePath, targetPath,
currentRetry, e);
-      setException(e);
-      return ErrorMsg.REPL_FILE_SYSTEM_OPERATION_RETRY.getErrorCode();
-    }
-    int sleepTime = FileUtils.getSleepTime(currentRetry);
-    LOG.info("Sleep for " + sleepTime + " milliseconds before retry no " + (currentRetry));
-    try {
-      Thread.sleep(sleepTime);
-    } catch (InterruptedException timerEx) {
-      LOG.info("Sleep interrupted", timerEx.getMessage());
-    }
-    try {
-      if (proxyUser == null) {
-        proxyUser = Utils.getUGI();
-      }
-      FileSystem.closeAllForUGI(proxyUser);
-    } catch (Exception ex) {
-      LOG.warn("Unable to closeAllForUGI for user " + proxyUser, ex);
-    }
-    return ErrorMsg.getErrorMsg(e.getMessage()).getErrorCode();
-  }
-
   @Override
   public int execute() {
     String distCpDoAsUser = conf.getVar(HiveConf.ConfVars.HIVE_DISTCP_DOAS_USER);
+    Retryable retryable = Retryable.builder()
+      .withHiveConf(conf)
+      .withRetryOnException(IOException.class).build();
+    try {
+      return retryable.executeCallable(() -> {
+        UserGroupInformation proxyUser = null;
+        Path sourcePath = work.getFullyQualifiedSourcePath();
+        Path targetPath = work.getFullyQualifiedTargetPath();
+        try {
+          if (conf.getBoolVar(HiveConf.ConfVars.REPL_ADD_RAW_RESERVED_NAMESPACE)) {
+            sourcePath = reservedRawPath(work.getFullyQualifiedSourcePath().toUri());
+            targetPath = reservedRawPath(work.getFullyQualifiedTargetPath().toUri());
+          }
+          UserGroupInformation ugi = Utils.getUGI();
+          String currentUser = ugi.getShortUserName();
+          if (distCpDoAsUser != null && !currentUser.equals(distCpDoAsUser)) {
+            proxyUser = UserGroupInformation.createProxyUser(
+              distCpDoAsUser, UserGroupInformation.getLoginUser());
+          }
 
-    Path sourcePath = work.getFullyQualifiedSourcePath();
-    Path targetPath = work.getFullyQualifiedTargetPath();
-    if (conf.getBoolVar(HiveConf.ConfVars.REPL_ADD_RAW_RESERVED_NAMESPACE)) {
-      sourcePath = reservedRawPath(work.getFullyQualifiedSourcePath().toUri());
-      targetPath = reservedRawPath(work.getFullyQualifiedTargetPath().toUri());
-    }
-    int currentRetry = 0;
-    int error = 0;
-    UserGroupInformation proxyUser = null;
-    while (currentRetry <= MAX_COPY_RETRY) {
-      try {
-        UserGroupInformation ugi = Utils.getUGI();
-        String currentUser = ugi.getShortUserName();
-        if (distCpDoAsUser != null && !currentUser.equals(distCpDoAsUser)) {
-          proxyUser = UserGroupInformation.createProxyUser(
-                  distCpDoAsUser, UserGroupInformation.getLoginUser());
-        }
-
-        setTargetPathOwner(targetPath, sourcePath, proxyUser);
-
-        // do we create a new conf and only here provide this additional option so that we
get away from
-        // differences of data in two location for the same directories ?
-        // basically add distcp.options.delete to hiveconf new object ?
-        FileUtils.distCp(
-                sourcePath.getFileSystem(conf), // source file system
-                Collections.singletonList(sourcePath),  // list of source paths
-                targetPath,
-                false,
-                proxyUser,
-                conf,
-                ShimLoader.getHadoopShims());
-        return 0;
-      } catch (Exception e) {
-        currentRetry++;
-        error = handleException(e, sourcePath, targetPath, currentRetry, proxyUser);
-        if (error == 0) {
-          return 0;
-        }
-      } finally {
-        if (proxyUser != null) {
+          setTargetPathOwner(targetPath, sourcePath, proxyUser);
           try {
-            FileSystem.closeAllForUGI(proxyUser);
-          } catch (IOException e) {
-            LOG.error("Unable to closeAllForUGI for user " + proxyUser, e);
-            if (error == 0) {
-              setException(e);
-              error = ErrorMsg.getErrorMsg(e.getMessage()).getErrorCode();
+            if (!checkIfPathExist(sourcePath, proxyUser)) {
+              LOG.info("Source path is missing. Ignoring exception.");
+              return 0;
             }
-            break;
+          } catch (Exception ex) {
+            LOG.warn("Source path missing check failed. ", ex);

Review comment:
       There is a change in behavior here. When we have "Source path missing check failed.
" we are still going for DistCp, earlier distCp wasn't attempted in this case only retry was.


