Slack Docker Pulls

This document describes the following security related features in Alluxio.

  1. User Authentication: Alluxio filesystem will differentiate users accessing the service when the authentication mode is SIMPLE. Alluxio also supports NOSASL and KERBEROS as authentication mechanisms. Having authentication mode to be SIMPLE, or KERBEROS is required for authorization.
  2. User Authorization: Alluxio filesystem will grant or deny user access based on the requesting user and the POSIX permissions model of the files or directories to access, when Note that, authentication cannot be NOSASL as authorization requires user information.
  3. Access Control Lists: In addition to the POSIX permission model, Alluxio implements an Access Control List (ACL) model similar to those found in Linux and HDFS. The ACL model is more flexible and allows administrators to manage any user or group’s permissions to any file system object.
  4. Client-Side Hadoop Impersonation: Alluxio supports client-side Hadoop impersonation so the Alluxio client can access Alluxio on the behalf of the Hadoop user. This can be useful if the Alluxio client is part of an existing Hadoop service.
  5. Auditing: If enabled, the Alluxio filesystem writes an audit log for all user accesses.
  6. Encryption: Alluxio supports end-to-end data encryption, and TLS for network communication.

See Security specific configuration for different security properties.


The authentication protocol is determined by the configuration property, with a default value of SIMPLE.


Authentication is enabled when the authentication type is SIMPLE.

A client must identify itself with a username to the Alluxio service. If the property is set on the Alluxio client, its value will be used as the login user, otherwise, the login user is inferred from the operating system user executing the client process. The provided user information is attached to the corresponding metadata when the client creates directories or files.


Authentication is disabled when the authentication type is NOSASL.

The Alluxio service will ignore the user of the client and no user information will be attached to the corresponding metadata when the client creates directories or files.


Authentication is enabled when the authentication type is CUSTOM.

Alluxio clients retrieves user information via the class provided by the property. The specified class must implement the interface

This mode is currently experimental and should only be used in tests.


Authentication is enabled and enforced via Kerberos. Kerberos is an authentication protocol that provides strong and mutual authentication between clients and servers.

The typical Kerberos principal format used for services is "primary/instance@REALM.COM". It is required to prepare the Kerberos principals as Alluxio service principal names (SPN).

It is recommended to use hostname-associated instance name of Alluxio service principals, such as: <alluxio-service-name>/<hostname>@REALM.COM. This way each Alluxio server node has a unique service principal. Note that the <hostname> in each principal must match the server (either master or worker) hostname.

On the other hand, Alluxio also supports cluster-wide unified instance name, like <alluxio-service-name>/<alluxio-cluster-name>@REALM.COM, so that all the Alluxio servers share the same principal. To use this feature, please set<alluxio-cluster-name>.

Alluxio Enterprise Edition supports two ways to setup Kerberos authentication:

  1. Java Kerberos
  2. MIT native Kerberos (through JGSS)

Java Kerberos

Please refer to Kerberos security setup instructions to set set up Alluxio with Java Kerberos enabled.

Alluxio system administrators are responsible for specifying the Alluxio servers Kerberos credentials, with principal name and keytab file. Alluxio clients need valid Kerberos credentials (either keytab files or local ticket cache) to access a Kerberos-enabled Alluxio cluster.

When KERBEROS authentication is enabled, the login user for different component is obtained as follows:

  1. For Alluxio servers, the login user is represented by the server-side Kerberos principal in A corresponding keytab file must be specified in The Alluxio user shown in Alluxio namespace is the short name of the Kerberos principal, which excludes the hostname and realm part.

  2. For Alluxio clients, the login user is represented by the client-side Kerberos principal in There are two ways for the Alluxio clients to login via Kerberos. One is specify a keytab file in The other way is to do kinit Kerberos login for the name on the client machine. Alluxio client first checks whether there is a valid If there is no valid keytab file which can login successfully, Alluxio client will fall back to find the login info in the ticket cache. If none of those Kerberos credentials exist, Alluxio client will throw a login failure, and ask the user to provide the keytab file, or login via kinit.

MIT Kerberos

Please refer to MIT Kerberos security setup for detailed instructions to setup Alluxio with MIT native Kerberos.

