You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@mesos.apache.org by "Yan Xu (JIRA)" <ji...@apache.org> on 2016/08/23 21:27:21 UTC
[jira] [Commented] (MESOS-6074) Master check failure if the metrics
endpoint is polled soon after it starts
[ https://issues.apache.org/jira/browse/MESOS-6074?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15433641#comment-15433641 ]
Yan Xu commented on MESOS-6074:
-------------------------------
The problem seems to be that: since 1.0 we introduced a few gauges including this one that depend on the sorters which are Owned pointers. These pointers are not initialized until in [Master::initialize|https://github.com/apache/mesos/blob/dcc8bd7d2a942889fe473c21ab64e863d0e6a13f/src/master/master.cpp#L699]. However the metrics, being hosted by a separate actor, can be accessed before the master is initialized, thus leaving the Owned sorter pointers uninitialized when called.
I don't see a problem initializing the sorters in the [HierarchicalAllocatorProcess constructor|https://github.com/apache/mesos/blob/dcc8bd7d2a942889fe473c21ab64e863d0e6a13f/src/master/allocator/mesos/hierarchical.hpp#L74-L85]?
/cc [~bmahler] [~jvanremoortere] [~alexr]
> Master check failure if the metrics endpoint is polled soon after it starts
> ---------------------------------------------------------------------------
>
> Key: MESOS-6074
> URL: https://issues.apache.org/jira/browse/MESOS-6074
> Project: Mesos
> Issue Type: Bug
> Components: master
> Affects Versions: 1.0.0
> Reporter: Yan Xu
> Priority: Critical
>
> We observed the following check failure
> {noformat:title=}
> F0822 22:27:10.364923 10489 owned.hpp:110] Check failed: 'get()' Must be non NULL
> {noformat}
> called from {{mesos::internal::master::allocator::internal::HierarchicalAllocatorProcess::_re
> sources_total}}
> The code:
> {code}
> double HierarchicalAllocatorProcess::_resources_total(
> const string& resource)
> {
> Option<Value::Scalar> total =
> roleSorter->totalScalarQuantities()
> .get<Value::Scalar>(resource);
> return total.isSome() ? total->value() : 0;
> }
> {code}
> See [github|https://github.com/apache/mesos/blob/dcc8bd7d2a942889fe473c21ab64e863d0e6a13f/src/master/allocator/mesos/hierarchical.cpp#L1804]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)