/**
* test 1. create DFS cluster with 3 storage directories - 2 EDITS_IMAGE, 1 EDITS 2. create a
* cluster and write a file 3. corrupt/disable one storage (or two) by removing 4. run
* doCheckpoint - it will fail on removed dirs (which will invalidate the storages) 5. write
* another file 6. check that edits and fsimage differ 7. run doCheckpoint 8. verify that all the
* image and edits files are the same.
*/
@Test
public void testStorageRestore() throws Exception {
int numDatanodes = 2;
cluster =
new MiniDFSCluster(0, config, numDatanodes, true, false, true, null, null, null, null);
cluster.waitActive();
SecondaryNameNode secondary = new SecondaryNameNode(config);
FileSystem fs = cluster.getFileSystem();
Path path = new Path("/", "test");
writeFile(fs, path, 2);
invalidateStorage(cluster.getNameNode().getFSImage());
path = new Path("/", "test1");
writeFile(fs, path, 2);
checkFiles(false);
secondary.doCheckpoint();
checkFiles(true);
secondary.shutdown();
cluster.shutdown();
}
/** Create a connection to the primary namenode. */
public SecondaryNameNode(Configuration conf) throws IOException {
try {
initialize(conf);
} catch (IOException e) {
shutdown();
throw e;
}
}
/**
* Test to simulate interleaved checkpointing by 2 2NNs after a storage directory has been taken
* offline. The first will cause the directory to come back online, but it won't have any valid
* contents. The second 2NN will then try to perform a checkpoint. The NN should not serve up the
* image or edits from the restored (empty) dir.
*/
@Test
public void testCheckpointWithRestoredDirectory() throws IOException {
SecondaryNameNode secondary = null;
try {
cluster = new MiniDFSCluster(0, config, 1, true, false, true, null, null, null, null);
cluster.waitActive();
secondary = new SecondaryNameNode(config);
FSImage fsImage = cluster.getNameNode().getFSImage();
FileSystem fs = cluster.getFileSystem();
Path path1 = new Path("/", "test");
writeFile(fs, path1, 2);
// Take name3 offline
fsImage.getEditLog().removeEditsAndStorageDir(2);
// Simulate a 2NN beginning a checkpoint, but not finishing. This will
// cause name3 to be restored.
cluster.getNameNode().rollEditLog();
// Now another 2NN comes along to do a full checkpoint.
secondary.doCheckpoint();
// The created file should still exist in the in-memory FS state after the
// checkpoint.
assertTrue("File missing after checkpoint", fs.exists(path1));
secondary.shutdown();
// Restart the NN so it reloads the edits from on-disk.
cluster.restartNameNode();
// The created file should still exist after the restart.
assertTrue("path should still exist after restart", fs.exists(path1));
} finally {
if (cluster != null) {
cluster.shutdown();
}
if (secondary != null) {
secondary.shutdown();
}
}
}
/** Check whether the secondary name-node can be started. */
private boolean canStartSecondaryNode(Configuration conf) throws IOException {
SecondaryNameNode sn = null;
try {
sn = new SecondaryNameNode(conf);
} catch (IOException e) {
if (e instanceof java.net.BindException) return false;
throw e;
}
sn.shutdown();
return true;
}
@Override
protected void doDelete(SecondaryNameNode service) throws Exception {
service.shutdown();
}