[MESOS-9674] Allocations can occurr between the multiple resource changes related to orphan operations.

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.8.0
• Component/s: allocation, master
• Labels: foundations
• Target Version/s: 1.9.0
• Assignee: Unassigned
• Reporter: Gastón Kleiman
• Created: 22/Mar/19 20:50
• Updated: 24/Mar/20 17:02

Description

Master::updateSlave() was recently updated to handle orphan operations. The new code doesn't pause the allocator before doing multiple dispatches to the allocator.

This means that an allocation can be interleaved while the allocator is still not fully updated or in an inconsistent state.

I tried just to quickly improve the handler by adding a couple of allocator->pause() and allocator->resume() calls, but that made quite some SLRP tests get stuck and time out waiting for offers.