public void startNodeRepairAndWaitFinish(boolean canResume, boolean crossVdc) throws Exception {
if (canResume && getLastSucceededRepairStatus(true) != null) {
log.info("Resume last successful repair");
return;
}
DbRepairStatus state = null;
for (int i = 0; i < DB_REPAIR_MAX_RETRY_COUNT; i++) {
startNodeRepair(canResume, crossVdc);
state = waitDbRepairFinish(true);
if (state != null) {
break;
}
// It could be cluster state changed, so we have to wait for ANY repair to finish here
// We don't care if it's NotFound, Success, Or Failed for other state, repair for current
// state is failed anyway.
log.error(
"No db repair found for current cluster state, waiting for possible stale repair to finish");
state = waitDbRepairFinish(false);
// Trigger a new db repair
log.info("Trigger a new db repair for current cluster state");
}
if (state.getStatus() == DbRepairStatus.Status.FAILED) {
log.error("Db node repair started at {} is failed", state.getStartTime());
throw new IllegalStateException("Repair failed");
}
log.info("Db node repair started at {} is finished", state.getStartTime());
}
public DbRepairStatus waitDbRepairFinish(boolean forCurrentStateOnly) throws Exception {
for (int lastProgress = -1; ; Thread.sleep(1000)) {
DbRepairStatus status = getLastRepairStatus(forCurrentStateOnly);
if (status == null) {
log.info(
"No db repair found(forCurrentStateOnly={})", forCurrentStateOnly ? "true" : "false");
return null;
}
if (status.getStatus() != DbRepairStatus.Status.IN_PROGRESS) {
log.info(
"Db repair(forCurrentStateOnly={}) finished with state: {}",
forCurrentStateOnly ? "true" : "false",
status.toString());
return status;
}
int newProgress = status.getProgress();
if (newProgress != lastProgress) {
log.info("Db repair started at {} is in progress {}%", status.getStartTime(), newProgress);
lastProgress = newProgress;
}
}
}
