Three node cluster doesn't have all peers set correctly?

[slfritchie]

Hi. I was fiddling with a 3 node cluster and saw some odd behavior on commit 097fcb2.

• For all nodes, rm -rf ./data, just in case.
• For all nodes, start Erlang with short node names p0@sbb4, p1@sbb4, and p2@sbb4. (Your hostname will vary, of course.) E.g. erl -sname p0 -pz ebin _build/*/*/*/ebin
• For all nodes, run plumtree:start().
• On node p0, run: [{goo, X} ! hello || X <- [p1@sbb4, p2@sbb4] ].
• On node p0, run, [plumtree_peer_service:join(X) || X <- nodes()].

Then, if I run the following:

f(All).
All = lists:sort([node()|nodes()]).
[{X, Y, catch plumtree_broadcast:debug_get_peers(X, Y)} || X <- All, Y <- All].

I see the following oddity on node p0:

[{p0@sbb4,p0@sbb4,{[],[]}},
{p0@sbb4,p1@sbb4,{[],[]}},
{p0@sbb4,p2@sbb4,{[],[]}},
{p1@sbb4,p0@sbb4,{[p2@sbb4],[p0@sbb4]}},
{p1@sbb4,p1@sbb4,{[p2@sbb4],[p0@sbb4]}},
{p1@sbb4,p2@sbb4,{[p2@sbb4],[p0@sbb4]}},
{p2@sbb4,p0@sbb4,{[p0@sbb4],[p1@sbb4]}},
{p2@sbb4,p1@sbb4,{[p0@sbb4],[p1@sbb4]}},
{p2@sbb4,p2@sbb4,{[p0@sbb4],[p1@sbb4]}}]

... which means that replication of any metadata put() on node p0 is needlessly (?) delayed by 10s of seconds before appearing on p1 and p2.

If I stop & start the VM of node p0 and restart Plumtree, the peer sets are non-empty as expected.

[tsantero]

Was on vacation past few days. Looking into this tonight. Thanks for opening the issue!

[benoitc]

what's the status of this issue?