Fix race in membership and ring consistency #112
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Without this change, membership and ring could be not in sync. Scenario:
A - current node.
B - alive (according to C).
B - suspect (according to D).
C - thinks B is faulty.
D - thinks B is alive.
At the same time:
Changes in the ring:
3. The lock on m.members is released, so updates on Ring due to (1) and
(2) can execute in any order. In our example, the operations are
reversed:
3.1. A adds B to the ring list due to (2).
3.2. A removes B to the ring list due to (1).
Now B is in the membership list, but not in the ring.
This is the second iteration of the change. First iteration did not
introduce a new lock, but a new invariant: a goroutine needs to lock
both memberlist and disseminator, lock memberlist first, then
disseminator. However, the invariant caused requestAdminStats to fail on
travis-ci, and I couldn't find out why.