(2nd round) fix under fs and glusterfs configuration issues

[rootfs]

This is an enhanced fix to #623. Previous fix failed on TestDFSIO test because in that test, the fs.defaultFS property is already set. If so, then avoid overriding this property. This new fix passed the tests.

When tachyon is used for MapReduce jobs, under fs may not be properly initialized.
This fix is to set those properties before under fs initialize() is called. If the properties are already set in core-site.xml, then don't overwrite them.

On my glutster setup, a wordcount test was run to validate this patch. Separately, @pfxuan validated part of the fix on his setup using orangefs as under fs.

The core-site.xml in use can be found at http://pastebin.com/WuR6Xn7v

The wordcount test output can be found at http://pastebin.com/aJJVSBK9

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Tachyon-Pull-Request-Builder/2624/
Test PASSed.

[haoyuan]

Thanks!

[haoyuan]

Seems still doesn't work. https://amplab.cs.berkeley.edu/jenkins/job/Tachyon-Snapshot/41/console

[rootfs]

@haoyuan not sure what is the problem but the test running well on my side
http://pastebin.com/izx3J7r3

[rootfs]

@haoyuan Run another test, all tests passed.

Regarding this failed test, the assert failure is that the usedbytes were expected to be 0. But with other tests creating tables on workers, the usedbytes could be non-zero.

public void deleteFileTest() throws IOException {
String uniqPath = TestUtils.uniqPath();
List workers = sTfs.getWorkersInfo();
Assert.assertEquals(1, workers.size());
Assert.assertEquals(WORKER_CAPACITY_BYTES, workers.get(0).getCapacityBytes());
Assert.assertEquals(0, workers.get(0).getUsedBytes()); <--- failed assert