See Master::exited() https://github.com/apache/mesos/blob/0.22.1/src/master/master.cpp#L878 which derives from Process::exited() https://github.com/apache/mesos/blob/0.22.1/3rdparty/libprocess/include/process/process.hpp#L55 In the event of a temporary network partition, the Mesos master will continue trying to send offer/status/etc. messages to the framework scheduler. Since status messages are reliable at-least once delivery, they are actually queued up (per task) on the slave until an acknowledgement is received from scheduler.
On Wed, Jul 1, 2015 at 5:08 PM, Ying Ji <[email protected]> wrote: > Hey, I am new to mesos and just start to investigate it. I have a > fundament question about Framework > > Assume i am using a long live framework, how the mesos master detect the > framework in unavailable , such as some network error or some internal > error from the framework ? (I can not find it at mater.cpp. Could you > please point out the source code for me ?)The framework has been registered > to the master successfully, and has successfully run for a whole. > > Thanks > > Ying >
