[MESOS-3287] downloadWithHadoop tries to access Error() for a valid Try<bool>

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.24.0
• Labels: hdfs mesosphere
• Sprint: Mesosphere Sprint 17
• Story Points: 1
• Assignee: Marco Massenzio
• Reporter: Marco Massenzio
• Created: 18/Aug/15 07:47
• Updated: 26/Nov/18 12:22
• Resolved: 26/Nov/18 12:22

Description

This was reported while trying to install Hadoop / Mesos integration:

I0818 05:36:35.058688 24428 fetcher.cpp:409] Fetcher Info: {"cache_directory":"\/tmp\/mesos\/fetch\/slaves\/20150706-075218-1611773194-5050-28439-S473\/hadoop","items":[{"action":"BYPASS_CACHE","uri":{"extract":true,"value":"hdfs:\/\/hdfs.prod:54310\/user\/ashwanth\/hadoop-with-mesos-2.6.0-cdh5.4.4.tar.gz"}}],"sandbox_directory":"\/var\/lib\/mesos\/slaves\/20150706-075218-1611773194-5050-28439-S473\/frameworks\/20150706-075218-1611773194-5050-28439-4532\/executors\/executor_Task_Tracker_4129\/runs\/c26f52d4-4055-46fa-b999-11d73f2096dd","user":"hadoop"}
I0818 05:36:35.059806 24428 fetcher.cpp:364] Fetching URI 'hdfs://hdfs.prod:54310/user/ashwanth/hadoop-with-mesos-2.6.0-cdh5.4.4.tar.gz'
I0818 05:36:35.059821 24428 fetcher.cpp:238] Fetching directly into the sandbox directory
I0818 05:36:35.059835 24428 fetcher.cpp:176] Fetching URI 'hdfs://hdfs.prod:54310/user/ashwanth/hadoop-with-mesos-2.6.0-cdh5.4.4.tar.gz'
mesos-fetcher: /tmp/mesos-build/mesos-repo/3rdparty/libprocess/3rdparty/stout/include/stout/try.hpp:90: const string& Try<T>::error() const [with T = bool; std::string = std::basic_string<char>]: Assertion `data.isNone()' failed.

This is, however, a genuine bug in src/launcher/fetcher.cpp#L99:

  Try<bool> available = hdfs.available();

  if (available.isError() || !available.get()) {
    return Error("Skipping fetch with Hadoop Client as"
                 " Hadoop Client not available: " + available.error());
  }

The root cause is that (probably) the HDFS client is not available on the slave; however, we do not error() but rather return a false result.

The bug is exposed in the return line, where we try to retrieve available.error() (which is not there - it's just `false`).

This was a 'latent' bug that has been exposed by (my) recent refactoring of os::shell which is used by hdfs.available() under the covers.