The default Java GSS implementation relies on JAAS KerberosLoginModule for initial credential acquisition. In contrast, when native platform Kerberos integration is enabled, the initial credential acquisition should happen prior to calling JGSS APIs, e.g. through kinit. When enabled, Java GSS would look for native GSS library using the operating system specific name, e.g. Solaris: vs Linux: If the desired GSS library has a different name or is not located under a directory for system libraries, then its full path should be specified using the system property

Connecting with Secure-HDFS

Note that in Alluxio Enterprise edition the way to configure Alluxio with secure-HDFS is different from that in Alluxio enterprise edition. alluxio.master.keytab.file, alluxio.master.principal, alluxio.worker.keytab.file and alluxio.worker.principal are deprecated. Please use and instead. Note that Alluxio can use a different principal other than Alluxio server principal to access secure HDFS. If is not specified, then Alluxio falls back to using

If you want to setup a Kerberos-enabled Alluxio cluster on top of Kerberos-enabled HDFS, please refer to “Kerberos-enabled Alluxio integration with Secure-HDFS” in Kerberos setup guide or MIT Kerberos security setup for more details.

Auth-to-local configuration

Alluxio supports configurable translation from Kerberos principal name to operating system user, via MIT Kerberos auth_to_local conf. To make it easier to configure together with HDFS, the syntax is the same as Hadoop auth_to_local. Please use the configuration property to set it up. By default the value is DEFAULT.


The Alluxio filesystem implements a permissions model similar to the POSIX permissions model.

Each file and directory is associated with:

  • An owner, which is the user of the client process to create the file or directory.
  • A group, which is the group fetched from user-groups-mapping service. See User group mapping.
  • Permissions, which consist of three parts:
    • Owner permission defines the access privileges of the file owner
    • Group permission defines the access privileges of the owning group
    • Other permission defines the access privileges of all users that are not in any of above two classes

Each permission has three actions:

  1. read (r)
  2. write (w)
  3. execute (x)

For files:

  • Read permissions are required to read files
  • Write permission are required to write files

For directories:

  • Read permissions are required to list its contents
  • Write permissions are required to create, rename, or delete files or directories under it
  • Execute permissions are required to access a child of the directory

The output of the ls shell command when authorization is enabled is:

$ ./bin/alluxio fs ls /
drwxr-xr-x jack           staff                       24       PERSISTED 06-14-2019 07:02:45:248  DIR /default_tests_files
-rw-r--r-- jack           staff                       80   NOT_PERSISTED 06-14-2019 07:02:26:487 100% /default_tests_files/BASIC_CACHE_PROMOTE_MUST_CACHE

User group mapping

For a given user, the list of groups is determined by a group mapping service, configured by the property, with a default implementation of This implementation executes the groups shell command on the local machine to fetch the group memberships of a particular user. Running the groups command for every query may be expensive, so the user group mapping is cached, with an expiration period configured by the property, with a default value of 60s. If set to a value of 0, the caching is disabled. If the cache timeout is too low or disabled, the groups command will be run very frequently, but may increase latency for operations. If the cache timeout is too high, the groups command will not be run frequently, but the cached results may become stale.

Alluxio has super user, a user with special privileges typically needed to administer and maintain the system. The super user is the operating system user executing the Alluxio master process. The property defines a super group. Any additional operating system users belong to this operating system group are also super users.


If your organization use OpenLDAP or Active Directory to manage identities, it is recommended to sync LDAP users and groups to machines’ operating system running Alluxio. Alternatively, Alluxio also supports direct connection to OpenLDAP or Active Directory for group mapping service. To use the Alluxio integration with OpenLDAP or Active Directory, set the following properties in

Property NameDefaultMeaning

Example configuration for an LDAP server without SSL:,dc=example,dc=com,cn=Users,dc=example,dc=com

Example configuration for an LDAP server with SSL:,dc=example,dc=com,cn=Users,dc=example,dc=com

If you have your own way of managing the SSL keystore, configure the properties related to LDAP SSL keystore according to your setup.

