@Override
public void acceptPromotion() {
try {
long startTime = System.currentTimeMillis();
Boolean success = false;
m_term =
createTerm(
m_messenger.getZK(), m_partitionId, getInitiatorHSId(), m_initiatorMailbox, m_whoami);
m_term.start();
while (!success) {
RepairAlgo repair = null;
repair = createPromoteAlgo(m_term.getInterestingHSIds(), m_initiatorMailbox, m_whoami);
// if rejoining, a promotion can not be accepted. If the rejoin is
// in-progress, the loss of the master will terminate the rejoin
// anyway. If the rejoin has transferred data but not left the rejoining
// state, it will respond REJOINING to new work which will break
// the MPI and/or be unexpected to external clients.
if (!m_initiatorMailbox.acceptPromotion()) {
tmLog.error(m_whoami + "rejoining site can not be promoted to leader. Terminating.");
VoltDB.crashLocalVoltDB("A rejoining site can not be promoted to leader.", false, null);
return;
}
m_initiatorMailbox.setRepairAlgo(repair);
// term syslogs the start of leader promotion.
Pair<Boolean, Long> result = repair.start().get();
success = result.getFirst();
if (success) {
m_initiatorMailbox.setLeaderState(result.getSecond());
tmLog.info(
m_whoami
+ "finished leader promotion. Took "
+ (System.currentTimeMillis() - startTime)
+ " ms.");
// THIS IS where map cache should be updated, not
// in the promotion algorithm.
LeaderCacheWriter iv2masters = new LeaderCache(m_messenger.getZK(), m_zkMailboxNode);
iv2masters.put(m_partitionId, m_initiatorMailbox.getHSId());
} else {
// The only known reason to fail is a failed replica during
// recovery; that's a bounded event (by k-safety).
// CrashVoltDB here means one node failure causing another.
// Don't create a cascading failure - just try again.
tmLog.info(
m_whoami
+ "interrupted during leader promotion after "
+ (System.currentTimeMillis() - startTime)
+ " ms. of "
+ "trying. Retrying.");
}
}
// Tag along and become the export master too
ExportManager.instance().acceptMastership(m_partitionId);
} catch (Exception e) {
VoltDB.crashLocalVoltDB("Terminally failed leader promotion.", true, e);
}
}
@Override
public void noticedTopologyChange() {
if (ExportManager.instance() != null) {
ExportManager.instance().notifyOfClusterTopologyChange();
}
}