Internal Authentication

Requests between Presto nodes are authenticated using a shared secret. The shared secret must be configured on all nodes in the cluster:

internal-communication.shared-secret=<secret>

A large random key is recommended, and can be generated with the following Linux command:

openssl rand 512 | base64

Internal SSL/TLS configuration

SSL/TLS is configured in the config.properties file. The SSL/TLS on the worker and coordinator nodes are configured using the same set of properties. Every node in the cluster must be configured. Nodes that have not been configured, or are configured incorrectly, are not able to communicate with other nodes in the cluster.

To enable SSL/TLS for Presto internal communication, do the following:

1. Disable HTTP endpoint.

   http-server.http.enabled=false
   

   Warning

   You can enable HTTPS, while leaving HTTP enabled. In most cases this is a security hole. If you are certain you want to use this configuration, you should consider using an firewall to limit access to the HTTP endpoint to only those hosts that should be allowed to use it.

2. Configure the cluster to communicate using the fully qualified domain name (fqdn) of the cluster nodes. This can be done in either of the following ways:

   • If the DNS service is configured properly, we can just let the nodes introduce themselves to the coordinator using the hostname taken from the system configuration (hostname --fqdn)

     node.internal-address-source=FQDN
     

   • It is also possible to specify each node’s fully-qualified hostname manually. This is different for every host. Hosts should be in the same domain to make it easy to create the correct SSL/TLS certificates. e.g.: coordinator.example.com, worker1.example.com, worker2.example.com.

     node.internal-address=<node fqdn>
     

3. Generate a Java Keystore File. Every Presto node must be able to connect to any other node within the same cluster. It is possible to create unique certificates for every node using the fully-qualified hostname of each host, create a keystore that contains all the public keys for all of the hosts, and specify it for the client (see step #8 below). In most cases it is simpler to use a wildcard in the certificate as shown below.

   keytool -genkeypair -alias example.com -keyalg RSA -keystore keystore.jks
   Enter keystore password:
   Re-enter new password:
   What is your first and last name?
     [Unknown]:  *.example.com
   What is the name of your organizational unit?
     [Unknown]:
   What is the name of your organization?
     [Unknown]:
   What is the name of your City or Locality?
     [Unknown]:
   What is the name of your State or Province?
     [Unknown]:
   What is the two-letter country code for this unit?
     [Unknown]:
   Is CN=*.example.com, OU=Unknown, O=Unknown, L=Unknown, ST=Unknown, C=Unknown correct?
     [no]:  yes
   
   Enter key password for <presto>
           (RETURN if same as keystore password):
   

4. Distribute the Java Keystore File across the Presto cluster.

5. Enable the HTTPS endpoint.

   http-server.https.enabled=true
   http-server.https.port=<https port>
   http-server.https.keystore.path=<keystore path>
   http-server.https.keystore.key=<keystore password>
   

6. Change the discovery uri to HTTPS.

   discovery.uri=https://<coordinator fqdn>:<https port>
   

7. Configure the internal communication to require HTTPS.

   internal-communication.https.required=true
   

8. Configure the internal communication to use the Java keystore file.

   internal-communication.https.keystore.path=<keystore path>
   internal-communication.https.keystore.key=<keystore password>
   

Performance with SSL/TLS enabled

Enabling encryption impacts performance. The performance degradation can vary based on the environment, queries, and concurrency.

For queries that do not require transferring too much data between the Presto nodes e.g. SELECT count(*) FROM table, the performance impact is negligible.

However, for CPU intensive queries which require a considerable amount of data to be transferred between the nodes (for example, distributed joins, aggregations and window functions, which require repartitioning), the performance impact can be considerable. The slowdown may vary from 10% to even 100%+, depending on the network traffic and the CPU utilization.

Advanced Performance Tuning

In some cases, changing the source of random numbers improves performance significantly.

By default, TLS encryption uses the /dev/urandom system device as a source of entropy. This device has limited throughput, so on environments with high network bandwidth (e.g. InfiniBand), it may become a bottleneck. In such situations, it is recommended to try to switch the random number generator algorithm to SHA1PRNG, by setting it via http-server.https.secure-random-algorithm property in config.properties on the coordinator and all of the workers:

http-server.https.secure-random-algorithm=SHA1PRNG

Be aware that this algorithm takes the initial seed from the blocking /dev/random device. For environments that do not have enough entropy to seed the SHAPRNG algorithm, the source can be changed to /dev/urandom by adding the java.security.egd property to jvm.config:

-Djava.security.egd=file:/dev/urandom