Otherwise, here is an example for generate the SSL keystore:

# get the LDAP server's certificate by
$ echo  | openssl s_client -connect 2>/dev/null | openssl x509 > /tmp/ldap.crt

# add the certificate to Java's trusted keystore by
$ sudo keytool -import -noprompt -trustcacerts -alias ldap -file /tmp/ldap.crt -keystore ${JAVA_HOME}/jre/lib/security/cacerts

# generate the keystore in JKS format, in the prompt, specify password as the value for property
# "", answer "yes" to the question of whether to trust the certificate
$ keytool -import -keystore /path/to/ldap.jks -file /tmp/ldap.crt

Initialized directory and file permissions

When a file is created, it is initially assigned fully opened permissions of 666 by default. Similarly, a directory is initially assigned with 777 permissions. A umask is applied on the initial permissions; this is configured by the property, with a default of 022. Without any property modifications, files and directories are created with 644 and 755 permissions respectively.

Update directory and file permission model

The owner, group, and permissions can be changed by two ways:

  1. User application invokes the setAttribute(...) method of FileSystem API or Hadoop API.
  2. CLI command in shell. See chown, chgrp, chmod.

The owner attribute can only be changed by a super user. The group and permission attributes can be changed by a super user or the owner of the path.

Access Control Lists

The POSIX permissions model allows administrators to grant permissions to owners, owning groups and other users. The permission bits model is sufficient for most cases. However, to help administrators express more complicated security policies, Alluxio also supports Access Control Lists (ACLs). ACLs allow administrators to grant permissions to any user or group.

A file or directory’s Access Control List consists of multiple entries. The two types of ACL entries are Access ACL entries and Default ACL entries.

1. Access ACL Entries:

This type of ACL entry specifies a particular user or group’s permission to read, write and execute.

Each ACL entry consists of:

  • a type, which can be one of user, group or mask
  • an optional name
  • a permission string similar to the POSIX permission bits

The following table shows the different types of ACL entries that can appear in the access ACL:

ACL Entry Type Description
user:userid:permission Sets the access ACLs for a user. Empty userid implies the permission is for the owner of the file.
group:groupid:permission Sets the access ACLs for a group. Empty groupid implies the permission is for the owning group of the file.
other::permission Sets the access ACLs for all users not specified above.
mask::permission Sets the effective rights mask. The ACL mask indicates the maximum permissions allowed for all users other than the owner and for groups.

Notice that ACL entries describing owner’s, owning group’s and other’s permissions already exist in the standard POSIX permission bits model. For example, a standard POSIX permission of 755 translates into an ACL list as follows:


These three entries are always present in each file and directory. When there are entries in addition to these standard entries, the ACL is considered an extended ACL.

A mask entry is automatically generated when an ACL becomes extended. Unless specifically set by the user, the mask’s value is adjusted to be the union of all permissions affected by the mask entry. This includes all the user entries other than the owner and all group entries.

For the ACL entry user::rw-:

  • the type is user
  • the name is empty, which implies the owner
  • the permission string is rw-

This culminates to the owner has read and write permissions, but not execute.

For the ACL entry group:interns:rwx and mask mask::r--:

  • the entry grants all permissions to the group interns
  • the mask only allows read permissions

This culminates to the interns group having only read access because the mask disallows all other permissions.

2. Default ACL Entries:

Default ACLs only apply to directories. Any new file or directory created within a directory with a default ACL will inherit the default ACL as its access ACL. Any new directory created within a directory with a default ACL will also inherit the default ACL as its default ACL.

Default ACLs also consists of ACL entries, similar to those found in access ACLs. The are distinguished by the default keyword as the prefix. For example, default:user:alluxiouser:rwx and default:other::r-x are both valid default ACL entries.

Given a documents directory, its default ACL can be set to default:user:alluxiouser:rwx. The user alluxiouser will have full access to any new files created in the documents directory. These new files will have an access ACL entry of user:alluxiouser:rwx. Note that the ACL does not grant the user alluxiouser any additional permissions to the directory.

Managing ACL entries

ACLs can be managed by two ways:

  1. User application invokes the setFacl(...) method of FileSystem API or Hadoop API to change the ACL and invokes the getFacl(...) to obtain the current ACL.
  2. CLI command in shell. See getfacl setfacl,

The ACL of a file or directory can only be changed by super user or its owner.

Data Path Authorization

In Alluxio Enterprise Edition, the access control on data transfer path (Client-Workers) is further enforced by an enhanced distributed authorization mechanism. Alluxio worker is able to check whether the client user has the right privilege to access the requested block, even though workers do not know about the file permission info.