##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/AtlasDumpTask.java
##########
@@ -177,7 +184,7 @@ long dumpAtlasMetaData(AtlasRequestBuilder atlasRequestBuilder, AtlasReplInfo
at
     } catch (SemanticException ex) {
       throw ex;
     } catch (Exception ex) {
-      throw new SemanticException(ex);
+      throw new SemanticException(ex.getMessage(), ex);
     } finally {
       if (inputStream != null) {

Review comment:
       At this point, we have  some  thing like this:
   if (inputStream != null) {
           try {
             inputStream.close();
           } catch (IOException e) {
             throw new SemanticException(e);
           }
         }
   Since we are ignoring the exception during close in lastStoredTimeStamp(), let's align
this also.

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/RangerDumpTask.java
##########
@@ -92,14 +92,21 @@ public int execute() {
       }
       URL url = work.getRangerConfigResource();
       if (url == null) {
-        throw new SemanticException("Ranger configuration is not valid "
-          + ReplUtils.RANGER_CONFIGURATION_RESOURCE_NAME);
+        throw new SemanticException(ErrorMsg.REPL_INVALID_CONFIG_FOR_SERVICE
+          .format("Ranger configuration is not valid "
+          + ReplUtils.RANGER_CONFIGURATION_RESOURCE_NAME, "ranger"));
       }
       conf.addResource(url);
       String rangerHiveServiceName = conf.get(ReplUtils.RANGER_HIVE_SERVICE_NAME);
       String rangerEndpoint = conf.get(ReplUtils.RANGER_REST_URL);
-      if (StringUtils.isEmpty(rangerEndpoint) || !rangerRestClient.checkConnection(rangerEndpoint,
conf)) {
-        throw new SemanticException("Ranger endpoint is not valid " + rangerEndpoint);
+      if (StringUtils.isEmpty(rangerEndpoint)) {
+        throw new SemanticException(ErrorMsg.REPL_INVALID_CONFIG_FOR_SERVICE
+          .format("Ranger endpoint is not valid "
+            + rangerEndpoint, "ranger"));
+      }
+      if (!rangerRestClient.checkConnection(rangerEndpoint, conf)) {
+        throw new SemanticException(ErrorMsg.REPL_EXTERNAL_SERVICE_CONNECTION_ERROR.format("ranger",
+          "Ranger endpoint is not valid " + rangerEndpoint));

Review comment:
       The error message could be bit different like "Ranger endpoint is either invalid or
not reachable"

##########
File path: ql/src/java/org/apache/hadoop/hive/ql/exec/repl/atlas/AtlasRestClientBuilder.java
##########
@@ -92,7 +94,7 @@ private void initializeAtlasApplicationProperties() throws SemanticException
{
       props.setProperty(ATLAS_PROPERTY_AUTH_KERBEROS, "true");
       ApplicationProperties.set(ConfigurationConverter.getConfiguration(props));
     } catch (AtlasException e) {
-      throw new SemanticException(e);
+      throw new SemanticException(ErrorMsg.REPL_INVALID_CONFIG_FOR_SERVICE.format(), e);

Review comment:
       Service name not needed, here?




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Issue Time Tracking
-------------------

    Worklog Id:     (was: 467346)
    Time Spent: 20m  (was: 10m)

> Classification of Error Codes in Replication
> --------------------------------------------
>
>                 Key: HIVE-23955
>                 URL: https://issues.apache.org/jira/browse/HIVE-23955
>             Project: Hive
>          Issue Type: Task
>            Reporter: Aasha Medhi
>            Assignee: Aasha Medhi
>            Priority: Major
>              Labels: pull-request-available
>         Attachments: HIVE-23955.01.patch, HIVE-23955.02.patch, Retry Logic for Replication.pdf
>
>          Time Spent: 20m
>  Remaining Estimate: 0h
>




--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Mime
View raw message