Merge druid-api, druid-common and java-util modules into druid-common

[leventov]

I had to collapse druid-api, druid-common and java-util, to avoid cyclic dependencies. It was a part of #4312 plan.

First part of #5488.

[b-slim]

can we fix this TODO?

[leventov]

I'm not going to do this in this PR

[b-slim]

okay, can we add more context to what the TODO means and link it to an issue?

[gianm]

I think it makes sense, in this PR, to change nothing about the content and just collapse the modules.

[leventov]

@b-slim previous comment was a lengthy explanation why those config classes couldn't be merged at that time. Now they could. I didn't want to add explanations why it should be done, because previously it wasn't as well (it is considered obvious, I suppose.)

Anyway, in general we don't usually add TODOs and favor issues on github; I've created #5491.

[b-slim]

looks good to me, but there are some merge conflicts.

[jihoonson]

@leventov would you resolve conflicts? I think this PR should be merged as soon as possible to avoid further conflicts.

[jihoonson]

This patch looks good to me. I'll finish my review if CI is fixed.

[leventov]

Unfortunately I don't know how to fix the CI. That are strange problems that I cannot reproduce locally. Any ideas welcome.

[gianm]

The failure in LoggingRequestLoggerTest looks like a bug in the test. It has this code,

  @BeforeClass
  public static void setUpStatic() throws Exception
  {
    appender = OutputStreamAppender
        .newBuilder()
        .setName("test stream")
        .setTarget(baos)
        .setLayout(JsonLayout.createLayout(false, true, false, true, true, Charsets.UTF_8))
        .build();
    final Logger logger = (Logger)
        LogManager.getLogger(LoggingRequestLogger.class);
    appender.start();
    logger.addAppender(appender);
  }

  @After
  public void tearDown()
  {
    baos.reset();
  }

But this is wrong. baos is created in a static @BeforeClass setup method but is cleared in a per-test @After method. It leads to the "No content" error from Jackson. I'm honestly not sure how this has ever passed properly. Maybe some test configuration had previously been doing forks per test and now isn't?

[gianm]

I'm not sure what causes the other, docker integration test failure.

[jihoonson]

I can reproduce the IT test failure locally. I'm still not sure what's the problem but I could see every Druid servers work well.

[jihoonson]

@leventov ok. The unit test failures in LoggingRequestLoggerTest are because of the wrong log level. I don't know why, but the log level is set to ERROR and INFO level logs are not written.

The integration tests failure looks somehow related to service discovery. I can see these logs a lot in the overlord log. I checked mysql, zookeeper, and druid coordinator were running well.

2018-03-07T20:57:02,601 ERROR [BasicAuthenticatorCacheManager-Exec--0] io.druid.curator.discovery.ServerDiscoverySelector - No server instance found for [druid/coordinator]

BTW, you should be able to reproduce the test failures locally by deleting the maven local cache and installing druid.

[leventov]

I give up trying to do this