-
Type:
Bug
-
Resolution: Done
-
Priority:
Major - P3
-
Affects Version/s: 3.2.2
-
Component/s: Cluster Management
-
Labels:None
From BaseCluster#selectServer():
long startTimeNanos = System.nanoTime();
long endTimeNanos = startTimeNanos + getUseableTimeoutInNanoseconds();
long curTimeNanos = startTimeNanos;
while (true) {
throwIfIncompatible(curDescription);
if (server != null) {
return server;
}
if (curTimeNanos > endTimeNanos) {
throw createTimeoutException(serverSelector, curDescription);
}
private long getUseableTimeoutInNanoseconds() {
if (settings.getServerSelectionTimeout(NANOSECONDS) < 0) {
return Long.MAX_VALUE;
}
return settings.getServerSelectionTimeout(NANOSECONDS);
}
When the server selection timeout is less than 0 (to produce 'infinite' waiting), endTimeNanos will overflow, and curTimeNanos > endTimeNanos will be trivially true.
I guess the work-around is to use a reasonably big value instead of something less-than-0, and hope the issue is fixed before that value leads to overflows 
Note that master has the same issue (determined by looking at the code), also note that the #getUseableTimeoutInNanoseconds() is called from multiple places, and at least #getDescription() has the same issue.