Delegation tokens

Delegation Tokens #

This document aims to explain and demystify delegation tokens as they are used by Flink. Before we into the details here is the high level architecture diagram:

High Level Architecture Diagram

What Are Delegation Tokens and Why Use Them? #

Delegation tokens (DTs from now on) are authentication tokens used by some services to replace long-lived credentials. Many services in the Hadoop ecosystem have support for DTs, since they have some very desirable advantages over long-lived credentials:

No need to distribute long-lived credentials #

In a distributed application, distributing long-lived credentials is tricky. Additionally, not all users wants to distribute them over the network as part of application data as it is an additional attack surface for malicious actors.

DTs allow for a single place e.g. the JobManager (JM from now on) to require long-lived credentials. That entity can then distribute the DTs to other parts of the distributed application e.g. TaskManagers (TM from now on), so they can authenticate to services.

A single token per service is used for authentication #

If Kerberos authentication were used, each client connection to a server would require a trip to the Key Distribution Center (KDC) and generation of a service ticket. In a distributed system, the number of service tickets can balloon pretty quickly due in proportion to the number of client processes (e.g. TMs) times the number of service processes (e.g. HDFS DataNodes). That generates unnecessary extra load on the KDC, and may even run into usage limits set up by the KDC admin.

Delegation tokens are only used for authentication #

DTs, unlike long-lived credentials, can only be used to authenticate to the specific service for which they were issued. You cannot use an existing DT to create new DTs or to create DTs for a different service.

So in short, DTs are not long-lived credentials. They are used by many services to replace Kerberos authentication, or even other forms of authentication, although there is nothing (aside from maybe implementation details) that ties them to the authentication mechanism.

Lifecycle of Delegation Tokens #

DTs, unlike some long-lived credentials, are service-specific. There is no centralized location you contact to create a DT for a service. So, the first step needed to get a DT is being able to authenticate to the service in question. In the Hadoop ecosystem, that is generally done using Kerberos.

This requires long-lived credentials to be available somewhere for the application to use. The user is generally responsible for providing those credentials, which is most commonly done by logging in to the KDC (e.g. using kinit). That generates a “credential cache” containing a ticket granting ticket (TGT), which can then be used to request service tickets.

There are other ways of obtaining TGTs, but, ultimately, one needs a TGT to bootstrap the process.

Once a TGT is available, the target service’s client library can then be used to authenticate to the service and request the creation of a delegation token. This token can now be sent to other processes and used to authenticate to different daemons belonging to that service. And thus the first drawback of DTs becomes apparent: you need service-specific logic to create and use them.

Flink implements a (somewhat) pluggable, internal DT creation API. Support for new services can be added by implementing a DelegationTokenProvider that is then called by the delegation token manager when generating delegation tokens for an application.

Once they are created, the semantics of how DTs operate are also service-specific. But, in general, they try to follow the semantics of Kerberos tokens:

  • A “renewable period” (equivalent to TGT’s “lifetime”) stands for the DT’s validity length before it requires renewal.
  • A “max lifetime” (equivalent to TGT’s “renewable life”) stands for the time until the DT can be renewed.

Once the token reaches its “max lifetime”, a new one needs to be created by contacting the appropriate service, restarting the above process.

Delegation Token Renewal and Renewers #

This is the most confusing part of DT handling, and part of it is because much of the system was designed with Apache Hadoop YARN in mind even though it extends to other services and mechanism.

As seen above, DTs need to be renewed periodically until they finally expire for good. An example of this is the default configuration of HDFS services: delegation tokens are valid for up to 7 days, and need to be renewed every 24 hours. If 24 hours pass without the token being renewed, the token cannot be used anymore. And the token cannot be renewed anymore after 7 days.

This raises the question: who renews tokens? And for a long time the answer was YARN.

When YARN applications are submitted, a set of DTs is also submitted with them. YARN takes care of distributing these tokens to containers (using conventions set by the UserGroupInformation API) and, also, keeping them renewed while the app is running. These tokens are used not just by the application, they are also used by YARN itself to implement features like log collection and aggregation.

But this has a few caveats.

Who renews the tokens? #

This is handled mostly transparently by the Hadoop libraries in the case of YARN. Some services have the concept of a token “renewer”. This “renewer” is the name of the service principal that is allowed to renew the DT. When submitting to YARN, that will be the principal that the YARN service is running as, which means that the client application needs to know that information.

For other resource managers, the renewer mostly does not matter, since there is no service that is doing the renewal.

Which tokens are renewed? #

This is probably the biggest caveat.