This data path authorization feature is disabled by default. It can be turned on with the following configuration on all masters and workers. When capability feature is enabled, Alluxio master verifies the permission and grants a signed capability to the client. The capability is a token which grants the bearer specified access rights. The capability is verified by Alluxio workers to see whether the granted permission matches with the client’s access request. A capability is only valid for a short amount of time, which can be configured via Alluxio server configuration (default to 1 hour) on all masters and workers.

Capabilities are generated using a scheme where the Master and all Workers share a secret key, called CapabilityKey. Only Master and Workers know the key, no third party can forge the capabilities. The capability key is generated and rotated by Alluxio master periodically. To avoid bulk capability invalidation errors, during key rotation the old key is still valid for a short time period to allow graceful key expiration. The workers will accept old capabilities for a certain time period (by default 25% of the key life time) after receiving a new version of capability key. Capability key life time can be configured via Alluxio server configuration (default to 1 day) on all masters.

Client-Side Hadoop Impersonation

When Alluxio is used in a Hadoop environment, a user, or identity, can be specified for both the Hadoop client and the Alluxio client. Since the Hadoop client user and the Alluxio client user can specified independently, the users could be different from each other. The Hadoop client user may even be in a separate namespace from the Alluxio client user.

Alluxio client-side Hadoop impersonation solves the issues when the Hadoop client user is different from the Alluxio client user. With this feature, the Alluxio client examines the Hadoop client user, and then attempts to impersonate as that Hadoop client user.

For example, a Hadoop application can be configured to run as the Hadoop client user foo, but the Alluxio client user is configured to be yarn. This means any data interactions will be attributed to user yarn. With client-side Hadoop impersonation, the Alluxio client will detect the Hadoop client user is foo, and then connect to Alluxio servers as user yarn impersonating as user foo. With this impersonation, the data interactions will be attributed to user ‘foo’.

This feature is only applicable when using the hadoop compatible client to access Alluxio.

In order to configure Alluxio for client-side Hadoop impersonation, both client and server configurations (master and worker) are required.

Server Configuration

To enable a particular Alluxio client user to impersonate other users server (master and worker) configuration are required. Set the<USERNAME>.users property, where <USERNAME> is the name of the Alluxio client user.

The property value is a comma-separated list of users that <USERNAME> is allowed to impersonate. The wildcard value * can be used to indicate the user can impersonate any other user. Some examples:

    • the Alluxio client user alluxio_user is allowed to impersonate user1 and user2
    • the Alluxio client user client is allowed to impersonate any user

To enable a particular user to impersonate other groups, set the<USERNAME>.groups property, where again <USERNAME> is the name of the Alluxio client user. Similar to above, the value is a comma-separated list of groups and the wildcard value * can be used to indicate all groups. Some examples:

    • the Alluxio client user alluxio_user is allowed to impersonate any users from groups group1 and group2
    • the Alluxio client user client is allowed to impersonate users from any group

In summary, to enable an Alluxio client user to impersonate other users, at least one of the two impersonation properties must be set on servers; setting both are allowed for the same Alluxio client user.

Client Configuration

After enabling impersonation on the servers for a given Alluxio client user, the client must indicate which user it wants to impersonate. This is configured by the property.

If the property is set to an empty string or _NONE_, impersonation is disabled, and the Alluxio client will interact with Alluxio servers as the Alluxio client user. If the property is set to _HDFS_USER_, the Alluxio client will connect to Alluxio servers as the Alluxio client user, but impersonate as the Hadoop client user when using the Hadoop compatible client.


The most common impersonation error applications may see is something like User yarn is not configured for any impersonation. impersonationUser: foo. This is most likely due to the fact that the Alluxio servers have not been configured to enable impersonation for that user. To fix this, the Alluxio servers must be configured to enable impersonation for the user in question (yarn in the example error message).

Please read this blog post for more tips.


Alluxio supports audit logging to allow system administrators to track users’ access to file metadata.

