shards_availability health indicator turns yellow when creating new index with replicas

[idegtiarenko]

Description

Unassigned shards caused by create index (or similar operations) are ignored and considered available in order to keep the shards_availability health indicator green:

elasticsearch/server/src/main/java/org/elasticsearch/cluster/routing/allocation/ShardsAvailabilityHealthIndicatorService.java

Lines 405 to 407 in d8e7373

	boolean isNew = isUnassignedDueToNewInitialization(routing);
	boolean isRestarting = isUnassignedDueToTimelyRestart(routing, shutdowns);
	available &= routing.active() || isRestarting || isNew;

However the isUnassignedDueToNewInitialization check itself is only implemented for primaries:

elasticsearch/server/src/main/java/org/elasticsearch/cluster/routing/allocation/ShardsAvailabilityHealthIndicatorService.java

Lines 457 to 459 in d8e7373

	private static boolean isUnassignedDueToNewInitialization(ShardRouting routing) {
	return routing.primary() && routing.active() == false && getInactivePrimaryHealth(routing) == ClusterHealthStatus.YELLOW;
	}

as a result:

• when creating a new index without replicas the health indicator remains green
• when creating the same index with replicas the health indicator turns yellow until replicas are initialized

The check could be extended for replicas like following (to verify the status of corresponding primary):

    private static boolean isUnassignedDueToNewInitialization(ShardRouting routing, ClusterState state) {
        if (routing.active() == false) {
            return false;
        }
        if (routing.primary()) {
            return getInactivePrimaryHealth(routing) == ClusterHealthStatus.YELLOW;
        } else {
            var primary = state.routingTable().shardRoutingTable(routing.shardId()).primaryShard();
            return getInactivePrimaryHealth(primary) == ClusterHealthStatus.YELLOW;
        }
    }

however overall indicator would still turn yellow when primary is active and replica is still initializing.

[elasticsearchmachine]

Pinging @elastic/es-data-management (Team:Data Management)