mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Benjamin Bannier <benjamin.bann...@mesosphere.io>
Subject Re: Review Request 63732: Reconciled offer operations between agent and master.
Date Thu, 30 Nov 2017 16:00:58 GMT


> On Nov. 30, 2017, 2:55 a.m., Jie Yu wrote:
> > src/master/master.cpp
> > Lines 7426 (patched)
> > <https://reviews.apache.org/r/63732/diff/9/?file=1903718#file1903718line7426>
> >
> >     See my comment in https://reviews.apache.org/r/63842/
> >     
> >     since removeOfferOperation will update 'used', you endup mutate 'used' twice.
> >     
> >     I think `Master::removeOfferOperation` should be responsible to maintain invariants
between allocator, Slave and Framework (similar to `Master::updateOfferOperation`). Ideally,
it should not be called  except in operation ack handler.
> >     
> >     I understand that `Master::addOfferOperation` is an exception because it does
not mutatae allocator state (because resources might already be allocated in `_accept` for
example), which is indeed a bit weird.
> >     
> >     But I would rather treat `addOfferOperation` as an exception, instead of treating
`updateOfferOperation` as an exception.
> >     
> >     As a result, I still suggest we update allocator state in `Master::removeOfferOperation`.
No need to call `slave->recoverResources(operation);` here because it'll be called by `Slave::removeOfferOperation`.

We now possibly recover resources from pending, non-speculated offer operation in `Master::removeOfferOperation`
in an updated https://reviews.apache.org/r/63842/.

I also simplified the `updateSlave` handler here to not explictly invoke `Slave::recoverResources`
anymore and instead delegate that to an adjusted `Slave::removeOfferOperation` which I also
adjusted in https://reviews.apache.org/r/63842/.


- Benjamin


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/63732/#review192214
-----------------------------------------------------------


On Nov. 30, 2017, 5 p.m., Benjamin Bannier wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/63732/
> -----------------------------------------------------------
> 
> (Updated Nov. 30, 2017, 5 p.m.)
> 
> 
> Review request for mesos, Jie Yu and Jan Schlicht.
> 
> 
> Bugs: MESOS-8207
>     https://issues.apache.org/jira/browse/MESOS-8207
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> See summary.
> 
> 
> Diffs
> -----
> 
>   src/master/master.cpp fadc78b2ca5d46b8cc12a794b428753aa79ac095 
> 
> 
> Diff: https://reviews.apache.org/r/63732/diff/10/
> 
> 
> Testing
> -------
> 
> `make check`, tested as part of https://reviews.apache.org/r/63843/.
> 
> This patch requires `protobuf::isSpeculativeOperation` from https://reviews.apache.org/r/63751/
which is _not_ part of this review chain (to avoid multiple dependent RRs).
> 
> 
> Thanks,
> 
> Benjamin Bannier
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message