Close transport on shutdown #92
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Travis integration is currently failing due to failure to build on go 1.2. Building on tip seems to work fine. Remove 1.2 (I don't know why it's failing, but 1.2 is prehistoric), and add 1.4, 1.5, 1.6. Signed-off-by: Alex Bligh <alex@alex.org.uk>
abligh
force-pushed
the
close-transport-on-shutdown
branch
from
bbf9947 to
d9a0fd2
Compare
Heartbeat timers are randomized. This means that (all other things being equal) they may elapse and cause a node to enter candidate state at any time. The heartbeat timer is currently reset when a candidate receives a valid AppendEntries RPC call. This causes an issue as a node that has just voted for a candidate may immediately declare itself a candidate. For instance in a new 3 node cluster, Node A declares itself a candidate, Node A votes for itself, and Node B votes for node A too. Shortly afterwards (but before an AppendEntries heartbeat is sent by Node A), Node B's heartbeat timer expires, and Node B starts an election. This causes an issue within the test suite, characterised by about 1%-2% of test runs failing. In particular A is detected as the leader, and B is subsequently detected as the leader. Issue hashicorp#90 describes this in more detail. The raft specification (section 5.2 of the paper, section 3.4 of the thesis), says: "A server remains in follower state as long as it receives valid RPCs from a leader OR CANDIDATE" (my emphasis). It is thus intended that a node in receipt of a 'valid' RequestVote RPC from a candidate should not declare itself a candidate (but should remain in follower state) for at least the Heartbeat interval. It isn't entirely clear what a 'valid' RequestVote RPC means, but certainly one that is voted on is valid, and this is the smallest change. This one line change thus resets the heartbeat timer if a vote is cast in favour, and thus removes the 1%-2% of tests failing for this reason. Signed-off-by: Alex Bligh <alex@alex.org.uk>
When a peer shuts down, close the transport. This needs to be done in the future (i.e. when all the routines have ended), else the remove peer does not propagate correctly - this was an issue with pull request hashicorp#72. Unfortunately we cannot simply add a Close() to the Transport interface as we would break back compatibility. Therefore add a type assertion to ensure that the interface supports Close before closing it. Add a Close() method to InmemTransport so this actually gets tested. InmemTransport has nothing much to close (though it does call DisconnectAll), but other transports may do. Signed-off-by: Alex Bligh <alex@alex.org.uk>
|
Rebased onto #93 in the hope that fixes travis tests. |
|
Thanks! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
(based on branch reset-heartbeat-on-requestvote which is in an earlier PR)
When a peer shuts down, close the transport. This needs to be done in the future (i.e. when all the routines have ended), else the remove peer does not propagate correctly - this was an issue with pull request #72.
Unfortunately we cannot simply add a
Close()to theTransportinterface as we would break back compatibility. Therefore add a type assertion to ensure that the interface supportsClosebefore closing it.Add a
Close()method toInmemTransportso this actually gets tested.InmemTransporthas nothing much to close (though it does callDisconnectAll), but other transports may do.