As discussed in the previous section, DTs are service-specific, and require service-specific libraries for creation and renewal. This means that for YARN to be able to renew application tokens, YARN needs:

  • The client libraries for all the services the application is using
  • Information about how to connect to the services the application is using
  • Permissions to connect to those services

In reality, though, most of the time YARN has access to a single HDFS cluster, and that will be the extent of its DT renewal features. Any other tokens sent to YARN will be distributed to containers, but will not be renewed.

This means that those tokens will expire way before their max lifetime, unless some other code takes care of renewing them.

Also, not all client libraries even implement token renewal. To use the example of a service supported by Flink, the renew() method of HBase tokens is a no-op. So the only way to “renew” an HBase token is to create a new one.

What happens when tokens expire for good? #

The final caveat is that DTs have a maximum life, regardless of renewal. And after that deadline is met, you need to create new tokens to be able to connect to the services. That means you need the ability to connect to the service without a delegation token, which requires some form of authentication aside from DTs.

This is especially important for long-running applications that run unsupervised. They must be able to keep on going without having someone logging into a terminal and typing a password every few days.

Delegation Token Renewal #

Because of the issues explained above, Flink implements a different way of doing renewal. The solution is a compromise: it targets the lowest common denominator, which is services like HBase that do not support actual token renewal. In the following examples we often drive the discussion via highlighting the Hadoop ecosystem components to cover all our bases, because they tend to be more complex from an authentication perspective compared to others e.g. AWS S3.

In Flink, DT “renewal” is enabled by giving the application long-lived credentials (e.g. keytab). A keytab is equivalent to your Kerberos password written into a plain text file, which is why it is so sensitive: if anyone is able to get hold of that keytab file, they can authenticate to any service as that user as long as the credentials stored in the keytab remain valid in the KDC.

By having the keytab, Flink can indefinitely maintain a valid Kerberos TGT.

With long-lived credentials available, Flink will create new DTs for the configured services as old ones expire. So Flink doesn’t renew tokens as explained in the previous section: it will create new tokens at every renewal interval instead, and distribute those tokens to TMs.

This also has another advantage on top of supporting services like HBase: it removes the dependency on an external renewal service (like YARN). That way, Flink’s renewal feature can be used with resource managers that are not DT-aware, such as Kubernetes, as long as the application has long-lived credentials.

Delegation Tokens and Proxy Users #

“Proxy users” is Hadoop-speak for impersonation. It allows user A to impersonate user B when connecting to a service, if that service allows it.

Flink simply doesn’t allow impersonation when submitting applications. Spark supports impersonation but doesn’t allow token renewal. Since Flink is mainly designed for streaming workloads there would be not much gain to add this feature.

Externally Generated Delegation Tokens #

Flink uses the UserGroupInformation (UGI) API to manage the Hadoop credentials. That means that Flink inherits the feature of loading DTs automatically from a file. The Hadoop classes will load the token cache pointed at by the HADOOP_TOKEN_FILE_LOCATION environment variable, when it’s defined.

This feature is mostly used by services that start workloads on behalf of users. Regular users rarely use this feature, given it would require them to figure out how to get those tokens outside Flink.

Note that Flink itself can obtain DTs. In case UGI contains a DT for a specific service and Flink is configured to obtain tokens for that service then the token first will be loaded and then will be overwritten by the loading mechanism in Flink.

Limitations of Delegation Token Support #

There are certain limitations to bear in mind when talking about DTs.

  • Not all DTs actually expose their renewal period. This is a service configuration that is not generally exposed to clients. For this reason, certain DT providers cannot provide a renewal period, thus requiring that the service’s configuration is in some way synchronized with another service that does provide that information.
    The HDFS service, which is generally available when DTs are needed in the first place, provides this information, so in general it’s a good idea for all services using DTs to use the same configuration as HDFS for the renewal period.

  • Flink is not parsing the user application code, so it doesn’t know which delegation tokens will be needed. This means that Flink will try to get as many delegation tokens as is possible based on the configuration available. That means that if an HBase token provider is enabled but the app doesn’t actually use HBase, a DT will still be generated. The user would have to explicitly disable the mentioned provider in that case.

  • It is challenging to create DTs “on demand”. Flink obtains/distributes tokens upfront and re-obtains/re-distributes them periodically.
    The advantage, though, is that user code does not need to worry about DTs, since Flink will handle them transparently when the proper configuration is available.

  • There are external file system plugins which are authenticating to the same service. One good example is s3-hadoop and s3-presto. They both authenticate to S3. They’re having different service names but obtaining tokens for the same service which might cause unintended consequences. Since they’re obtaining tokens for the same service they store these tokens at the same place. It’s easy to see that if they’re used together with the same credentials then there will be no issues since the tokens are going to be overwritten by each other in a single-threaded way (which belongs to a single user). However, if the plugins are configured with different user credentials then the token which will be used for data processing can belong to any of the users which is non-deterministic.