@Test
public void testGetCompanionBLocks() throws IOException {
try {
setupCluster(false, 1L, racks1, hosts1);
String[] files = new String[] {"/dir/file1", "/dir/file2", "/dir/file3"};
Codec codec = Codec.getCodec("rs");
for (String file : files) {
TestRaidDfs.createTestFile(fs, new Path(file), 3, 2, 8192L);
}
FileStatus stat = fs.getFileStatus(new Path("/dir"));
RaidNode.doRaid(
conf,
stat,
new Path(codec.parityDirectory),
codec,
new RaidNode.Statistics(),
RaidUtils.NULL_PROGRESSABLE,
false,
1,
1);
Collection<LocatedBlock> companionBlocks;
for (int i = 0; i < 2; i++) {
for (int j = 0; j < 2; j++) {
companionBlocks =
getCompanionBlocks(
namesystem, policy, getBlocks(namesystem, files[i]).get(j).getBlock());
Assert.assertEquals(8, companionBlocks.size());
}
}
companionBlocks =
getCompanionBlocks(namesystem, policy, getBlocks(namesystem, files[2]).get(0).getBlock());
Assert.assertEquals(8, companionBlocks.size());
companionBlocks =
getCompanionBlocks(namesystem, policy, getBlocks(namesystem, files[2]).get(1).getBlock());
Assert.assertEquals(4, companionBlocks.size());
String parityFile = "/raidrs/dir";
for (int i = 0; i < 3; i++) {
companionBlocks =
getCompanionBlocks(
namesystem, policy, getBlocks(namesystem, parityFile).get(i).getBlock());
Assert.assertEquals(8, companionBlocks.size());
}
for (int i = 3; i < 6; i++) {
companionBlocks =
getCompanionBlocks(
namesystem, policy, getBlocks(namesystem, parityFile).get(i).getBlock());
Assert.assertEquals(4, companionBlocks.size());
}
} finally {
closeCluster();
}
}
@After
public void tearDown() throws Exception {
if (rnode != null) {
rnode.stop();
rnode.join();
rnode = null;
}
if (cluster != null) {
cluster.shutdown();
cluster = null;
}
dfs = null;
LOG.info("Test cluster shut down");
}
/*
* This test start datanodes with simulated mode and keep running
* chooseReplicaToDelete multiple times to get the average processing time
* and number of allocated objects
*/
@Test
public void testDirXORChooseReplicasToDeletePerformance() throws Exception {
try {
setupCluster(true, 1L, racks1, hosts1);
// create test files
int numFiles = 1000;
long blockSize = 1024L;
String parentDir = "/dir/";
for (int i = 0; i < numFiles; i++) {
String file = parentDir + "file" + i;
TestRaidDfs.createTestFile(fs, new Path(file), 3, 1, blockSize);
}
LOG.info("Created " + numFiles + " files");
Codec code = Codec.getCodec("xor");
FSNamesystem fsNameSys = cluster.getNameNode().namesystem;
for (DatanodeDescriptor dd : fsNameSys.datanodeMap.values()) {
LOG.info(dd);
}
// create fake parity file
long numStripes = RaidNode.numStripes(numFiles, code.stripeLength);
TestRaidDfs.createTestFile(
fs,
new Path(code.parityDirectory, "dir"),
3,
(int) numStripes * code.parityLength,
blockSize);
long startTime = System.currentTimeMillis();
long total = 0L;
fsNameSys.readLock();
for (BlocksMap.BlockInfo bi : fsNameSys.blocksMap.getBlocks()) {
fsNameSys.replicator.chooseReplicaToDelete(
bi.getINode(),
bi,
(short) 3,
fsNameSys.datanodeMap.values(),
new ArrayList<DatanodeDescriptor>());
total++;
}
fsNameSys.readUnlock();
LOG.info(
"Average chooseReplicaToDelete time: "
+ ((double) (System.currentTimeMillis() - startTime) / total));
} finally {
closeCluster();
}
}
/*
* This test creates a directory with 3 files and its fake parity file.
* We decommissioned all nodes in the rack2 to make sure all data are stored
* in rack1 machine.
* Then we bring rack2 machines to normal state and create a non-raided file
* which is too small to be raided in the directory with 4 replicas
* (1 in rack1 and 3 in rack2).
* Then we reduce the replication to 3 to trigger chooseReplicatToDelete.
* We verify remaining replicas has 1 in rack1 and 2 in rack2.
