Use the health api to view the health of the es cluster The usage of the health api is as follows:

curl 'http://localhost:9200/_cluster/health' 

There is a core field status in the return value of health api. Status has three values: green, yellow, red. They represent three states of the cluster: the primary partition and the replica have been allocated, the primary partition has been allocated and the replica partition has not been allocated, and the designated partition has not been allocated.

That is to say, the core of health api is the high availability of data.

See the implementation of health api:
The request will be routed to the master node, and then read the routing table in the clusterState

Based on routing table, judge the partition status of index

    public ClusterShardHealth(int shardId, final IndexShardRoutingTable shardRoutingTable) {
        this.shardId = shardId;
        for (ShardRouting shardRouting : shardRoutingTable) {
            if (shardRouting.active()) {
                activeShards++;
                if (shardRouting.relocating()) {
                    // the shard is relocating, the one it is relocating to will be in initializing state, so we don't count it
                    relocatingShards++;
                }
                if (shardRouting.primary()) {
                    primaryActive = true;
                }
            } else if (shardRouting.initializing()) {
                initializingShards++;
            } else if (shardRouting.unassigned()) {
                unassignedShards++;
            }
        }
        if (primaryActive) {
            if (activeShards == shardRoutingTable.size()) {
                status = ClusterHealthStatus.GREEN;
            } else {
                status = ClusterHealthStatus.YELLOW;
            }
        } else {
            status = ClusterHealthStatus.RED;
        }
    }

Determine the state of index based on fragmentation

    public ClusterIndexHealth(IndexMetaData indexMetaData, IndexRoutingTable indexRoutingTable) {
        this.index = indexMetaData.getIndex();
        this.numberOfShards = indexMetaData.getNumberOfShards();
        this.numberOfReplicas = indexMetaData.getNumberOfReplicas();
        this.validationFailures = indexRoutingTable.validate(indexMetaData);

        for (IndexShardRoutingTable shardRoutingTable : indexRoutingTable) {
            int shardId = shardRoutingTable.shardId().id();
            shards.put(shardId, new ClusterShardHealth(shardId, shardRoutingTable));
        }

        // update the index status
        status = ClusterHealthStatus.GREEN;

        for (ClusterShardHealth shardHealth : shards.values()) {
            if (shardHealth.isPrimaryActive()) {
                activePrimaryShards++;
            }
            activeShards += shardHealth.getActiveShards();
            relocatingShards += shardHealth.getRelocatingShards();
            initializingShards += shardHealth.getInitializingShards();
            unassignedShards += shardHealth.getUnassignedShards();

            if (shardHealth.getStatus() == ClusterHealthStatus.RED) {
                status = ClusterHealthStatus.RED;
            } else if (shardHealth.getStatus() == ClusterHealthStatus.YELLOW && status != ClusterHealthStatus.RED) {
                // do not override an existing red
                status = ClusterHealthStatus.YELLOW;
            }
        }
        if (!validationFailures.isEmpty()) {
            status = ClusterHealthStatus.RED;
        } else if (shards.isEmpty()) { // might be since none has been created yet (two phase index creation)
            status = ClusterHealthStatus.RED;
        }
    }

Based on the index, the state of the cluster is determined.

    public ClusterStateHealth(ClusterState clusterState, String[] concreteIndices) {
        RoutingTableValidation validation = clusterState.routingTable().validate(clusterState.metaData());
        validationFailures = validation.failures();
        numberOfNodes = clusterState.nodes().size();
        numberOfDataNodes = clusterState.nodes().dataNodes().size();

        for (String index : concreteIndices) {
            IndexRoutingTable indexRoutingTable = clusterState.routingTable().index(index);
            IndexMetaData indexMetaData = clusterState.metaData().index(index);
            if (indexRoutingTable == null) {
                continue;
            }

            ClusterIndexHealth indexHealth = new ClusterIndexHealth(indexMetaData, indexRoutingTable);

            indices.put(indexHealth.getIndex(), indexHealth);
        }

        status = ClusterHealthStatus.GREEN;

        for (ClusterIndexHealth indexHealth : indices.values()) {
            activePrimaryShards += indexHealth.getActivePrimaryShards();
            activeShards += indexHealth.getActiveShards();
            relocatingShards += indexHealth.getRelocatingShards();
            initializingShards += indexHealth.getInitializingShards();
            unassignedShards += indexHealth.getUnassignedShards();
            if (indexHealth.getStatus() == ClusterHealthStatus.RED) {
                status = ClusterHealthStatus.RED;
            } else if (indexHealth.getStatus() == ClusterHealthStatus.YELLOW && status != ClusterHealthStatus.RED) {
                status = ClusterHealthStatus.YELLOW;
            }
        }

        if (!validationFailures.isEmpty()) {
            status = ClusterHealthStatus.RED;
        } else if (clusterState.blocks().hasGlobalBlock(RestStatus.SERVICE_UNAVAILABLE)) {
            status = ClusterHealthStatus.RED;
        }

        // shortcut on green
        if (status.equals(ClusterHealthStatus.GREEN)) {
            this.activeShardsPercent = 100;
        } else {
            List<ShardRouting> shardRoutings = clusterState.getRoutingTable().allShards();
            int activeShardCount = 0;
            int totalShardCount = 0;
            for (ShardRouting shardRouting : shardRoutings) {
                if (shardRouting.active()) activeShardCount++;
                totalShardCount++;
            }
            this.activeShardsPercent = (((double) activeShardCount) / totalShardCount) * 100;
        }
    }

To understand the health api, you need to understand clusterState Fortunately, these are read-only information, not difficult to understand.