Added comment to master logic for downgrading checkpointed resources. Review: https://reviews.apache.org/r/60415
Project: http://git-wip-us.apache.org/repos/asf/mesos/repo Commit: http://git-wip-us.apache.org/repos/asf/mesos/commit/ea7c9807 Tree: http://git-wip-us.apache.org/repos/asf/mesos/tree/ea7c9807 Diff: http://git-wip-us.apache.org/repos/asf/mesos/diff/ea7c9807 Branch: refs/heads/master Commit: ea7c9807b649f7305c20515ec60785cd08ea5f33 Parents: 09fcc7f Author: Neil Conway <neil.con...@gmail.com> Authored: Sat Jun 24 12:06:15 2017 -0700 Committer: Neil Conway <neil.con...@gmail.com> Committed: Sat Jun 24 20:57:15 2017 -0700 ---------------------------------------------------------------------- src/master/master.cpp | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/mesos/blob/ea7c9807/src/master/master.cpp ---------------------------------------------------------------------- diff --git a/src/master/master.cpp b/src/master/master.cpp index 33eca0d..09c8beb 100644 --- a/src/master/master.cpp +++ b/src/master/master.cpp @@ -6389,6 +6389,14 @@ void Master::___reregisterSlave( message.mutable_resources()->CopyFrom(slave->checkpointedResources); if (!slave->capabilities.reservationRefinement) { + // If the agent is not refinement-capable, don't send it + // checkpointed resources that contain refined reservations. This + // might occur if a reservation refinement is created but never + // reaches the agent (e.g., due to network partition), and then + // the agent is downgraded before the partition heals. + // + // TODO(neilc): It would probably be better to prevent the agent + // from re-registering in this scenario. Try<Nothing> result = downgradeResources(message.mutable_resources()); if (result.isError()) { LOG(WARNING) << "Not sending updated checkpointed resouces " @@ -8870,6 +8878,14 @@ void Master::_apply(Slave* slave, const Offer::Operation& operation) { message.mutable_resources()->CopyFrom(slave->checkpointedResources); if (!slave->capabilities.reservationRefinement) { + // If the agent is not refinement-capable, don't send it + // checkpointed resources that contain refined reservations. This + // might occur if a reservation refinement is created but never + // reaches the agent (e.g., due to network partition), and then + // the agent is downgraded before the partition heals. + // + // TODO(neilc): It would probably be better to prevent the agent + // from re-registering in this scenario. Try<Nothing> result = downgradeResources(message.mutable_resources()); if (result.isError()) { LOG(WARNING) << "Not sending updated checkpointed resouces "
