[
https://issues.apache.org/jira/browse/MESOS-675?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13757264#comment-13757264
]
Benjamin Mahler commented on MESOS-675:
---------------------------------------
There is a definite race here in that the master repeatedly calls {{link()}} on
the same slave whenever it gets the duplicate re-registration messages. The
CHECK failure above can occur if:
-> Slave run #1 terminates
-> delayed re-registration messages from Slave #1 are enqueued on the Master
-> delayed exited event from Slave run #1 is enqueued on the Master (now
possible to re-link the slave)
-> Slave starts again (run #2)
-> Master::reregisterSlave() calls link() (this unintentionally links against
Slave run #2 since the UPID is the same)
-> Slave run #2 crashes because it received a re-registered message while
recovering
-> An additional exited event is now enqueued on the Master
-> Master goes through its queue, eventually processes both exited events
We can do either or both of the following:
1. Only call {{link()}} the slave when re-registering if it was disconnected,
rather than always calling {{link()}}.
2. Remove the CHECK(!slave->disconnected) and instead ignore duplicate exited
events.
> CHECK failure in the Master.
> ----------------------------
>
> Key: MESOS-675
> URL: https://issues.apache.org/jira/browse/MESOS-675
> Project: Mesos
> Issue Type: Bug
> Reporter: Benjamin Mahler
> Assignee: Benjamin Mahler
> Priority: Blocker
> Fix For: 0.14.0
>
>
> Observed this failure in a staging cluster running 0.14.0-rc2.
> {noformat}
> F0902 06:01:11.105391 11876 master.cpp:564] Check failed:
> !slave->disconnected Slave 201308270033-1937777162-5050-50911-137 (<scrub>)
> already disconnected!
> *** Check failure stack trace: ***
> @ 0x7fb470894d8d google::LogMessage::Fail()
> @ 0x7fb470898d77 google::LogMessage::SendToLog()
> @ 0x7fb470897674 google::LogMessage::Flush()
> @ 0x7fb4708978a6 google::LogMessageFatal::~LogMessageFatal()
> @ 0x7fb4704aaea4 mesos::internal::master::Master::exited()
> @ 0x7fb470786af4 process::ProcessManager::resume()
> @ 0x7fb47078754f process::schedule()
> @ 0x7fb46fef483d start_thread
> @ 0x7fb46e8d6f8d clone
> {noformat}
> Grepping for this slave in the logs:
> {noformat}
> $ grep 201308270033-1937777162-5050-50911-137 /var/log/mesos/mesos-master.log
> W0902 06:01:10.607168 11876 master.cpp:1317] Ignoring unknown exited executor
> thermos-1377831261464-mesos-slave-recovery-spinner-60-f0bcfda6-4f8d-4df4-bd74-0b15f32d0502
> on slave 201308270033-1937777162-5050-50911-137 (<scrub>)
> ...
> W0902 06:01:10.646383 11876 master.cpp:1317] Ignoring unknown exited executor
> thermos-1377964938274-mesos-slave-recovery-spinner-184-3a25b824-5d73-4be0-984d-606230c5e8ac
> on slave 201308270033-1937777162-5050-50911-137 (<scrub>)
> W0902 06:01:10.699635 11876 master.cpp:1123] Slave at
> slave(1)@10.34.110.125:5051 (<scrub>) is being allowed to re-register with an
> already in use id (201308270033-1937777162-5050-50911-137)
> I0902 06:01:10.700628 11868 hierarchical_allocator_process.hpp:434] Added
> slave 201308270033-1937777162-5050-50911-137 (<scrub>) with cpus(*):14;
> mem(*):21913; ports(*):[31000-32000]; disk(*):400000 (and cpus(*):10.96;
> mem(*):19866; ports(*):[31000-31003, 31005-31449, 31451-31580, 31582-31801,
> 31803-31927, 31929-32000]; disk(*):397809 available)
> W0902 06:01:10.866525 11876 master.cpp:1123] Slave at
> slave(1)@10.34.110.125:5051 (<scrub>) is being allowed to re-register with an
> already in use id (201308270033-1937777162-5050-50911-137)
> W0902 06:01:10.919178 11876 master.cpp:1123] Slave at
> slave(1)@10.34.110.125:5051 (<scrub>) is being allowed to re-register with an
> already in use id (201308270033-1937777162-5050-50911-137)
> W0902 06:01:11.070862 11876 master.cpp:1123] Slave at
> slave(1)@10.34.110.125:5051 (<scrub>) is being allowed to re-register with an
> already in use id (201308270033-1937777162-5050-50911-137)
> I0902 06:01:11.085773 11876 master.cpp:553] Slave
> 201308270033-1937777162-5050-50911-137 (<scrub>) disconnected
> W0902 06:01:11.086096 11876 master.cpp:1404] Master returning resources
> offered because slave 201308270033-1937777162-5050-50911-137 is disconnected
> I0902 06:01:11.086145 11867 hierarchical_allocator_process.hpp:459] Removed
> slave 201308270033-1937777162-5050-50911-137
> I0902 06:01:11.104651 11876 master.cpp:553] Slave
> 201308270033-1937777162-5050-50911-137 (<scrub>) disconnected
> F0902 06:01:11.105391 11876 master.cpp:564] Check failed:
> !slave->disconnected Slave 201308270033-1937777162-5050-50911-137 (<scrub>)
> already disconnected!
> {noformat}
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
