[
https://issues.apache.org/jira/browse/MESOS-3090?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14634173#comment-14634173
]
Andrew Jorgensen commented on MESOS-3090:
-----------------------------------------
The 3.9999999999999973 came from the aurora offers endpoint.
We are using version 0.22.0 of mesos.
Here are the logs, it looks like the slave thinks it has 4 cpu
{quote}
I0720 22:26:43.982659 25402 main.cpp:200] Starting Mesos slave
2015-07-20 22:26:43,982:25402(0x7f0e2fddb700):ZOO_INFO@log_env@712: Client
environment:zookeeper.version=zookeeper C client 3.4.5
2015-07-20 22:26:43,982:25402(0x7f0e2fddb700):ZOO_INFO@log_env@716: Client
environment:host.name=mesos-worker2019
2015-07-20 22:26:43,982:25402(0x7f0e2fddb700):ZOO_INFO@log_env@723: Client
environment:os.name=Linux
2015-07-20 22:26:43,982:25402(0x7f0e2fddb700):ZOO_INFO@log_env@724: Client
environment:os.arch=3.13.0-40-generic
2015-07-20 22:26:43,982:25402(0x7f0e2fddb700):ZOO_INFO@log_env@725: Client
environment:os.version=#69-Ubuntu SMP Thu Nov 13 17:53:56 UTC 2014
I0720 22:26:43.983403 25402 slave.cpp:174] Slave started on 1)@10.95.200.84:5051
I0720 22:26:43.983973 25402 slave.cpp:299] Slave resources: cpus(*):4;
mem(*):14018; disk(*):418048; ports(*):[31000-32000]
2015-07-20 22:26:43,984:25402(0x7f0e2fddb700):ZOO_INFO@log_env@733: Client
environment:user.name=(null)
2015-07-20 22:26:43,984:25402(0x7f0e2fddb700):ZOO_INFO@log_env@741: Client
environment:user.home=/root
2015-07-20 22:26:43,984:25402(0x7f0e2fddb700):ZOO_INFO@log_env@753: Client
environment:user.dir=/
{quote}
> Calculated CPU should be rounded
> --------------------------------
>
> Key: MESOS-3090
> URL: https://issues.apache.org/jira/browse/MESOS-3090
> Project: Mesos
> Issue Type: Bug
> Reporter: Andrew Jorgensen
>
> My mesos worker pool consists of 4 core machines in AWS. When I look at the
> calculated amount of CPU that each machine is offering it looks like
> {quote}
> "name":"cpus",
> "scalar":3.9999999999999973
> {quote}
> It feels like this number should be rounded up to the nearest thousands so
> that a job does not need to specify 3.9 CPU but can just specify the full 4.
> This can be remedied by adding the following to the worker machines:
> {quote}
> --resources='cpus:4'
> {quote}
> But this would be hard to manage if the worker pool is at all heterogenous.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
