----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/71182/#review216951 -----------------------------------------------------------
In the description: it's not correct that the timeout would be reset if the agent fails over. Since the grace period is sent to the executor when the first KILL event is set, an agent failover will cause another KILL event to be sent, but the grace period will not be extended. The timeout _would_ be extended, however, if the agent fails over before it receives the `DrainSlaveMessage`. include/mesos/master/master.proto Lines 473 (patched) <https://reviews.apache.org/r/71182/#comment304185> It would seem a bit more consistent to use `TimeInfo` here, since we have a couple other timestamp fields in this message which use that type. WDYT? include/mesos/v1/master/master.proto Lines 474 (patched) <https://reviews.apache.org/r/71182/#comment304186> Consider `estimated_drain_start_time` for this field name, to help communicate that this is a best guess. - Greg Mann On July 29, 2019, 9:18 p.m., Benjamin Bannier wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/71182/ > ----------------------------------------------------------- > > (Updated July 29, 2019, 9:18 p.m.) > > > Review request for mesos and Greg Mann. > > > Bugs: MESOS-9907 > https://issues.apache.org/jira/browse/MESOS-9907 > > > Repository: mesos > > > Description > ------- > > This information can be used to calculate an approximate time when > draining an agent would be finished, e.g., by comparing > `DrainConfig.max_grace_period` and the field `drain_start_time_seconds` > added here, both obtained from the master API simultaneously. This > information is necessarily approximate as the agent might e.g., fail > over before it has finished draining which would reset the timeout; for > that specific case the master would send the drain request again so > after some time the value calculated from the master API would be in > line with the expected true value. > > WIP: drain start time in master > > > Diffs > ----- > > include/mesos/master/master.proto 07bf4e7194399ef660d7282275f25e1d4563e473 > include/mesos/v1/master/master.proto > 9bcef2f358b50fd63cdfdec5dcad742930de3ecc > src/common/protobuf_utils.cpp 7778e7f2475e9d6125d1c599715c91715f3654d3 > src/master/http.cpp 765d5052ab6a8d731113f8922d20fb280b843003 > src/master/master.hpp 8bdbac90b9000e9126a7fad375642dcf0a03378e > src/master/readonly_handler.cpp a8931153dc1b3f204c1b6437b972fe6607f900bd > src/tests/api_tests.cpp 641eb15153ddb85df322aa6a133ca8e4c6d6a510 > > > Diff: https://reviews.apache.org/r/71182/diff/1/ > > > Testing > ------- > > `make check` > > > Thanks, > > Benjamin Bannier > >
