-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/15143/
-----------------------------------------------------------
Review request for mesos, Benjamin Hindman and Vinod Kone.
Repository: mesos-git
Description
-------
This implements the persistence of Slave information in the master.
Upon being elected as a leading Master, we recover the set of admitted slaves
from the Registrar.
Once recovered, we allow time for these slaves to re-register with the Master.
If they do not re-register, we will remove them from the Registrar and notify
frameworks.
The new {re-}registration process is asynchronous: the Master replies to
re-registrations once the slave has been admitted in the Registrar.
A key implementation choice here is whether to remove slave's from the Master's
structs before or after being removed from the Registrar. I've implemented this
so that we remove slaves from the Master's structs _immediately_, but only send
out updates and lost slave messages _after_ being removed from the Registrar.
Subsequent patches will implement:
-> Update the tests to always use a persistent Master! (Already done but I'll
send out separately).
-> Add the ability to upgrade running clusters from a stateless Master to a
registrar-backed Master.
-> Implement the ZK based registry wiring in master/main.cpp and
local/local.cpp
Diffs
-----
src/master/master.hpp e377af8b3ccd932ae411fa2df4c19642a7310d02
src/master/master.cpp 8e14a070e87ebe579b54d05fb1e8b286edb5e459
Diff: https://reviews.apache.org/r/15143/diff/
Testing
-------
Some of the tests need to be altered to handle the stateful master. I've
already done this but I'll be sending it out separately.
Thanks,
Ben Mahler
