Yes, if you re-register with the master, this will invalidate all outstanding offers.
On Mon, Oct 31, 2016 at 2:28 PM, Hendrik Haddorp <[email protected]> wrote: > Right, I have written my own scheduler and sometimes end up in a state > that Mesos believes that there are outstanding offers for my framework but > I don't seem to have received them and the normal Mesos trace is now > showing the IDs when it offers resources just when they get declined or > used. I'll look into using that trace. > > Beside that the question is how one can get back to a state where there > are no outstanding offers. For tasks I can call "reconcileTasks" to check > with Mesos on the tasks state. But there does not seem to be an equivalent > for offers, which is odd given that offers don't timeout by default. Thus I > was wondering what happens if there are communication problems and Mesos > sends out an offer that I never receive. And what happens if my framework > gets reregistered with Mesos, do outstanding offers get automatically > reseted or not? > > On 31.10.2016 18:49, Vinod Kone wrote: > >> Are you running a custom framework? >> >> Can you see in scheduler logs which offers you are receiving? Am I >> understanding your question correctly that Mesos thinks offers are being >> sent to your framework but (you think) your framework hasn't received them? >> >> Note that you can increase logging on the framework (driver) and Mesos >> master by setting GLOG_v=1 in the environment. >> >> On Mon, Oct 31, 2016 at 12:42 AM, Hendrik Haddorp < >> [email protected] <mailto:[email protected]>> wrote: >> >> Hi, >> >> I have a Mesos 0.28.2 system and generally things seem to run >> fine. The "Outstanding Offers" normally shows nothing, which I >> believe is normal. However at some point my framework gets >> disconnected for some odd reason, might be due to some high load >> or so. A few seconds later I receive a reregistered call from >> Mesos. However it looks like around this time offers start to get >> listed on the "Oustanding Offers" page. Even more strangely no >> Mesos log file contains any information for the offer IDs shown. >> Unfortunately the default logging does not show what offer IDs are >> being send out while it shows the IDs that are being declined or >> got accepted. So I don't know when these actually offers got send out. >> >> How can I deal with such situation? Should I: >> Stop the SchedulerDriver when I get disconnected instead of >> waiting for a reregistered call? >> Is it advised to set --offer_timeout to recover from such a >> situation? >> Is there any way to reconcile offers like one can do for tasks? >> >> thanks, >> Hendrik >> >> >> >
