Cluster Managment

Emcache provides a public interface for making different cluster opeations, by calling the method emcache.Client.cluster_managment() an instance of a emcache.ClusterManagment will be returned, the instance returned will use behind the scenes the cluster - for example nodes and configuration - that were provided for creating the CLient instance.

Following example shows how an instance of emcache.ClusterManagment can be retrieved:

client = await emcache.create_client(
    [
        emcache.MemcachedHostAddress('localhost', 11211),
        emcache.MemcachedHostAddress('localhost', 11212)
    ]
)
client.cluster_managment().nodes()
await client.close()

We might be interested on provide an intermediate factory for creating a client cache that will retrieve the cluster managment instance for using it later, for example:

async def create_client_cache() -> emcache.Client:
    global cache_cluster_managment
    client = await emcache.create_client(
        [
            emcache.MemcachedHostAddress('localhost', 11211),
            emcache.MemcachedHostAddress('localhost', 11212)
        ]
    )
    cache_cluster_managment = client.cluster_managment()
    return client

emcache.ClusterManagment provides the following methods:

• emcache.ClusterManagment.nodes() Returns the list of nodes that belong to the cluster, where each node is a emcache.MemcachedHostAddress.

• emcache.ClusterManagment.healthy_nodes() Returns the list of nodes that belong to the cluster that are considered healthy, where each node is a emcache.MemcachedHostAddress. Take a look to the advanced options chapter and more specifically at the Healty and Unhealthy nodes for understanding what node healthiness means.

• emcache.ClusterManagment.unhealthy_nodes() Returns the list of nodes that belong to the cluster that are considered unhealthy, where each node is a emcache.MemcachedHostAddress. Take a look to the advanced options chapter and more specifically at the Healty and Unhealthy nodes for understanding what node healthiness means.

• emcache.ClusterManagment.connection_pool_metrics() Returns the most important metrics that have been observed per each connection pool, take a look to the following section for understanding what metrics are being reported.

Connection pool metrics

The emcache.ClusterManagment.connection_pool_metrics() method returns the main metrics observed by each connection pool of each of the cluster nodes. A dictionary having as a key the emcache.MemcachedHostAddress class instances representing each node, and as a value an instance of emcache.ConnectionPoolMetrics. The metrics that are returned as attributes of this class are:

• curr_connections Tells you how many connections are oppened at that specific moment.

• connections_created Tells you how many connections have been created until now.

• connections_created_with_error Tells you how many connections have been created but ended up by having an exception during the creation time.

• connections_purged Tells you how many connections have been purged.

• connections_closed Tells you how many connections have been closed.

• operations_executed Tells you how many operations have been executed.

• operations_executed_with_error Tells you how many operations have been executed but ended up by having an exception during the execution.

• operations_waited Tells you how many operations were delayed becausew they had to wait for an available connection.

• create_connection_avg Tells you what’s the average time that creating a connection took, using the last 100 observed values.

• create_connection_p50 Tells you what’s the percentil 50 time for creating a connection, using the last 100 observed values.

• create_connection_p99 Tells you what’s the percentil 99 time for creating a connection, using the last 100 observed values.

• create_connection_upper Tells you what’s the worst time that creating a connection took, using the last 100 observed values.

These metrics can be push to a time series database for monitoring the execution of the Emcache driver, the user will need to take care of calculating the deltas, if the user is intereseted on them, of each historical value since historical values are accumulated values.

When autobatching is enabled, each count for operations_executed, operations_executed_with_error and operations_waited could imply multiple keys, since autobatching uses a single command for retrieveing multiple keys.