> On Nov. 18, 2015, 8:05 a.m., Qian Zhang wrote: > > src/master/http.cpp, line 2097 > > <https://reviews.apache.org/r/40244/diff/2/?file=1125056#file1125056line2097> > > > > This comment may not be related to this patch. I am just curious, for > > ```required```, we have called ```flatten()``` to remove role from it when > > passing it into ```Master::Http::_operation()```. But here for > > ```offer->resources()```, we do not call ```flatten()``` to remove its > > role, so is it subtractable between ```required``` and > > ```offer->resources()```? > > Michael Park wrote: > We `flatten` the resources for `/reserve` because the `required` > represents the resources that we need in order to perform the reservation. > Suppose we want to end up with 2 reserved cpus. We `flatten` it, and > search for 2 unreserved cpus in the received offers. > If we were to `flatten` the `offer.resources()` also, then an offer with > 1 reserved cpu and 1 unreserved cpu would incorrectly satisfy our condition. > > In terms of subtractibility, that is determined on a per-resource basis, > and non-subtractable resources are ignored. > For example, `[(2 unreserved cpus)] - [(1 reserved cpu), (1 unreserved > cpu)]` == `[(1 unreserved cpu)]`. > > Qian Zhang wrote: > Suppose ```offer.resources()``` is ```[(1 reserved cpu), (1 unreserved > cpu)]```, and ```required``` is ```[(2 unreserved cpus)]```, so the result of > ```required == required - offer.resources()``` is ```false```, that mean we > will rescind this offer, right? But it may not make sense to me, because I > think the **reserved** resource should not be rescinded.
Yes, it's clear that this approach has limitations. It was just the best known approach given the current state of things since we cannot currently rescind offers partially. While rescinding the __reserved__ resource is undesirable, it still meets the requirements as the resource will still be offered to the role. - Michael ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/40244/#review106992 ----------------------------------------------------------- On Nov. 18, 2015, 11:27 a.m., Neil Conway wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/40244/ > ----------------------------------------------------------- > > (Updated Nov. 18, 2015, 11:27 a.m.) > > > Review request for mesos, Greg Mann and Michael Park. > > > Bugs: MESOS-2455 > https://issues.apache.org/jira/browse/MESOS-2455 > > > Repository: mesos > > > Description > ------- > > Clarified comments in Master::Http::_operation. > > > Diffs > ----- > > src/master/http.cpp 1c4f1406f5d917f5d655a7d61d311365f8999ce0 > > Diff: https://reviews.apache.org/r/40244/diff/ > > > Testing > ------- > > > Thanks, > > Neil Conway > >
