-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/52723/
-----------------------------------------------------------
(Updated Oct. 11, 2016, 11:29 a.m.)
Review request for mesos and Vinod Kone.
Changes
-------
Backward compatibility with old masters and new agents.
Summary (updated)
-----------------
Changed agent to send TASK_DROPPED during reconciliation.
Bugs: MESOS-6332
https://issues.apache.org/jira/browse/MESOS-6332
Repository: mesos
Description
-------
If a framework attempts to launch a task but the launch message is
dropped after it reaches the master but before it reaches the slave, the
failed launch will be detected during master <-> agent reconciliation
when the agent re-registers. Previously, the agent would generate a
TASK_LOST status update for such dropped tasks; now it will generate
TASK_DROPPED if the framework is partition-aware.
Note that we'll only send TASK_DROPPED if the agent is running a
sufficiently recent version of Mesos (>= 1.1.0). That means that in a
mixed cluster where the master has been upgraded to Mesos 1.1 but some
of the agents have not been, a partition-aware framework might still see
TASK_LOST in this situation.
Diffs (updated)
-----
src/slave/slave.cpp 119fb36c27b25739f2a86a55d48e964ca4a84ff7
src/tests/master_slave_reconciliation_tests.cpp
2983c1b074c2d4179e95e619083f5dd4e9ac6730
src/tests/slave_recovery_tests.cpp 703948f7a6861a4401ee45ce9cae2644106083f3
Diff: https://reviews.apache.org/r/52723/diff/
Testing
-------
`make check`
Thanks,
Neil Conway
