Apache Storm offers a range of configuration options when trying to secure your cluster. By default all authentication and authorization is disabled but can be turned on as needed.
You can still have a secure storm cluster without turning on formal Authentication and Authorization. But to do so usually requires configuring your Operating System to restrict the operations that can be done. This is generally a good idea even if you plan on running your cluster with Auth.
The exact detail of how to setup these precautions varies a lot and is beyond the scope of this document.
It is generally a good idea to enable a firewall and restrict incoming network connections to only those originating from the cluster itself and from trusted hosts and services, a complete list of ports storm uses are below.
If the data your cluster is processing is sensitive it might be best to setup IPsec to encrypt all traffic being sent between the hosts in the cluster.
| Default Port | Storm Config | Client Hosts/Processes | Server |
|---|---|---|---|
| 2181 | storm.zookeeper.port |
Nimbus, Supervisors, and Worker processes | Zookeeper |
| 6627 | nimbus.thrift.port |
Storm clients, Supervisors, and UI | Nimbus |
| 8080 | ui.port |
Client Web Browsers | UI |
| 8000 | logviewer.port |
Client Web Browsers | Logviewer |
| 3772 | drpc.port |
External DRPC Clients | DRPC |
| 3773 | drpc.invocations.port |
Worker Processes | DRPC |
| 670{0,1,2,3} | supervisor.slots.ports |
Worker Processes | Worker Processes |
The UI and logviewer processes provide a way to not only see what a cluster is doing, but also manipulate running topologies. In general these processes should not be exposed except to users of the cluster.
Some form of Authentication is typically required, with using java servlet filters
ui.filter: "filter.class"
ui.filter.params: "param1":"value1"or by restricting the UI/log viewers ports to only accept connections from local hosts, and then front them with another web server, like Apache httpd, that can authenticate/authorize incoming connections and proxy the connection to the storm process. To make this work the ui process must have logviewer.port set to the port of the proxy in its storm.yaml, while the logviewers must have it set to the actual port that they are going to bind to.
The servlet filters are preferred because it allows individual topologies to specificy who is and who is not allowed to access the pages associated with them.
Storm UI can be configured to use AuthenticationFilter from hadoop-auth.
ui.filter: "org.apache.hadoop.security.authentication.server.AuthenticationFilter"
ui.filter.params:
"type": "kerberos"
"kerberos.principal": "HTTP/nimbus.witzend.com"
"kerberos.keytab": "/vagrant/keytabs/http.keytab"
"kerberos.name.rules": "RULE:[2:$1@$0]([jt]t@.*EXAMPLE.COM)s/.*/$MAPRED_USER/ RULE:[2:$1@$0]([nd]n@.*EXAMPLE.COM)s/.*/$HDFS_USER/DEFAULT"make sure to create a principal 'HTTP/{hostname}' (here hostname should be the one where UI daemon runs
Once configured users needs to do kinit before accessing UI. Ex: curl -i --negotiate -u:anyUser -b ~/cookiejar.txt -c ~/cookiejar.txt http://storm-ui-hostname:8080/api/v1/cluster/summary
- Firefox: Goto about:config and search for network.negotiate-auth.trusted-uris double-click to add value "http://storm-ui-hostname:8080"
- Google-chrome: start from command line with: google-chrome --auth-server-whitelist="*storm-ui-hostname" --auth-negotiate-delegate-whitelist="*storm-ui-hostname"
- IE: Configure trusted websites to include "storm-ui-hostname" and allow negotiation for that website
Caution: In AD MIT Keberos setup the key size is bigger than the default UI jetty server request header size. Make sure you set ui.header.buffer.bytes to 65536 in storm.yaml. More details are on STORM-633
Storm offers pluggable authentication support through thrift and SASL. This example only goes off of Kerberos as it is a common setup for most big data projects.
Setting up a KDC and configuring kerberos on each node is beyond the scope of this document and it is assumed that you have done that already.
Each Zookeeper Server, Nimbus, and DRPC server will need a service principal, which, by convention, includes the FQDN of the host it will run on. Be aware that the zookeeper user MUST be zookeeper.
The supervisors and UI also need a principal to run as, but because they are outgoing connections they do not need to be service principals.
The following is an example of how to setup kerberos principals, but the
details may vary depending on your KDC and OS.
# Zookeeper (Will need one of these for each box in teh Zk ensamble)
sudo kadmin.local -q 'addprinc zookeeper/[email protected]'
sudo kadmin.local -q "ktadd -k /tmp/zk.keytab zookeeper/[email protected]"
# Nimbus and DRPC
sudo kadmin.local -q 'addprinc storm/[email protected]'
sudo kadmin.local -q "ktadd -k /tmp/storm.keytab storm/[email protected]"
# All UI logviewer and Supervisors
sudo kadmin.local -q 'addprinc [email protected]'
sudo kadmin.local -q "ktadd -k /tmp/storm.keytab [email protected]"be sure to distribute the keytab(s) to the appropriate boxes and set the FS permissions so that only the headless user running ZK, or storm has access to them.
Both storm and Zookeeper use jaas configuration files to log the user in. Each jaas file may have multiple sections for different interfaces being used.