*/
@Test
public void testChooseReplicasToDeleteForSmallFile() throws Exception {
try {
setupCluster(false, 512L, racks2, hosts2);
// create test files
int numFiles = 4;
long blockSize = 1024L;
String parentDir = "/dir/";
DFSClient client = getDfsClient(cluster.getNameNode(), conf);
DatanodeInfo[] infos = client.datanodeReport(DatanodeReportType.LIVE);
ArrayList<String> rack2nodes = new ArrayList<String>();
ArrayList<DatanodeInfo> rack2di = new ArrayList<DatanodeInfo>();
for (DatanodeInfo di : infos) {
if (di.getHostName().contains("rack2")) {
rack2nodes.add(di.getName());
rack2di.add(cluster.getNameNode().namesystem.getDatanode(di));
}
}
LOG.info("Decommission rack2 nodes");
writeConfigFile(excludeFile, rack2nodes);
cluster.getNameNode().namesystem.refreshNodes(conf);
waitState(rack2di, AdminStates.DECOMMISSIONED);
for (int i = 0; i < numFiles; i++) {
if (i == 2) {
continue;
}
String file = parentDir + "file" + i;
Path filePath = new Path(file);
TestRaidDfs.createTestFile(fs, filePath, 1, 1, blockSize);
printLocatedBlocks(filePath);
}
LOG.info("Created " + (numFiles - 1) + " files");
// create fake parity file
Codec code = Codec.getCodec("xor");
long numStripes = RaidNode.numStripes(numFiles, code.stripeLength);
Path parityPath = new Path(code.parityDirectory, "dir");
TestRaidDfs.createTestFile(
fs, parityPath, 1, (int) numStripes * code.parityLength, blockSize);
LOG.info("Create parity file: " + parityPath);
printLocatedBlocks(parityPath);
LOG.info("Bring back rack2 nodes out of decommission");
writeConfigFile(excludeFile, null);
cluster.getNameNode().namesystem.refreshNodes(conf);
waitState(rack2di, AdminStates.NORMAL);
Path smallFilePath = new Path(parentDir + "file2");
TestRaidDfs.createTestFile(fs, smallFilePath, 4, 1, 256L);
assertEquals(
"all datanodes should have replicas", hosts2.length, printLocatedBlocks(smallFilePath));
LOG.info("Created small file: " + smallFilePath);
LOG.info("Reduce replication to 3");
dfs.setReplication(smallFilePath, (short) 3);
long startTime = System.currentTimeMillis();
while (System.currentTimeMillis() - startTime < 120000
&& printLocatedBlocks(smallFilePath) == 4) {
Thread.sleep(1000);
}
LocatedBlocks lbs = dfs.getLocatedBlocks(smallFilePath, 0L, Integer.MAX_VALUE);
boolean hasRack1 = false;
for (DatanodeInfo di : lbs.getLocatedBlocks().get(0).getLocations()) {
if (di.getNetworkLocation().contains("rack1")) {
hasRack1 = true;
break;
}
}
assertTrue("We should keep the nodes in rack1", hasRack1);
} finally {
closeCluster();
}
}
/** raids test file */
private void raidTestFiles(Path raidPath, Path[] filePaths, boolean doHar)
throws IOException, ClassNotFoundException {
// create RaidNode
raidConf = new Configuration(conf);
raidConf.set(RaidNode.RAID_LOCATION_KEY, RAID_DIR);
raidConf.setInt("raid.blockfix.interval", 1000);
// the RaidNode does the raiding inline (instead of submitting to MR node)
conf.set("raid.classname", "org.apache.hadoop.raid.LocalRaidNode");
rnode = RaidNode.createRaidNode(null, raidConf);
for (Path filePath : filePaths) {
long waitStart = System.currentTimeMillis();
boolean raided = false;
Path parityFilePath = new Path(RAID_DIR, filePath.toString().substring(1));
while (!raided) {
try {
FileStatus[] listPaths = dfs.listStatus(raidPath);
if (listPaths != null) {
if (doHar) {
// case with HAR
for (FileStatus f : listPaths) {
if (f.getPath().toString().endsWith(".har")) {
// check if the parity file is in the index
final Path indexPath = new Path(f.getPath(), "_index");
final FileStatus indexFileStatus = dfs.getFileStatus(indexPath);
final HarIndex harIndex =
new HarIndex(dfs.open(indexPath), indexFileStatus.getLen());
final HarIndex.IndexEntry indexEntry =
harIndex.findEntryByFileName(parityFilePath.toString());
if (indexEntry != null) {
LOG.info(
"raid file "
+ parityFilePath.toString()
+ " found in Har archive: "
+ f.getPath().toString()
+ " ts="
+ indexEntry.mtime);
raided = true;
break;
}
}
}
} else {
// case without HAR
for (FileStatus f : listPaths) {
Path found = new Path(f.getPath().toUri().getPath());
if (parityFilePath.equals(found)) {
LOG.info("raid file found: " + f.getPath().toString());
raided = true;
break;
}
}
}
}
} catch (FileNotFoundException ignore) {
}
if (!raided) {
if (System.currentTimeMillis() > waitStart + 40000L) {
LOG.error("parity file not created after 40s");
throw new IOException("parity file not HARed after 40s");
} else {
try {
Thread.sleep(1000);
} catch (InterruptedException ignore) {
}
}
}
}
}
rnode.stop();
rnode.join();
rnode = null;
LOG.info("test file raided");
}