-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38876/
-----------------------------------------------------------
(Updated Nov. 10, 2015, 2:22 p.m.)
Review request for mesos, Ben Mahler, Isabel Jimenez, and Vinod Kone.
Changes
-------
Address comments from BenM + Updated Description
Bugs: MESOS-3515
https://issues.apache.org/jira/browse/MESOS-3515
Repository: mesos
Description (updated)
-------
This change adds the relevant functionality to `src/slave/paths.cpp/hpp` to
store a marker file to denote HTTP based executors. We create the file when
`checkpointing` is enabled as part of handling the `Subscribe` request. This is
then used by the agent when recovering to ascertain if the executor was
connected via `HTTP` before the agent restart.
-- Detailed Explanation of Changes ( not to be included in the commit message )
This marker file is used when recovering HTTP based executors (assuming
framework checkpointing is enabled). Currently we support the following
recovery options on the agent.
1. `--cleanup` : If `PID` marker file is not found, the current behavior is to
directly destroy the container the executor was running in. With the help of
this `HTTP` marker file, we can now check if the executor was connected via
HTTP previously and if so, send it a `Event::SHUTDOWN` when it retries the
`Subscribe` call.
2. `--reconnect` : If `PID` marker file is not found, the current behavior is
to just `LOG` that we were not able to reconnect back to the executor. With the
help of the `HTTP` marker file, we are able to correctly distinguish between
the cases when a `PID` based executor failed to checkpoint its PID and it being
an `HTTP` based executor. An example:
https://github.com/apache/mesos/blob/master/src/slave/slave.cpp#L4177
Diffs (updated)
-----
src/slave/paths.hpp f743fb4b1ca278fade9134e0ae8f6a6450d4a977
src/slave/paths.cpp aab7a4b63f0e7c2104097077369bb10bcd28c6a1
Diff: https://reviews.apache.org/r/38876/diff/
Testing
-------
make check
Thanks,
Anand Mazumdar
