I think you noticed the main problem with turning logging off entirely.

I'd be inclined to have two files: one which defaults to INFO and another
that defaults to ERROR for Travis. We can give a
-Dlog4j.configuration=file:log4j.config.set.to.ERROR.only for travis, which
I think Otto suggested.

That said, we're still going to have to suppress the

[ERROR|org.apache.curator.framework.imps.CuratorFrameworkImpl|CuratorFrameworkImpl]
Background retry gave up

type errors if they're not actual errors. Really, any of the exhaust
logging from the integration tests should be captured and discarded.

-D...


On Fri, Oct 28, 2016 at 2:41 PM, Michael Miklavcic <
[email protected]> wrote:

> To Otto and David's points: We could default 
> src/test/resources/log4j.properties
> to log4j.rootLogger=OFF
>
> Or alternatively, we could programmatically disable logging in a @Before
> method in the tests. e.g.
>
> package org.apache.metron;
>
> import org.apache.log4j.Level;
> import org.apache.log4j.Logger;
> import org.junit.BeforeClass;
> import org.junit.Test;
>
> public class MikeTest {
>
>   @BeforeClass
>   public static void setup() {
>     Logger.getRootLogger().setLevel(Level.OFF);
>   }
>
>   @Test
>   public void noMessage() {
>     new SomeClass().doSomething();
>   }
>
>   private class SomeClass {
>     private final Logger LOG = Logger.getLogger(SomeClass.class);
>
>     public void doSomething() {
>       LOG.error("don't want");
>     }
>   }
>
> }
>
> But that would disable logging for every test, regardless of intent. And
> we might miss valuable messages for tests that unexpectedly fail. I guess
> we could do this on a per-test basis, but that feels like a lot of work.
> I'm curious what our biggest offender is currently - is it the zookeeper
> and other services from the integration tests, as Casey pointed out?
>
>
> On Fri, Oct 28, 2016 at 1:46 PM, Otto Fowler <[email protected]>
> wrote:
>
>> What if we had a set of log4j.properties just for travis?  I think we can
>> script travis and deploy them then.
>> We could have full logging for local builds, and travis specific high
>> level
>> logging at the same time.
>>
>>
>> On October 27, 2016 at 20:52:46, David Lyle ([email protected]) wrote:
>>
>> I'd prefer not to see a file output by default, it's non-standard and
>> people won't know to look there without additional documentation. If we
>> default to ERROR logging and suppress ERROR output in the case of
>> "expected" errors, we should be able to keep the test output under 4MB.
>> Then, if additional logging is desired, one could enable a different log
>> level and carry on.
>>
>> For the out of order components, we can capture the output and suppress
>> it.
>>
>> -D...
>>
>>
>> On Thu, Oct 27, 2016 at 8:33 PM, Casey Stella <[email protected]> wrote:
>>
>> > I can get behind logging to a file, but can we see the file as an output
>> of
>> > Travis? Sometimes the problem is that it is failing in Travis but not
>> > locally or it is a situation that you want to give guidance on a PR but
>> > don't want to pull a branch and rerun tests. If we log to a file, I'd
>> like
>> > to make sure we can get that file as part of the Travis build.
>> >
>> > On Fri, Oct 28, 2016 at 00:42 Otto Fowler <[email protected]>
>> wrote:
>> >
>> > > I think we should have all the tests log to file verbosely and to
>> console
>> > > reduced. We can gather verbose logs on fail and have them part of the
>> > > clean etc. Error messages from integration components that do not
>> result
>> > > in test failures will still be a problem otherwise. I do not think
>> > > personally that the Travis visible logs need to show more than what
>> > failed.
>> > >
>> > > On October 27, 2016 at 17:05:28, Ryan Merriman ([email protected])
>> > > wrote:
>> > >
>> > > > I definitely agree with Otto that we need to make our build logs
>> less
>> > > > verbose. I think the first step is to get control of the log levels
>> > > across
>> > > > various components. Are there specific examples of the types of
>> > messages
>> > > > we want to suppress? For instance:
>> > > >
>> > > > org.apache.metron.profiler.integration.ProfilerIntegrationTest -
>> > > zookeeper
>> > > > and curator log levels are set to INFO
>> > > > org.apache.metron.enrichment.bolt.ThreatIntelJoinBoltTest -
>> expected
>> > > > exceptions are printed out when they should be suppressed (to Dave's
>> > > > point)
>> > > > ...
>> > > >
>> > > > Maybe building a list of examples will be easier after we take a
>> pass
>> > at
>> > > > setting all the log levels to ERROR. Ideally we only see the test
>> > results
>> > > > and an error if something really did fail. Does that sound like a
>> > > > reasonable approach?
>> > > >
>> > > > Ryan
>> > > >
>> > > > On Wed, Oct 19, 2016 at 7:44 AM, David Lyle <[email protected]>
>> > > wrote:
>> > > >
>> > > > Hi Otto,
>> > > >
>> > > > Good point. One of the practices I'd like to see is to suppress any
>> > > > exception/error output in tests when it's expected.
>> > > >
>> > > > Thanks!
>> > > >
>> > > > -D...
>> > > >
>> > > >
>> > > > On Wed, Oct 19, 2016 at 8:18 AM, Otto Fowler <
>> [email protected]>
>>
>> > > > wrote:
>> > > >
>> > > > The Metron build logs contain a great deal of information, including
>> a
>> > > > large number of warnings / exceptions that do not result in test
>> > failures
>> > > > but still fill the log. This makes troubleshooting test/build issues
>> > > > difficult. For example - the travis ci log for PR #276’s recent
>> failure
>> > > > doesn’t even have the final errors because the log is too long.
>> > > >
>> > > > What are some things that can be done to make the situation better?
>> > > >
>> > > >
>> > > >
>> > > >
>> > > >
>> > >
>> >
>>
>
>

Reply via email to