You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mesos.apache.org by Krishnanarayanan VR <kr...@phonepe.com> on 2016/11/26 09:05:54 UTC

Force offer from all of the slaves

Hello:

Is there a way to force ResourceOffers to get offers from all available
slaves ?

Let me clarify:

I have a single framework in my cluster. Each time ResourceOffers gets the
list of offers, I choose the right offer and decline the rest. But I notice
that next time a callback to ResourceOffers occurs, only a subset of slaves
is present in the offer. The slave from offer that was chosen in the
previous iteration is invariably absent.

I also tried to set refuse_seconds to 0 in  both LaunchTasks and
Decline(egs below):

driver.DeclineOffer(offer.Id, &mesos.Filters{RefuseSeconds:
proto.Float64(0)})

^^ but that didn't seem to help.

Any pointers how I can make sure am presented with offers from all the
slaves all the time ?

Thanks

Re: Force offer from all of the slaves

Posted by Krishnanarayanan VR <kr...@phonepe.com>.
Thanks for the responses. Am unable to try "GLOG_v=1" in my production
setup at this time. However, I tweaked the framework logic a bit to reject
offers that don't match my requirement and "wait" till the right offer
comes by.



On Tue, Nov 29, 2016 at 12:02 AM, Vinod Kone <vi...@apache.org> wrote:

> Once you set GLOG_v, you should be able to see lines like these "Framework
> <adfaf> filtered agent <adfaf>  for <123> seconds"
>
> On Sun, Nov 27, 2016 at 8:18 AM, haosdent <ha...@gmail.com> wrote:
>
>> > I choose the right offer and decline the rest.
>> Hi, @krishnanvr Do you use up all available resources in that agent's
>> offer? If so, that agent could not provide offers anymore until the
>> resource release.
>>
>> And you may consider starting the master with the `GLOG_v=1` environment
>> variable which would print more detail logs to help you debug this.
>>
>> On Sat, Nov 26, 2016 at 5:05 PM, Krishnanarayanan VR <
>> krishnanvr@phonepe.com> wrote:
>>
>>> Hello:
>>>
>>> Is there a way to force ResourceOffers to get offers from all available
>>> slaves ?
>>>
>>> Let me clarify:
>>>
>>> I have a single framework in my cluster. Each time ResourceOffers gets
>>> the list of offers, I choose the right offer and decline the rest. But I
>>> notice that next time a callback to ResourceOffers occurs, only a subset of
>>> slaves is present in the offer. The slave from offer that was chosen in the
>>> previous iteration is invariably absent.
>>>
>>> I also tried to set refuse_seconds to 0 in  both LaunchTasks and
>>> Decline(egs below):
>>>
>>> driver.DeclineOffer(offer.Id, &mesos.Filters{RefuseSeconds:
>>> proto.Float64(0)})
>>>
>>> ^^ but that didn't seem to help.
>>>
>>> Any pointers how I can make sure am presented with offers from all the
>>> slaves all the time ?
>>>
>>> Thanks
>>>
>>>
>>>
>>>
>>
>>
>> --
>> Best Regards,
>> Haosdent Huang
>>
>
>

Re: Force offer from all of the slaves

Posted by Krishnanarayanan VR <kr...@phonepe.com>.
Thanks for the responses. Am unable to try "GLOG_v=1" in my production
setup at this time. However, I tweaked the framework logic a bit to reject
offers that don't match my requirement and "wait" till the right offer
comes by.



