mesos-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Michael Park (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (MESOS-5114) empty quorum config causes masters to fail replica recovery and fail
Date Tue, 05 Apr 2016 17:16:25 GMT

    [ https://issues.apache.org/jira/browse/MESOS-5114?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15225569#comment-15225569
] 

Michael Park edited comment on MESOS-5114 at 4/5/16 5:16 PM:
-------------------------------------------------------------

[~jieyu] In the test that you posted above, it should be an error caused during {{load}} rather
than {{Option}} being set to {{None}}.
{{None}} should represent an absence of the value, as opposed to a parse failure from a specified
value.

{code}
TEST(FlagsTest, LoadFromEnvironmentEmptyInteger)
{
  TestFlags flags;

  Option<int> name6;

  flags.add(&name6,
            "name6",
            "Optional name6");

  os::setenv("FLAGSTEST_name6", "");

  Try<Nothing> load = flags.load("FLAGSTEST_");
  EXPECT_ERROR(load);

  os::unsetenv("FLAGSTEST_name6");
}
{code}

As we discussed briefly offline, {{!in.good() ||| !in.eof()}} will always return {{true}}.
Refer to the matrix of states in http://en.cppreference.com/w/cpp/io/ios_base/iostate.
{{good()}} is a pretty useless state to check for, what we really want is to use the {{operator
bool()}} along with the {{eof}} check to make sure that there are no left-over value.

{code}
if (in && in.eof()) {
  return t;
}

return Error(...);
{code}


was (Author: mcypark):
[~jieyu] In the test that you posted above, it should be an error caused during {{load}} rather
than {{Option}} being set to {{None}}.
{{None}} should represent an absence of the value, as opposed to a parse failure from a specified
value.

{code}
TEST(FlagsTest, LoadFromEnvironmentEmptyInteger)
{
  TestFlags flags;

  Option<int> name6;

  flags.add(&name6,
            "name6",
            "Optional name6");

  os::setenv("FLAGSTEST_name6", "");

  Try<Nothing> load = flags.load("FLAGSTEST_");
  EXPECT_ERROR(load);
}
{code}

As we discussed briefly offline, {{!in.good() ||| !in.eof()}} will always return {{true}}.
Refer to the matrix of states in http://en.cppreference.com/w/cpp/io/ios_base/iostate.
{{good()}} is a pretty useless state to check for, what we really want is to use the {{operator
bool()}} along with the {{eof}} check to make sure that there are no left-over value.

{code}
if (in && in.eof()) {
  return t;
}

return Error(...);
{code}

> empty quorum config causes masters to fail replica recovery and fail
> --------------------------------------------------------------------
>
>                 Key: MESOS-5114
>                 URL: https://issues.apache.org/jira/browse/MESOS-5114
>             Project: Mesos
>          Issue Type: Bug
>          Components: stout
>    Affects Versions: 0.23.1, 0.24.1, 0.25.0, 0.26.0, 0.28.0, 0.27.2
>         Environment: CentOS 7.1
>            Reporter: Cosmin Lehene
>            Assignee: Michael Park
>              Labels: mesosphere
>             Fix For: 0.26.1, 0.25.1, 0.24.2, 0.28.1, 0.27.3, 0.23.2
>
>
> A missing default for quorum size has generated the following master config 
> {code}
> MESOS_WORK_DIR="/var/lib/mesos/master"
> MESOS_ZK="zk://zk1:2181,zk2:2181,zk3:2181/mesos"
> MESOS_QUORUM=
> MESOS_PORT=5050
> MESOS_CLUSTER="mesos"
> MESOS_LOG_DIR="/var/log/mesos"
> MESOS_LOGBUFSECS=1
> MESOS_LOGGING_LEVEL="INFO"
> {code}
> This was causing each elected leader to attempt replica recovery.
> E.g. {{group.cpp:700] Trying to get '/mesos/log_replicas/0000000012' in ZooKeeper}}
> And eventually:
> {{master.cpp:1458] Recovery failed: Failed to recover registrar: Failed to perform fetch
within 1mins}}
> Full log on one of the masters https://gist.github.com/clehene/09a9ddfe49b92a5deb4c1b421f63479e
> All masters and zk nodes were reachable over the network. 
> Also once the quorum was configured the master recovery protocol finished gracefully.




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message