HDFS-11796. Ozone: MiniOzoneCluster should set "ozone.handler.type" key correctly. Contributed by Mukul Kumar Singh.
This commit is contained in:
parent
aa5722dc44
commit
1b2d0b4fec
|
@ -73,18 +73,23 @@ public class KeyManagerImpl implements KeyManager {
|
|||
throw new KSMException("Bucket not found",
|
||||
KSMException.ResultCodes.FAILED_BUCKET_NOT_FOUND);
|
||||
}
|
||||
// TODO throw exception if key exists, may change to support key
|
||||
// overwrite in the future
|
||||
//Check if key already exists.
|
||||
if(metadataManager.get(keyKey) != null) {
|
||||
LOG.error("key already exist: {}/{}/{} ", volumeName, bucketName,
|
||||
keyName);
|
||||
throw new KSMException("Key already exist",
|
||||
KSMException.ResultCodes.FAILED_KEY_ALREADY_EXISTS);
|
||||
}
|
||||
|
||||
// TODO: Garbage collect deleted blocks due to overwrite of a key.
|
||||
// FIXME: BUG: Please see HDFS-11922.
|
||||
// If user overwrites a key, then we are letting it pass without
|
||||
// corresponding process.
|
||||
// In reality we need to garbage collect those blocks by telling SCM to
|
||||
// clean up those blocks when it can. Right now making this change
|
||||
// allows us to pass tests that expect ozone can overwrite a key.
|
||||
|
||||
|
||||
// When we talk to SCM make sure that we ask for at least a byte in the
|
||||
// block. This way even if the call is for a zero length key, we back it
|
||||
// with a actual SCM block.
|
||||
// TODO : Review this decision later. We can get away with only a
|
||||
// metadata entry in case of 0 length key.
|
||||
AllocatedBlock allocatedBlock =
|
||||
scmBlockClient.allocateBlock(args.getDataSize());
|
||||
scmBlockClient.allocateBlock(Math.max(args.getDataSize(), 1));
|
||||
KsmKeyInfo keyBlock = new KsmKeyInfo.Builder()
|
||||
.setVolumeName(args.getVolumeName())
|
||||
.setBucketName(args.getBucketName())
|
||||
|
@ -98,7 +103,7 @@ public class KeyManagerImpl implements KeyManager {
|
|||
LOG.debug("Key {} allocated in volume {} bucket {}",
|
||||
keyName, volumeName, bucketName);
|
||||
return keyBlock;
|
||||
} catch (DBException ex) {
|
||||
} catch (Exception ex) {
|
||||
LOG.error("Key allocation failed for volume:{} bucket:{} key:{}",
|
||||
volumeName, bucketName, keyName, ex);
|
||||
throw new KSMException(ex.getMessage(),
|
||||
|
|
|
@ -423,6 +423,8 @@ public final class MiniOzoneCluster extends MiniDFSCluster
|
|||
if (!ozoneHandlerType.isPresent()) {
|
||||
throw new IllegalArgumentException(
|
||||
"The Ozone handler type must be specified.");
|
||||
} else {
|
||||
conf.set(OzoneConfigKeys.OZONE_HANDLER_TYPE_KEY, ozoneHandlerType.get());
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
@ -360,12 +360,14 @@ public class TestKeySpaceManager {
|
|||
try (OutputStream stream = storageHandler.newKeyWriter(keyArgs)) {
|
||||
stream.write(dataString.getBytes());
|
||||
}
|
||||
// try to put the same keyArg, should raise KEY_ALREADY_EXISTS exception
|
||||
exception.expect(IOException.class);
|
||||
exception.expectMessage("KEY_ALREADY_EXISTS");
|
||||
|
||||
// We allow the key overwrite to be successful. Please note : Till HDFS-11922
|
||||
// is fixed this causes a data block leak on the data node side. That is
|
||||
// this overwrite only overwrites the keys on KSM. We need to garbage
|
||||
// collect those blocks from datanode.
|
||||
KeyArgs keyArgs2 = new KeyArgs(volumeName, bucketName, keyName, userArgs);
|
||||
storageHandler.newKeyWriter(keyArgs2);
|
||||
Assert.assertEquals(1 + numKeyAllocateFails,
|
||||
Assert.assertEquals(numKeyAllocateFails,
|
||||
ksmMetrics.getNumKeyAllocateFails());
|
||||
}
|
||||
|
||||
|
|
|
@ -62,6 +62,7 @@ public class TestOzoneRestWithMiniCluster {
|
|||
conf = new OzoneConfiguration();
|
||||
cluster = new MiniOzoneCluster.Builder(conf).numDataNodes(1)
|
||||
.setHandlerType(OzoneConsts.OZONE_HANDLER_DISTRIBUTED).build();
|
||||
cluster.waitOzoneReady();
|
||||
ozoneClient = cluster.createOzoneClient();
|
||||
}
|
||||
|
||||
|
|
Loading…
Reference in New Issue