Understanding Keystores and Truststores
Configuring Cloudera Manager Server and cluster components to use TLS/SSL requires obtaining keys, certificates, and related security artifacts. The following provides a brief overview.
Java Keystore and Truststore
All clients in a Cloudera Manager cluster configured for TLS/SSL need access to the truststore to validate certificates presented during TLS/SSL session negotiation. The certificates assure the client or server process that the issuing authority for the certificate is part of a legitimate chain of trust.
The standard Oracle Java JDK distribution includes a default truststore (cacerts) that contains root certificates for many well-known CAs, including Symantec. Rather than using the default truststore, Cloudera recommends using the alternative truststore, jssecacerts. The alternative truststore is created by copying cacerts to that filename (jssecacerts). Certificates can be added to this truststore when needed for additional roles or services. This alternative truststore is loaded by Hadoop daemons at startup.
The private keys are maintained in the keystore.
Although the keystore and truststore in some environments may comprise the same file, as configured for Cloudera Manager Server and CDH clusters, the keystore and truststore are distinct files. For Cloudera Manager Server clusters, each host should have its own keystore, while several hosts can share the same truststore. This table summarizes the general differences between keystore and the truststore in Cloudera Manager Server clusters.
| Keystore | Truststore |
|---|---|
| Used by the server side of a TLS/SSL client-server connection. | Used by the client side of a TLS/SSL client-server connection. |
| Typically contains 1 private key for the host system. | Contains no keys of any kind. |
| Contains the certificate for the host's private key. | Contains root certificates for well-known public certificate authorities. May contain certificates for intermediary certificate authorities. |
| Password protected. Use the same password for the key and its keystore. | Password-protection not needed. However, if password has been used for the truststore, never use the same password as used for a key and keystore. |
| Password stored in a plaintext file read permissions granted to a specific group only (OS filesystem permissions set to 0440, hadoop:hadoop). | Password (if there is one for the truststore) stored in a plaintext file readable by all (OS filesystem permissions set to 0440). |
| No default. Provide a keystore name and password when you create the private key and CSR for any host system. | For Java JDK, cacerts is the default unless the alternative default jssecacerts is available. |
| Must be owned by hadoop user and group so that HDFS, MapReduce, YARN can access the private key. | HDFS, MapReduce, and YARN need client access to truststore. |
The details in the table above are specific to the Java KeyStore (JKS) format, which is used by Java-based cluster services such as Cloudera Manager Server, Cloudera Management Service, and many (but not all) CDH components and services. See Certificate Formats (JKS, PEM) and Cluster Components for information about certificate and key file type used various processes.
CDH Services as TLS/SSL Servers and Clients
Cluster services function as a TLS/SSL server, client, or both:
| Component | Client | Server |
|---|---|---|
| Flume |  |
|
| HBase |  |
|
| HDFS | |
|
| Hive | |
|
| Hue (Hue is a TLS/SSL client of HDFS, MapReduce, YARN, HBase, and Oozie.) | |
|
| MapReduce | |
|
| Oozie | |
|
| YARN | |
|
Daemons that function as TLS/SSL servers load the keystores when starting up. When a client connects to an TLS/SSL server daemon, the server transmits the certificate loaded at startup time to the client, and the client then uses its truststore to validate the certificate presented by the server.
Certificate Formats (JKS, PEM) and Cluster Components
Cloudera Manager Server, Cloudera Management Service, and many other CDH services use JKS formatted keystores and certificates. Cloudera Manager Agent, Hue, Key Trustee Server, Impala, and other Python or C++ based services require PEM formatted certificates and keystores rather than Java. Specifically, PEM certificates conform to PKCS #8, which requires individual Base64-encoded text files for certificate and password-protected private key file. The table summarizes certificate types required by several components.
| Component | JKS | PEM |
|---|---|---|
| Flume | |
|
| HBase | |
|
| HDFS | |
|
| Hive (Hive clients and HiveServer 2) | |
|
| Hue | |
|
| Impala | |
|
| MapReduce | |
|
| Oozie | |
|
| Solr | |
|
| YARN | |
|
For more information, see:
Recommended Keystore and Truststore Configuration
- Create a separate keystore for each host. Each keystore should have a name that helps identify it as to the type of host—server or agent, for example. The keystore contains the private key and should be password protected.
- Create a single truststore that can be used by the entire cluster. This truststore contains the root CA and intermediate CAs used to authenticate certificates presented during TLS/SSL handshake. The truststore does not need to be password protected. (See How To Add Root and Intermediary CAs to Truststore for more information about the truststore for TLS/SSL and Cloudera clusters.)
The steps included in Generate TLS Certificates follow this approach.