[
https://issues.apache.org/jira/browse/MESOS-8350?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16311275#comment-16311275
]
Benjamin Bannier commented on MESOS-8350:
-----------------------------------------
{noformat}
commit c29b7cd7b5627964ca75001dd3195656816f870c
Author: Benjamin Bannier <[email protected]>
Date: Thu Jan 4 11:12:27 2018 +0100
Fixed handling of checkpointed resources for RP-capable agents.
The master will not resend checkpointed resources when a resource
provider-capable agent reregisters. Instead the checkpointed resources
sent as part of the agent reregistration should be evaluated by the
master and be used to update its state.
This patch fixes the handling of checkpointed resources sent as part
of the agent reregistration so that the resources are used to update
the master state.
Review: https://reviews.apache.org/r/64889/
commit 6f134c93e52120ad6f29ef9057e2045ad8f24c7c
Author: Benjamin Bannier <[email protected]>
Date: Thu Jan 4 11:12:17 2018 +0100
Added test of handling of checkpointed resources in reregistration.
This patch adds a test that confirms that the master resends
checkpointed resources to the agent on reregistration.
Review: https://reviews.apache.org/r/64888/
commit 4e5b8cfe86061451747899ef516aa0c4bea12bca
Author: Benjamin Bannier <[email protected]>
Date: Thu Jan 4 11:12:01 2018 +0100
Future-proofed use of agent capabilities in tests.
Even though currently the resource provider capability is the only
capability which can be toggled by users, when examining agent flags
we expect either no capabilities at all or a full set of capabilities
(including both toggleable and required capabilities).
This patch cleans up test code agent capabilities by delegating the
bulk of the work to a helper function providing a set of
default-enabled capabilities and only adds a single capability to that
set. This not only makes it clearer which exact capability a test
cares about, but also future-proofs the code for the case where we
extend the set of required capabilities in the future.
Review: https://reviews.apache.org/r/64891/
{noformat}
> Resource provider-capable agents not correctly synchronizing checkpointed
> agent resources on reregistration
> -----------------------------------------------------------------------------------------------------------
>
> Key: MESOS-8350
> URL: https://issues.apache.org/jira/browse/MESOS-8350
> Project: Mesos
> Issue Type: Bug
> Components: master
> Reporter: Benjamin Bannier
> Assignee: Benjamin Bannier
> Priority: Critical
> Fix For: 1.5.0, 1.6.0
>
>
> For resource provider-capable agents the master does not re-send checkpointed
> resources on agent reregistration; instead the checkpointed resources sent as
> part of the {{ReregisterSlaveMessage}} should be used.
> This is not what happens in reality. If e.g., checkpointing of an offer
> operation fails and the agent fails over the checkpointed resources would, as
> expected, not be reflected in the agent, but would still be assumed in the
> master.
> A workaround is to fail over the master which would lead to the newly elected
> master bootstrapping agent state from {{ReregisterSlaveMessage}}.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
