HDFS-9533. seen_txid in the shared edits directory is modified during bootstrapping. Contributed by Kihwal Lee.

(cherry picked from commit 5cb1e0118b)

Conflicts:
	hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/ha/BootstrapStandby.java
	hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/ha/TestBootstrapStandby.java
(cherry picked from commit 3c0ae5914c)
This commit is contained in:
Kihwal Lee 2015-12-17 17:23:55 -06:00
parent 8812be3dc1
commit e5e5ba495f
5 changed files with 42 additions and 3 deletions

View File

@ -1629,6 +1629,9 @@ Release 2.7.3 - UNRELEASED
HDFS-9516. Truncate file fails with data dirs on multiple disks.
(Plamen Jeliazkov via shv)
HDFS-9533. seen_txid in the shared edits directory is modified during
bootstrapping (kihwal)
Release 2.7.2 - UNRELEASED
INCOMPATIBLE CHANGES

View File

@ -482,8 +482,24 @@ long getMostRecentCheckpointTime() {
* @param txid the txid that has been reached
*/
public void writeTransactionIdFileToStorage(long txid) {
writeTransactionIdFileToStorage(txid, null);
}
/**
* Write a small file in all available storage directories that
* indicates that the namespace has reached some given transaction ID.
*
* This is used when the image is loaded to avoid accidental rollbacks
* in the case where an edit log is fully deleted but there is no
* checkpoint. See TestNameEditsConfigs.testNameEditsConfigsFailure()
* @param txid the txid that has been reached
* @param type the type of directory
*/
public void writeTransactionIdFileToStorage(long txid,
NameNodeDirType type) {
// Write txid marker in all storage directories
for (StorageDirectory sd : storageDirs) {
for (Iterator<StorageDirectory> it = dirIterator(type); it.hasNext();) {
StorageDirectory sd = it.next();
try {
writeTransactionIdFile(sd, txid);
} catch(IOException e) {

View File

@ -52,6 +52,7 @@
import org.apache.hadoop.hdfs.server.namenode.FSNamesystem;
import org.apache.hadoop.hdfs.server.namenode.NNStorage;
import org.apache.hadoop.hdfs.server.namenode.NNStorage.NameNodeFile;
import org.apache.hadoop.hdfs.server.namenode.NNStorage.NameNodeDirType;
import org.apache.hadoop.hdfs.server.namenode.NNUpgradeUtil;
import org.apache.hadoop.hdfs.server.namenode.NameNode;
import org.apache.hadoop.hdfs.server.namenode.TransferFsImage;
@ -313,13 +314,14 @@ private int downloadImage(NNStorage storage, NamenodeProtocol proxy)
return ERR_CODE_LOGS_UNAVAILABLE;
}
image.getStorage().writeTransactionIdFileToStorage(curTxId);
// Download that checkpoint into our storage directories.
MD5Hash hash = TransferFsImage.downloadImageToStorage(
otherHttpAddr, imageTxId, storage, true, true);
image.saveDigestAndRenameCheckpointImage(NameNodeFile.IMAGE, imageTxId,
hash);
// Write seen_txid to the formatted image directories.
storage.writeTransactionIdFileToStorage(imageTxId, NameNodeDirType.IMAGE);
} catch (IOException ioe) {
throw ioe;
} finally {

View File

@ -568,4 +568,11 @@ public static void assertNNFilesMatch(MiniDFSCluster cluster) throws Exception {
FSImageTestUtil.assertParallelFilesAreIdentical(curDirs,
ignoredFiles);
}
public static long getStorageTxId(NameNode node, URI storageUri)
throws IOException {
StorageDirectory sDir = getFSImage(node).getStorage().
getStorageDirectory(storageUri);
return NNStorage.readTransactionIdFile(sDir);
}
}

View File

@ -135,6 +135,13 @@ public void testDownloadingLaterCheckpoint() throws Exception {
.getFSImage().getMostRecentCheckpointTxId();
assertEquals(6, expectedCheckpointTxId);
// advance the current txid
cluster.getFileSystem(0).create(new Path("/test_txid"), (short)1).close();
// obtain the content of seen_txid
URI editsUri = cluster.getSharedEditsDir(0, 1);
long seen_txid_shared = FSImageTestUtil.getStorageTxId(nn0, editsUri);
int rc = BootstrapStandby.run(
new String[]{"-force"},
cluster.getConfiguration(1));
@ -145,6 +152,10 @@ public void testDownloadingLaterCheckpoint() throws Exception {
ImmutableList.of((int)expectedCheckpointTxId));
FSImageTestUtil.assertNNFilesMatch(cluster);
// Make sure the seen_txid was not modified by the standby
assertEquals(seen_txid_shared,
FSImageTestUtil.getStorageTxId(nn0, editsUri));
// We should now be able to start the standby successfully.
cluster.restartNameNode(1);
}