> On June 24, 2017, 2:21 a.m., Adam B wrote: > > src/slave/slave.cpp > > Lines 1405-1407 (original), 1405-1408 (patched) > > <https://reviews.apache.org/r/60405/diff/2/?file=1761692#file1761692line1405> > > > > If this agent has refinements, and we send post format to an old > > master, will the old master safely reject the registration, crash and burn, > > or something in between?
The master will basically consider the resources to be unreserved; because the master and agent will have inconsistent views of the resource state at the agent, this will cause problems. Since you need a new master to create reservation refinements in the first place, you can only arrive in this situation by: Upgrading master Upgrading agent Creating res refinement Downgrading master Which arguably falls under the "don't downgrade if you are using new featues" bucket. But yes, this is certainly unfortunate. Hard to prevent without introducing something similar to master capabilities, which we definitely need (MESOS-5675). I'll drop this issue for now, since AFAIK there's not much we can do to improve this in the short term. > On June 24, 2017, 2:21 a.m., Adam B wrote: > > src/slave/slave.cpp > > Line 1408 (original), 1412-1414 (patched) > > <https://reviews.apache.org/r/60405/diff/2/?file=1761692#file1761692line1412> > > > > We could at least log an INFO/WARN if we aren't able to downgrade, and > > still send it anyway. Hmm, not sure a warning/log is warranted here. In the common case (refined reservation, >= 1.4 master, >= 1.4 agent), we will fail to downgrade the resources, but that is fine and expected. Should we really be cluttering the logs with this information? - Neil ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/60405/#review178834 ----------------------------------------------------------- On June 24, 2017, 1:48 a.m., Neil Conway wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/60405/ > ----------------------------------------------------------- > > (Updated June 24, 2017, 1:48 a.m.) > > > Review request for mesos, Adam B and Michael Park. > > > Repository: mesos > > > Description > ------- > > Documented resource format in agent <-> master protocol. > > > Diffs > ----- > > src/messages/messages.proto 2c086263fdcee4d54a76a61379c2d4dba5271d23 > src/slave/slave.cpp f808458849bb9667a91abe18868751d377d36e0c > > > Diff: https://reviews.apache.org/r/60405/diff/2/ > > > Testing > ------- > > > Thanks, > > Neil Conway > >
