> On Nov. 15, 2016, 6:42 a.m., Sivaram Kannan wrote: > > src/slave/containerizer/mesos/containerizer.cpp, line 1388 > > <https://reviews.apache.org/r/53473/diff/6/?file=1563497#file1563497line1388> > > > > Passing None() has the agent crashing when we have switch_user=false. > > Sivaram Kannan wrote: > That is the reason I thought I would have the current user of the agent > set to logger. > > Joseph Wu wrote: > The reason the agent/test would fail is because of this line in your > fourth diff: > https://reviews.apache.org/r/53473/diff/4#7 > > `container->config.user()` has a return type of `string`. When the value > is not set, calling `container->config.user()` returns the empty string. You > need to manually convert it to an `Option<string>`. > > `container->config` is a protobuf defined here: > > https://github.com/apache/mesos/blob/1.1.x/include/mesos/slave/containerizer.proto#L112 > > Sivaram Kannan wrote: > https://reviews.apache.org/r/53473/diff/6/ has an explicit conversion to > Option<string>. When container->config.user() is empty, I tried to set > > Option<string> user = None() > > But the call logger->prepare(..., user), still crashes the agent. Doesn't > the below code more logical than setting the value to None()? > > Result<std::string> result = os::user(); > user = result.get(); > > Joseph Wu wrote: > What's the stack trace for the crash? (And what are the steps taken to > repro?) > > Sivaram Kannan wrote: > Sorry - should have added when I created the RBR. > > Steps to reproduce: > 1. In the containerizer.cpp, change the below code in the rbr > 2. Option<string> user = None(); > if (container->config.has_user()) { > user = container->config.user(); > } else { > Result<std::string> result = os::user(); > //user = result.get(); <============= Current code > user = None(); <=========== Change it to this > } > 3. make and make check. > 4. Run the below test > 5. sudo bin/mesos-tests.sh --verbose > --gtest_filter="UserContainerLoggerTest.ROOT_LOGROTATE_RotateAsSwitchUserFalse" > 6. The above test would end up in a crash like below. > 7. Stack Trace: > 8. I1116 15:58:16.660096 15282 slave.cpp:2031] Queued task > 'bffed9be-9830-4840-a313-e2ff3741b2fc' for executor > 'bffed9be-9830-4840-a313-e2ff3741b2fc' of framework > 16b6e12e-3344-4aad-adf1-a3460b3ad2a1-0000 > lt-mesos-tests: ../../3rdparty/stout/include/stout/option.hpp:111: const > T& Option<T>::get() const & [with T = std::basic_string<char>]: Assertion > `isSome()' failed. > *** Aborted at 1479311896 (unix time) try "date -d @1479311896" if you > are using GNU date *** > PC: @ 0x7f89e65b0c37 (unknown) > *** SIGABRT (@0x3ba2) received by PID 15266 (TID 0x7f89dd0db700) from PID > 15266; stack trace: *** > @ 0x7f89e694f330 (unknown) > @ 0x7f89e65b0c37 (unknown) > @ 0x7f89e65b4028 (unknown) > @ 0x7f89e65a9bf6 (unknown) > @ 0x7f89e65a9ca2 (unknown) > @ 0xa8542f _ZNKR6OptionISsE3getEv > @ 0x7f89dfba6a9b > mesos::internal::logger::LogrotateContainerLoggerProcess::prepare() > @ 0x7f89dfbbcb5e > _ZZN7process8dispatchIN5mesos5slave15ContainerLogger14SubprocessInfoENS1_8internal6logger31LogrotateContainerLoggerProcessERKNS1_12ExecutorInfoERKSsRK6OptionISsES8_SsSE_EENS_6FutureIT_EERKNS_3PIDIT0_EEMSL_FSJ_T1_T2_T3_ET4_T5_T6_ENKUlPNS_11ProcessBaseEE_clESY_ > @ 0x7f89dfbcb54d > _ZNSt17_Function_handlerIFvPN7process11ProcessBaseEEZNS0_8dispatchIN5mesos5slave15ContainerLogger14SubprocessInfoENS5_8internal6logger31LogrotateContainerLoggerProcessERKNS5_12ExecutorInfoERKSsRK6OptionISsESC_SsSI_EENS0_6FutureIT_EERKNS0_3PIDIT0_EEMSP_FSN_T1_T2_T3_ET4_T5_T6_EUlS2_E_E9_M_invokeERKSt9_Any_dataS2_ > @ 0x7f89eca9168d std::function<>::operator()() > @ 0x7f89eca73e0d process::ProcessBase::visit() > @ 0x7f89eca7c2ec process::DispatchEvent::visit() > @ 0xa674b2 process::ProcessBase::serve() > @ 0x7f89eca700fa process::ProcessManager::resume() > @ 0x7f89eca6cc39 > _ZZN7process14ProcessManager12init_threadsEvENKUt_clEv > @ 0x7f89eca7ba94 > _ZNSt12_Bind_simpleIFZN7process14ProcessManager12init_threadsEvEUt_vEE9_M_invokeIIEEEvSt12_Index_tupleIIXspT_EEE > @ 0x7f89eca7b9eb > _ZNSt12_Bind_simpleIFZN7process14ProcessManager12init_threadsEvEUt_vEEclEv > @ 0x7f89eca7b984 > _ZNSt6thread5_ImplISt12_Bind_simpleIFZN7process14ProcessManager12init_threadsEvEUt_vEEE6_M_runEv > @ 0x7f89e712aa60 (unknown) > @ 0x7f89e6947184 start_thread > @ 0x7f89e667437d (unknown)
With your latest review diff, this stack trace shouldn't be possible, as there is no code in the `LogrotateContainerLoggerProcess` that calls `user.get()`. Can you double check? If you've added a debug print statement or something else excluded the from the diff, that may be causing it. - Joseph ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/53473/#review155916 ----------------------------------------------------------- On Nov. 14, 2016, 1:52 p.m., Sivaram Kannan wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/53473/ > ----------------------------------------------------------- > > (Updated Nov. 14, 2016, 1:52 p.m.) > > > Review request for mesos and Joseph Wu. > > > Bugs: MESOS-5856 > https://issues.apache.org/jira/browse/MESOS-5856 > > > Repository: mesos > > > Description > ------- > > Add new param user to logrotate's prepare function. > > > Diffs > ----- > > include/mesos/slave/container_logger.hpp > 939974736f9eb744c83036e074718d2a1eba8b0a > src/slave/container_loggers/lib_logrotate.hpp > 28fdf3bdcc66d473521b377f66ab0b48f6900f58 > src/slave/container_loggers/lib_logrotate.cpp > 53698d339f0f4d2dc916b53239ca0c36bbebcd42 > src/slave/container_loggers/logrotate.hpp > d1db69236f5a9b1dbb3113ad02218a512afdb46b > src/slave/container_loggers/sandbox.hpp > e0aeb32a9ec83af049af8a10010b819c1d8b25d8 > src/slave/container_loggers/sandbox.cpp > cc263ebef7e0c3e778fabafa49faa6dd315adc45 > src/slave/containerizer/docker.cpp c2ed5240aab9ea9d1a386c44c94e5ae7e98d313c > src/slave/containerizer/mesos/containerizer.cpp > ec4ae32485a7ab6c9f73c512004d1220482a188e > src/tests/container_logger_tests.cpp > 1bb94a8461e481983f25a44737e4011ed5fc4b1f > > Diff: https://reviews.apache.org/r/53473/diff/ > > > Testing > ------- > > Run the mesos-logrotate-logger with un-priviledged user and verify whether > the logs are getting rotated. > Run the mesos-logrotate-logger as root user and verify whether the logs are > getting rotated. > > > Thanks, > > Sivaram Kannan > >
