Please see inline... On Tue, Jul 12, 2016 at 2:01 AM, Attila Simon <[email protected]> wrote:
> "TestFileChannel.testInOut:117 ? Failed to locate tar-ball > distribution. Pleas..." > These seem transient and build related. OOM killer, disk swipe, some > restart, etc. Would you mind retriggring the job? I don't have > perimissions for that. > I retriggered the job and it passed! Good or bad, you decide. :) @Mike StagedInstall was last edited by you, do you happen to remember > whether is there any documentation to catch the idea how this supposed > to work in details, what are the weak points if there is any? It was Arvind's baby. :) And he basically did it as a quick hack to get an integration test framework working. I don't remember doing major surgery to it. LMK if you have specific questions. One drawback, IIRC, is that we didn't put in the necessary plumbing to make it start up quickly, so there are a bunch of sleeps and the tests run slowly. @Mike: decomposing what you wrote: ~5.6G is the max memory and flume > would like to allocate its first ~1M block of direct memory. I guess > jenkins was configured with a non sun/oracle jvm but that shouldn't be > an issue. If OOM killer did this then it might be related but unlikely > that 1M triggered it (unless the allowed max direct memory was set to > something smaller than 1M). Otherwise it seems normal in aspect of > file channel. > I agree that 1MB should not trigger the OOM killer. This failure is still a mystery to me. Mike