The audit log file at master_audit.log contains entries corresponding to file metadata access operations. The format of Alluxio audit log entry is shown in the table below:

succeeded True if the command has succeeded. To succeed, it must also have been allowed.
allowed True if the command has been allowed. Note that a command can still fail even if it has been allowed.
ugi User group information, including username, primary group, and authentication type.
ip Client IP address.
cmd Command issued by the user.
src Path of the source file or directory.
dst Path of the destination file or directory. If not applicable, the value is null.
perm User:group:mask or null if not applicable.

This is similar to the format of HDFS audit log wiki.

To enable Alluxio audit logging, set the JVM property alluxio.master.audit.logging.enabled to true in See Configuration settings.


Alluxio supports encryption of the network communication between services with TLS, and supports end-to-end data encryption.

TLS Encryption for Network Communication

For Alluxio network communication (rpcs, data transfers), Alluxio supports TLS encryption. In order to configure Alluxio to use TLS encryption, keystores and truststores must be created for Alluxio. A keystore is used by the server side of the TLS connection, and the truststore is used by the client side of the TLS connection.


Alluxio servers (masters and workers) require a keystore in order to enable TLS. The keystore typically stores the key and certificate for the server. This keystore file must be readable by the OS user which launches the Alluxio server processes.

An example, self-signed keystore can be created like:

$ keytool -genkeypair -alias key -keyalg RSA -keysize 2048 -dname "cn=localhost, ou=Department, o=Company, l=City, st=State, c=US" -keystore /alluxio/keystore.jks -keypass keypass -storepass storepass

This will generate a keystore file to /alluxio/keystore.jks, with a key password of keypass and the keystore password as storepass.


All clients of a TLS connection must have access to a truststore to trust all the certificates of the servers. Clients include Alluxio clients, as well as Alluxio workers (since Alluxio workers create client connections to the Alluxio master). The truststore stores the trusted certificates, and must be readable by the process initiating the client connection (clients, workers).

An example truststore (based on the previous keystore) can be created like:

$ keytool -export -alias key -keystore /alluxio/keystore.jks -storepass storepass -rfc -file selfsigned.cer
$ keytool -import -alias key -noprompt -file selfsigned.cer -keystore /alluxio/truststore.jks -storepass trustpass

The first command extracts the certificate from the previously created keystore (using the keystore password storepass). Then, the second command creates a truststore file using that extracted certificate, and saves the truststore to /alluxio/truststore.jks, with a truststore pasword of trustpass.

Configuring Alluxio servers and clients

Once the keystores and truststores are created for all the machines involved, Alluxio needs to be configured to understand how to access those files.

On Alluxio servers (masters and workers), you must add these properties to

# enables TLS
# keystore properties for the server side of connections
# truststore properties for the client side of connections (worker to master)

Once the servers are configured, additional Alluxio clients need to be configured with the client side properties:

# enables TLS
# truststore properties for the client side of connections (worker to master)

Setting these configuration properties will be dependent on the specific application or computation framework you are using.

Once the servers and clients are configured, all network communication will be encrypted with TLS.

End-to-End Data Encryption

Alluxio Enterprise Edition supports transparent end-to-end data encryption. Once configured, data will be written to the Alluxio cluster encrypted and be read decrypted, without requiring changes in application code. Data can only be encrypted and decrypted by the client, and the client is responsible to get the crypto keys from an external Key Management Service (KMS). Alluxio servers and admins are not able to access the crypto keys, thus can not make sense of the data stored in Alluxio servers and under storage.

Alluxio end-to-end encryption achieves both data at-rest encryption and data in-flight encryption. The end-to-end encryption is not applicable to Alluxio metadata.

Alluxio data encryption brings the following benefits:

  1. Data Protection at Rest: malicious users can not make sense of the encrypted data residing on RAM/SSD/HDD.
  2. Security on All Under Storages: applications can use various under storages via Alluxio, with no worry about under storage encryption.
  3. Secure Transmit over Network: eavesdroppers can not make sense of the encrypted data in flight.
  4. Data Integrity: if encrypted data is manipulated, users can easily notice that it has been tampered with.

Here is an overview of a file write and read with encryption:

Alluxio is the single access point to all the encrypted data. Alluxio client is responsible for encrypting the data, and writes the encrypted data to an Alluxio worker, which optionally (depending on the write type) writes the encrypted data to UFS. When Alluxio client reads the data from an Alluxio worker, it decrypts the data and then serves the decrypted data back to the application.