On Tue, Nov 29, 2016 at 12:02 AM, Vinod Kone <vi...@apache.org> wrote:

> Once you set GLOG_v, you should be able to see lines like these "Framework
> <adfaf> filtered agent <adfaf>  for <123> seconds"
>
> On Sun, Nov 27, 2016 at 8:18 AM, haosdent <ha...@gmail.com> wrote:
>
>> > I choose the right offer and decline the rest.
>> Hi, @krishnanvr Do you use up all available resources in that agent's
>> offer? If so, that agent could not provide offers anymore until the
>> resource release.
>>
>> And you may consider starting the master with the `GLOG_v=1` environment
>> variable which would print more detail logs to help you debug this.
>>
>> On Sat, Nov 26, 2016 at 5:05 PM, Krishnanarayanan VR <
>> krishnanvr@phonepe.com> wrote:
>>
>>> Hello:
>>>
>>> Is there a way to force ResourceOffers to get offers from all available
>>> slaves ?
>>>
>>> Let me clarify:
>>>
>>> I have a single framework in my cluster. Each time ResourceOffers gets
>>> the list of offers, I choose the right offer and decline the rest. But I
>>> notice that next time a callback to ResourceOffers occurs, only a subset of
>>> slaves is present in the offer. The slave from offer that was chosen in the
>>> previous iteration is invariably absent.
>>>
>>> I also tried to set refuse_seconds to 0 in  both LaunchTasks and
>>> Decline(egs below):
>>>
>>> driver.DeclineOffer(offer.Id, &mesos.Filters{RefuseSeconds:
>>> proto.Float64(0)})
>>>
>>> ^^ but that didn't seem to help.
>>>
>>> Any pointers how I can make sure am presented with offers from all the
>>> slaves all the time ?
>>>
>>> Thanks
>>>
>>>
>>>
>>>
>>
>>
>> --
>> Best Regards,
>> Haosdent Huang
>>
>
>

Re: Force offer from all of the slaves

Posted by Vinod Kone <vi...@apache.org>.
Once you set GLOG_v, you should be able to see lines like these "Framework
<adfaf> filtered agent <adfaf>  for <123> seconds"

On Sun, Nov 27, 2016 at 8:18 AM, haosdent <ha...@gmail.com> wrote:

> > I choose the right offer and decline the rest.
> Hi, @krishnanvr Do you use up all available resources in that agent's
> offer? If so, that agent could not provide offers anymore until the
> resource release.
>
> And you may consider starting the master with the `GLOG_v=1` environment
> variable which would print more detail logs to help you debug this.
>
> On Sat, Nov 26, 2016 at 5:05 PM, Krishnanarayanan VR <
> krishnanvr@phonepe.com> wrote:
>
>> Hello:
>>
>> Is there a way to force ResourceOffers to get offers from all available
>> slaves ?
>>
>> Let me clarify:
>>
>> I have a single framework in my cluster. Each time ResourceOffers gets
>> the list of offers, I choose the right offer and decline the rest. But I
>> notice that next time a callback to ResourceOffers occurs, only a subset of
>> slaves is present in the offer. The slave from offer that was chosen in the
>> previous iteration is invariably absent.
>>
>> I also tried to set refuse_seconds to 0 in  both LaunchTasks and
>> Decline(egs below):
>>
>> driver.DeclineOffer(offer.Id, &mesos.Filters{RefuseSeconds:
>> proto.Float64(0)})
>>
>> ^^ but that didn't seem to help.
>>
>> Any pointers how I can make sure am presented with offers from all the
>> slaves all the time ?
>>
>> Thanks
>>
>>
>>
>>
>
>
> --
> Best Regards,
> Haosdent Huang
>

Re: Force offer from all of the slaves

Posted by Vinod Kone <vi...@apache.org>.
Once you set GLOG_v, you should be able to see lines like these "Framework
<adfaf> filtered agent <adfaf>  for <123> seconds"

On Sun, Nov 27, 2016 at 8:18 AM, haosdent <ha...@gmail.com> wrote:

