MongoDB replica heartbeat request time exceeded

I have replica set (hosted on amazon) which has:

• primary
• secondary
• arbiter

All of them are version 3.2.6 and this replica is making one shard in my sharded cluster (if that is important although I think it is not).

When I type rs.status() on primary it says that cannot reach secondary (the same thing is on arbiter):

        {
        "_id" : 1,
        "name" : "secondary-ip:27017",
        "health" : 0,
        "state" : 8,
        "stateStr" : "(not reachable/healthy)",
        "uptime" : 0,
        "optime" : {
            "ts" : Timestamp(0, 0),
            "t" : NumberLong(-1)
        },
        "optimeDate" : ISODate("1970-01-01T00:00:00Z"),
        "lastHeartbeat" : ISODate("2016-07-20T15:40:50.479Z"),
        "lastHeartbeatRecv" : ISODate("2016-07-20T15:40:51.793Z"),
        "pingMs" : NumberLong(0),
        "lastHeartbeatMessage" : "Couldn't get a connection within the time limit",
        "configVersion" : -1
    }

(btw look at the optimeDate O.o)

Error in my log is:

[ReplicationExecutor] Error in heartbeat request to secondary-ip:27017; ExceededTimeLimit: Couldn't get a connection within the time limit

Strange thing is that when I go on secondary and type rs.status() everything looks OK. Also I am able to connect to secondary from my primary instance (with mongo --host secondary) so I guess it is not network issue. Yesterday it was all working fine.

TL;DR my primary cannot see secondary and arbiter cannot see secondary and my secondary sees primary and it was all working fine just day ago and I am able manually connect to secondary from primary instance.

Anyone has an idea what could go wrong?

Tnx, Ivan