Key Management Service

Alluxio assumes there is an external Key Management Service (KMS) in the enterprise organization, and integrates with it. Alluxio Enterprise Edition supports Hadoop KMS out-of-box. Alluxio is not responsible for encryption key maintenance and lifecycle management. Data is considered as deleted if the secret key is deleted.

Please contact the Alluxio team for customized integration with other KMS types.

Hadoop KMS

An example configuration in for using Hadoop KMS is below:

If the Hadoop KMS is SSL enabled, import the kMS’s certificate to your Java truststore by

$ openssl s_client -showcerts -connect host:port </dev/null 2>/dev/null | openssl x509 -outform PEM > /tmp/cert
$ keytool -import -noprompt -trustcacerts -alias localhost -file /tmp/cert -keystore ${JAVA_HOME}/jre/lib/security/cacerts

and set the to something like kms://https@localhost:16000/kms.

If the Hadoop KMS is Kerberos enabled, you need to specify the Kerberos principal and keytab file for authenticating to the KMS by<kms client principal><path to the kms client principal’s keytab file>

Cipher type and mode

By default, Alluxio Enterprise Edition uses the Advanced Encryption Standard (AES) algorithm in Galois/Counter Mode (GCM), known as AES-GCM. Alluxio supports both 128-bit and 256-bit secret keys, which is determined by the secret key length provided by the KMS. Alluxio uses symmetric encryption, where the same secret key is used to perform both encryption and decryption.

AES-GCM is an authenticated encryption algorithm designed to provide both data authenticity (integrity) and confidentiality. Authentication tags are produced during GCM encryption and must be supplied during decryption. Alluxio stores the authentication tags within the ciphertext data and records some encryption metadata (such as encryption id and encryption layout) in the footer of the encrypted file. Therefore, the encrypted file size will be slightly larger than the plaintext size. Alluxio clients are not aware of this space overhead because the logical plaintext sizes are always shown to the Alluxio clients. Alluxio administrators will see slightly bigger files and blocks on Alluxio server side.


By default, encryption is disabled in Alluxio Enterprise Edition.

To enable encryption in an Alluxio cluster, simply add the following configuration properties.

It is highly recommended to use the integration with OpenSSL crypto library for better encryption performance. The Alluxio Enterprise Edition comes with the pre-compiled JNI library that connects with the OpenSSL library. OpenSSL libcrypto is a prerequisite to use Alluxio encryption with OpenSSL enabled. On unix servers, installing openssl or openssl-devel will install the required OpenSSL packages. The Alluxio Enterprise Edition’s pre-compiled JNI library is compiled with OpenSSL 1.1. If you need to integrate with other versions of OpenSSL, please contact us.

Note that the pre-compiled JNI library is located at ${ALLUXIO_HOME}/lib/native/. If you encountered the following error indicating the required .so not found, please make sure the ${ALLUXIO_HOME} or alluxio.home is set properly.

java.lang.NoClassDefFoundError: Could not initialize class

Caused by: java.lang.UnsatisfiedLinkError: Can't load library: /opt/alluxio/lib/native/

For testing purpose, Alluxio provides a dummy KMS which provides a fixed testing-only encryption key. In order to connect Alluxio with a Key Management Service, please set the and Please refer to the example setup page for Alluxio with Hadoop KMS.

Encryption with UFS

Data is encrypted and decrypted in Alluxio clients, so only ciphertext will be persisted to the under storage systems. If the under storage system supports encryption, it is recommended to disable UFS encryption to avoid unnecessary double encryption. It is also recommended to setup an encrypted Alluxio cluster without any pre-existing unencrypted files, because files in one Alluxio cluster should be either all encrypted or all non-encrypted. In other words, the under storage system is mounted to Alluxio as an empty UFS, and all UFS I/O happens through Alluxio.

Alluxio encryption supports re-mounting the UFS. Data encrypted by Alluxio and persisted in UFS can be read and decrypted when UFS is unmounted and mounted back or across Alluxio restarts.


It is required to start Alluxio master and workers using the same operating system user. In the case where there is a user mismatch, secondary master healthcheck, the command all and certain file operations may fail because of permission checks.