> > I choose the right offer and decline the rest.
> Hi, @krishnanvr Do you use up all available resources in that agent's
> offer? If so, that agent could not provide offers anymore until the
> resource release.
>
> And you may consider starting the master with the `GLOG_v=1` environment
> variable which would print more detail logs to help you debug this.
>
> On Sat, Nov 26, 2016 at 5:05 PM, Krishnanarayanan VR <
> krishnanvr@phonepe.com> wrote:
>
>> Hello:
>>
>> Is there a way to force ResourceOffers to get offers from all available
>> slaves ?
>>
>> Let me clarify:
>>
>> I have a single framework in my cluster. Each time ResourceOffers gets
>> the list of offers, I choose the right offer and decline the rest. But I
>> notice that next time a callback to ResourceOffers occurs, only a subset of
>> slaves is present in the offer. The slave from offer that was chosen in the
>> previous iteration is invariably absent.
>>
>> I also tried to set refuse_seconds to 0 in  both LaunchTasks and
>> Decline(egs below):
>>
>> driver.DeclineOffer(offer.Id, &mesos.Filters{RefuseSeconds:
>> proto.Float64(0)})
>>
>> ^^ but that didn't seem to help.
>>
>> Any pointers how I can make sure am presented with offers from all the
>> slaves all the time ?
>>
>> Thanks
>>
>>
>>
>>
>
>
> --
> Best Regards,
> Haosdent Huang
>

Re: Force offer from all of the slaves

Posted by haosdent <ha...@gmail.com>.
> I choose the right offer and decline the rest.
Hi, @krishnanvr Do you use up all available resources in that agent's
offer? If so, that agent could not provide offers anymore until the
resource release.

And you may consider starting the master with the `GLOG_v=1` environment
variable which would print more detail logs to help you debug this.

On Sat, Nov 26, 2016 at 5:05 PM, Krishnanarayanan VR <krishnanvr@phonepe.com
> wrote:

> Hello:
>
> Is there a way to force ResourceOffers to get offers from all available
> slaves ?
>
> Let me clarify:
>
> I have a single framework in my cluster. Each time ResourceOffers gets the
> list of offers, I choose the right offer and decline the rest. But I notice
> that next time a callback to ResourceOffers occurs, only a subset of slaves
> is present in the offer. The slave from offer that was chosen in the
> previous iteration is invariably absent.
>
> I also tried to set refuse_seconds to 0 in  both LaunchTasks and
> Decline(egs below):
>
> driver.DeclineOffer(offer.Id, &mesos.Filters{RefuseSeconds:
> proto.Float64(0)})
>
> ^^ but that didn't seem to help.
>
> Any pointers how I can make sure am presented with offers from all the
> slaves all the time ?
>
> Thanks
>
>
>
>


-- 
Best Regards,
Haosdent Huang

Re: Force offer from all of the slaves

Posted by haosdent <ha...@gmail.com>.
> I choose the right offer and decline the rest.
Hi, @krishnanvr Do you use up all available resources in that agent's
offer? If so, that agent could not provide offers anymore until the
resource release.

And you may consider starting the master with the `GLOG_v=1` environment
variable which would print more detail logs to help you debug this.

On Sat, Nov 26, 2016 at 5:05 PM, Krishnanarayanan VR <krishnanvr@phonepe.com
> wrote:

> Hello:
>
> Is there a way to force ResourceOffers to get offers from all available
> slaves ?
>
> Let me clarify:
>
> I have a single framework in my cluster. Each time ResourceOffers gets the
> list of offers, I choose the right offer and decline the rest. But I notice
> that next time a callback to ResourceOffers occurs, only a subset of slaves
> is present in the offer. The slave from offer that was chosen in the
> previous iteration is invariably absent.
>
> I also tried to set refuse_seconds to 0 in  both LaunchTasks and
> Decline(egs below):
>
> driver.DeclineOffer(offer.Id, &mesos.Filters{RefuseSeconds:
> proto.Float64(0)})
>
> ^^ but that didn't seem to help.
>
> Any pointers how I can make sure am presented with offers from all the
> slaves all the time ?
>
> Thanks
>
>
>
>


-- 
Best Regards,
Haosdent Huang