[
https://issues.apache.org/jira/browse/MESOS-5629?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15336423#comment-15336423
]
Greg Mann commented on MESOS-5629:
----------------------------------
I just did some testing as well - reliably reproduced the segfault before the
fix, and was unable to induce it after the fix. LGTM!
> Agent segfaults after request to '/files/browse'
> ------------------------------------------------
>
> Key: MESOS-5629
> URL: https://issues.apache.org/jira/browse/MESOS-5629
> Project: Mesos
> Issue Type: Bug
> Environment: CentOS 7, Mesos 1.0.0-rc1 with patches
> Reporter: Greg Mann
> Assignee: Joerg Schad
> Priority: Blocker
> Labels: authorization, mesosphere, security
> Fix For: 1.0.0
>
> Attachments: test-browse.py
>
>
> We observed a number of agent segfaults today on an internal testing cluster.
> Here is a log excerpt:
> {code}
> Jun 16 17:12:28 ip-10-10-0-87 mesos-slave[24818]: I0616 17:12:28.522925 24830
> status_update_manager.cpp:392] Received status update acknowledgement (UUID:
> e79ab0f4-2fa2-4df2-9b59-89b97a482167) for task
> datadog-monitor.804b138b-33e5-11e6-ac16-566ccbdde23e of framework
> 6d4248cd-2832-4152-b5d0-defbf36f6759-0000
> Jun 16 17:12:28 ip-10-10-0-87 mesos-slave[24818]: I0616 17:12:28.523006 24830
> status_update_manager.cpp:824] Checkpointing ACK for status update
> TASK_RUNNING (UUID: e79ab0f4-2fa2-4df2-9b59-89b97a482167) for task
> datadog-monitor.804b138b-33e5-11e6-ac16-566ccbdde23e of framework
> 6d4248cd-2832-4152-b5d0-defbf36f6759-0000
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: I0616 17:12:29.147181 24824
> http.cpp:192] HTTP GET for /slave(1)/state from 10.10.0.87:33356
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: *** Aborted at 1466097149
> (unix time) try "date -d @1466097149" if you are using GNU date ***
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: PC: @ 0x7ff4d68b12a3
> (unknown)
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: *** SIGSEGV (@0x0) received
> by PID 24818 (TID 0x7ff4d31ab700) from PID 0; stack trace: ***
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d6431100
> (unknown)
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d68b12a3
> (unknown)
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d7eced33
> process::dispatch<>()
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d7e7aad7
> _ZNSt17_Function_handlerIFN7process6FutureIbEERK6OptionISsEEZN5mesos8internal5slave9Framework15recoverExecutorERKNSA_5state13ExecutorStateEEUlS6_E_E9_M_invokeERKSt9_Any_dataS6_
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d7bd1752
> mesos::internal::FilesProcess::authorize()
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d7bd1bea
> mesos::internal::FilesProcess::browse()
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d7bd6e43
> std::_Function_handler<>::_M_invoke()
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d85478cb
> _ZZZN7process11ProcessBase5visitERKNS_9HttpEventEENKUlRKNS_6FutureI6OptionINS_4http14authentication20AuthenticationResultEEEEE0_clESC_ENKUlRKNS4_IbEEE1_clESG_
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d8551341
> process::ProcessManager::resume()
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d8551647
> _ZNSt6thread5_ImplISt12_Bind_simpleIFZN7process14ProcessManager12init_threadsEvEUt_vEEE6_M_runEv
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d6909220
> (unknown)
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d6429dc5
> start_thread
> Jun 16 17:12:29 ip-10-10-0-87 mesos-slave[24818]: @ 0x7ff4d615728d __clone
> Jun 16 17:12:29 ip-10-10-0-87 systemd[1]: dcos-mesos-slave.service: main
> process exited, code=killed, status=11/SEGV
> Jun 16 17:12:29 ip-10-10-0-87 systemd[1]: Unit dcos-mesos-slave.service
> entered failed state.
> Jun 16 17:12:29 ip-10-10-0-87 systemd[1]: dcos-mesos-slave.service failed.
> Jun 16 17:12:34 ip-10-10-0-87 systemd[1]: dcos-mesos-slave.service holdoff
> time over, scheduling restart.
> {code}
> In every case, the stack trace indicates one of the {{/files/*}} endpoints; I
> observed this a number of times coming from {{browse()}}, and twice from
> {{read()}}.
> The agent was built from the 1.0.0-rc1 branch, with two cherry-picks applied:
> [this|https://reviews.apache.org/r/48563/] and
> [this|https://reviews.apache.org/r/48566/], which were done to repair a
> different [segfault issue|https://issues.apache.org/jira/browse/MESOS-5587]
> on the master and agent.
> Thanks go to [~bmahler] for digging into this a bit and discovering a
> possible cause
> [here|https://github.com/apache/mesos/blob/master/src/slave/slave.cpp#L5737-L5745],
> where use of {{defer()}} may be necessary to keep execution in the correct
> context.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
