ascherbakoff commented on a change in pull request #126:
URL: https://github.com/apache/ignite-3/pull/126#discussion_r631819435
##########
File path:
modules/network/src/main/java/org/apache/ignite/network/scalecube/ScaleCubeTopologyService.java
##########
@@ -54,37 +52,16 @@ void setLocalMember(Member member) {
void onMembershipEvent(MembershipEvent event) {
ClusterNode member = fromMember(event.member());
- String memberName = member.name();
-
- switch (event.type()) {
- case ADDED:
- members.put(memberName, member);
-
- fireAppearedEvent(member);
-
- break;
-
- case LEAVING:
- members.remove(memberName);
-
- fireDisappearedEvent(member);
+ if (event.isAdded()) {
+ this.members.put(member.address(), member);
- break;
-
- case REMOVED:
- // In case if member left non-gracefully, without sending
LEAVING event.
- if (members.remove(memberName) != null)
- fireDisappearedEvent(member);
-
- break;
-
- case UPDATED:
- // No-op.
- break;
-
- default:
- throw new IgniteInternalException("This event is not
supported: event = " + event);
+ fireAppearedEvent(member);
+ }
+ else if (event.isRemoved()) {
+ this.members.compute(member.address(), // Ignore stale remove
event.
+ (k, v) -> v.id().equals(member.id()) ? null : v);
+ fireDisappearedEvent(member);
Review comment:
The event is "stale" only in the context of membership view.
But, because each event now is bound to specific cluster id, it has to be
fired or it will be lost.
For example, it's a correct event reordering on some cluster node:
join A(id=1)
fail A(id=1)
join A(id=2)
--->
join A(id=1)
join A(id=2)
fail(A(id=1)
##########
File path:
modules/network/src/main/java/org/apache/ignite/network/scalecube/ScaleCubeTopologyService.java
##########
@@ -54,37 +52,16 @@ void setLocalMember(Member member) {
void onMembershipEvent(MembershipEvent event) {
ClusterNode member = fromMember(event.member());
- String memberName = member.name();
-
- switch (event.type()) {
- case ADDED:
- members.put(memberName, member);
-
- fireAppearedEvent(member);
-
- break;
-
- case LEAVING:
- members.remove(memberName);
-
- fireDisappearedEvent(member);
+ if (event.isAdded()) {
+ this.members.put(member.address(), member);
- break;
-
- case REMOVED:
- // In case if member left non-gracefully, without sending
LEAVING event.
- if (members.remove(memberName) != null)
- fireDisappearedEvent(member);
-
- break;
-
- case UPDATED:
- // No-op.
- break;
-
- default:
- throw new IgniteInternalException("This event is not
supported: event = " + event);
+ fireAppearedEvent(member);
+ }
+ else if (event.isRemoved()) {
+ this.members.compute(member.address(), // Ignore stale remove
event.
+ (k, v) -> v.id().equals(member.id()) ? null : v);
+ fireDisappearedEvent(member);
Review comment:
The event is "stale" only in the context of membership view.
But, because each event now is bound to specific cluster id, it has to be
fired or it will be lost.
For example, it's a correct event reordering on some cluster node B:
join A(id=1)
fail A(id=1)
join A(id=2)
--->
join A(id=1)
join A(id=2)
fail A(id=1)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
