[MESOS-8403] Add agent HTTP API operator call to mark local resource providers as gone

• Type: Improvement
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.5.0
• Fix Version/s: 1.8.0
• Component/s: agent, storage
• Labels: mesosphere
• Epic Link: Resource Provider and CSI Tech Debt
• Sprint: Mesosphere Sprint 2018-25, Mesosphere Sprint 2018-26, Mesosphere Sprint 2018-27, Mesosphere Sprint 2018-28, Mesosphere Sprint 2018-29, Mesosphere RI-6 Sprint 2018-30, Mesosphere RI-6 Sprint 2018-31, Storage R7 Sprint 33, Storage R8 Sprint 34, Storage R8 Sprint 35, Storage R9 Sprint 36, Storage R9 Sprint 37, Storage R10 Sprint 38
• Story Points: 3
• Assignee: Benjamin Bannier
• Reporter: Benjamin Bannier
• Shepherd: Chun-Hung Hsiao
• Created: 05/Jan/18 15:03
• Updated: 18/Jan/19 10:04
• Resolved: 18/Jan/19 10:04

Description

It is currently not possible to mark local resource providers as gone (e.g., after agent reconfiguration). As resource providers registered at earlier times could still be cached in a number of places, e.g., the agent or the master, the only way to e.g., prevent this cache from growing too large is to fail over caching components (to e.g., prevent an agent cache to update a fresh master cache during reconciliation).

Showing unavailable and known to be gone resource providers in various endpoints is likely also confusing to users.

We should add an operator call to mark resource providers as gone. While the entity managing resource provider subscription state is the resource provider manager, it still seems to make sense to add this operator call to the agent API as currently only local resource providers are supported. The agent would then forward the call to the resource provider manager which would transition its state for the affected resource provider, e.g., setting its state to GONE and removing it from the list of known resource providers, and then send out an update to its subscribers.

Issue Links

• blocks: MESOS-9303 Clean up SLRP checkpoints after they are marked as gone.
• is blocked by: MESOS-8402 Resource provider manager should persist resource provider information
• supercedes: MESOS-7537 Add functionality to disconnect resource providers in the master