Sync on iterating thru a ConcurrentHashMap?

Question

i've read that synchronizing on collections while iterating thru them is tricky.
i "think" one should normally do this:
Map<Connection, Boolean> connPool = new HashMap<Connection, Boolean>();
Set<Connection> connSet = map.keySet();
Iterator it = connSet.iterator();
syncrhonize(it) {  // correct object to synchronize on
  while(it.hasNext()) {
}as opposed to doing this:
Map<Connection, Boolean> connPool = new HashMap<Connection, Boolean>();
Set<Connection> connSet = map.keySet();
Iterator it = connSet.iterator();
syncrhonize(connSet) {  // <-- _WRONG_ object to synchronize on
  while(it.hasNext()) {
}now, what if i am using
java.util.concurrent.ConcurrentHashMap
i read the API:
+....... However, iterators are designed to be used by only one thread at a time. ......+
my interpretation of this is that
if i iterate on a ConcurrentHashMap i would still need to synchronize on the Iterator (just like in a standard HashMap ), right?

Answer 1

the situation is:
i have many threads that analyze data and they retrieve from a database.
opening a java.sql.Connection is extremely slow, so i use a Connection pool:
ConcurrentHashMap<Connection, Boolean> connPool;
the Boolean value is "true" if the Connection is not being used by another thread.
for example, i might have 15 data analysis Threads and 5 Connections in the connectionPool.
when a Thread needs to query the database, it asks for an idle Connection.
so, i Iterate over the keySet() of the connPool Map looking for a "true" value. if i iterate thru the whole Map and do not find an
idle Connection, i add a new Connection. (so far, there is no need for synchronizing).
but then what i do to increase performance, because opening a new Connection is so slow, i start a background thread to create and then add a new Connection and the iteration continues in parallel. (note: an already open Connection might be released before a new Connection is created)
once the background thread opens a new Connection, it will connPool.put(dbConn, Boolean.TRUE); while the iteration over connPool is continuing in the primary Connection pool object.