egor-ryashin commented on a change in pull request #6349: maintenance mode for
Historical
URL: https://github.com/apache/incubator-druid/pull/6349#discussion_r221383921
##########
File path:
server/src/main/java/org/apache/druid/server/coordinator/helper/DruidCoordinatorBalancer.java
##########
@@ -106,35 +104,76 @@ private void balanceTier(
return;
}
- final List<ServerHolder> toMoveFrom = Lists.newArrayList(servers);
- final List<ServerHolder> toMoveTo = Lists.newArrayList(servers);
+ Map<Boolean, List<ServerHolder>> partitions = servers.stream()
+
.collect(Collectors.partitioningBy(ServerHolder::isMaintenance));
+ final List<ServerHolder> maintenance = partitions.get(true);
+ final List<ServerHolder> general = partitions.get(false);
+ log.info("Found %d servers in maintenance, %d general servers",
maintenance.size(), general.size());
- if (toMoveTo.size() <= 1) {
- log.info("[%s]: One or fewer servers found. Cannot balance.", tier);
+ if (general.size() == 0) {
+ log.info("[%s]: 0 general servers found. Cannot balance.", tier);
return;
}
int numSegments = 0;
- for (ServerHolder sourceHolder : toMoveFrom) {
+ for (ServerHolder sourceHolder : servers) {
numSegments += sourceHolder.getServer().getSegments().size();
}
-
if (numSegments == 0) {
log.info("No segments found. Cannot balance.");
return;
}
- final BalancerStrategy strategy = params.getBalancerStrategy();
final int maxSegmentsToMove =
Math.min(params.getCoordinatorDynamicConfig().getMaxSegmentsToMove(),
numSegments);
+ int priority =
params.getCoordinatorDynamicConfig().getMaintenanceModeSegmentsPriority();
+ int maintenanceSegmentsToMove = (int) Math.ceil(maxSegmentsToMove *
priority / 10.0);
+ log.info("Balancing %d segments for servers in maintenance mode",
maintenanceSegmentsToMove);
+ Pair<Integer, Integer> maintenanceResult = balanceServers(params,
maintenance, general, maintenanceSegmentsToMove);
+ int generalSegmentsToMove = maxSegmentsToMove - maintenanceResult.lhs;
+ log.info("Balancing %d segments for servers in general mode",
generalSegmentsToMove);
+ Pair<Integer, Integer> generalResult = balanceServers(params, general,
general, generalSegmentsToMove);
+ Pair<Integer, Integer> result = new Pair(
+ generalResult.lhs + maintenanceResult.lhs,
+ generalResult.rhs + maintenanceResult.rhs
+ );
+
+ int moved = result.lhs;
+ int unmoved = result.rhs;
+ if (unmoved == maxSegmentsToMove) {
+ // Cluster should be alive and constantly adjusting
+ log.info("No good moves found in tier [%s]", tier);
+ }
+ stats.addToTieredStat("unmovedCount", tier, unmoved);
+ stats.addToTieredStat("movedCount", tier, moved);
+
+ if (params.getCoordinatorDynamicConfig().emitBalancingStats()) {
+ final BalancerStrategy strategy = params.getBalancerStrategy();
+ strategy.emitStats(tier, stats, Lists.newArrayList(servers));
Review comment:
Ah, I got it, yes, every segment contributes the cost of its current
location to the Total Cost, if we remove servers in maintenance then it removes
those segments from the calculation too, but those servers are still
operational (moreover they can be removed from the maintenance list at any
time).
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