To enable Kerberos authentication in storm you need to set the following storm.yaml configs
storm.thrift.transport: "backtype.storm.security.auth.kerberos.KerberosSaslTransportPlugin"
java.security.auth.login.config: "/path/to/jaas.conf"Nimbus and the supervisor processes will also connect to ZooKeeper(ZK) and we want to configure them to use Kerberos for authentication with ZK. To do this append
-Djava.security.auth.login.config=/path/to/jaas.conf
to the childopts of nimbus, ui, and supervisor. Here is an example given the default childopts settings at the time of writing:
nimbus.childopts: "-Xmx1024m -Djava.security.auth.login.config=/path/to/jaas.conf"
ui.childopts: "-Xmx768m -Djava.security.auth.login.config=/path/to/jaas.conf"
supervisor.childopts: "-Xmx256m -Djava.security.auth.login.config=/path/to/jaas.conf"The jaas.conf file should look something like the following for the storm nodes. The StormServer section is used by nimbus and the DRPC Nodes. It does not need to be included on supervisor nodes. The StormClient section is used by all storm clients that want to talk to nimbus, including the ui, logviewer, and supervisor. We will use this section on the gateways as well but the structure of that will be a bit different. The Client section is used by processes wanting to talk to zookeeper and really only needs to be included with nimbus and the supervisors. The Server section is used by the zookeeper servers. Having unused sections in the jaas is not a problem.
StormServer {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="$keytab"
storeKey=true
useTicketCache=false
principal="$principal";
};
StormClient {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="$keytab"
storeKey=true
useTicketCache=false
serviceName="$nimbus_user"
principal="$principal";
};
Client {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="$keytab"
storeKey=true
useTicketCache=false
serviceName="zookeeper"
principal="$principal";
};
Server {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="$keytab"
storeKey=true
useTicketCache=false
principal="$principal";
};
The following is an example based off of the keytabs generated
StormServer {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="/keytabs/storm.keytab"
storeKey=true
useTicketCache=false
principal="storm/[email protected]";
};
StormClient {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="/keytabs/storm.keytab"
storeKey=true
useTicketCache=false
serviceName="storm"
principal="[email protected]";
};
Client {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="/keytabs/storm.keytab"
storeKey=true
useTicketCache=false
serviceName="zookeeper"
principal="[email protected]";
};
Server {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="/keytabs/zk.keytab"
storeKey=true
useTicketCache=false
serviceName="zookeeper"
principal="zookeeper/[email protected]";
};
Nimbus also will translate the principal into a local user name, so that other services can use this name. To configure this for Kerberos authentication set
storm.principal.tolocal: "backtype.storm.security.auth.KerberosPrincipalToLocal"
This only needs to be done on nimbus, but it will not hurt on any node. We also need to inform the topology who the supervisor daemon and the nimbus daemon are running as from a ZooKeeper perspective.
storm.zookeeper.superACL: "sasl:${nimbus-user}"
Here nimbus-user is the Kerberos user that nimbus uses to authenticate with ZooKeeper. If ZooKeeeper is stripping host and realm then this needs to have host and realm stripped too.
Complete details of how to setup a secure ZK are beyond the scope of this document. But in general you want to enable SASL authentication on each server, and optionally strip off host and realm
authProvider.1 = org.apache.zookeeper.server.auth.SASLAuthenticationProvider
kerberos.removeHostFromPrincipal = true
kerberos.removeRealmFromPrincipal = true
And you want to include the jaas.conf on the command line when launching the server so it can use it can find the keytab.
-Djava.security.auth.login.config=/jaas/zk_jaas.conf
Ideally the end user will only need to run kinit before interacting with storm. To make this happen seamlessly we need the default jaas.conf on the gateways to be something like
StormClient {
com.sun.security.auth.module.Krb5LoginModule required
doNotPrompt=false
useTicketCache=true
serviceName="$nimbus_user";
};
The end user can override this if they have a headless user that has a keytab.
Authentication does the job of verifying who the user is, but we also need authorization to do the job of enforcing what each user can do.
The preferred authorization plug-in for nimbus is The SimpleACLAuthorizer. To use the SimpleACLAuthorizer, set the following:
nimbus.authorizer: "backtype.storm.security.auth.authorizer.SimpleACLAuthorizer"DRPC has a separate authorizer configuration for it. Do not use SimpleACLAuthorizer for DRPC.
The SimpleACLAuthorizer plug-in needs to know who the supervisor users are, and it needs to know about all of the administrator users, including the user running the ui daemon.
These are set through nimbus.supervisor.users and nimbus.admins respectively. Each can either be a full Kerberos principal name, or the name of the user with host and realm stripped off.
The Log servers have their own authorization configurations. These are set through logs.users and logs.groups. These should be set to the admin users or groups for all of the nodes in the cluster.
When a topology is submitted, the submitting user can specify users in this list as well. The users and groups specified-in addition to the users in the cluster-wide setting-will be granted access to the submitted topology's worker logs in the logviewers.
To ensure isolation of users in multi-tenancy, there is need to run supervisors and headless user and group unique to execution on the supervisor nodes. To enable this follow below steps.
- Add headlessuser to all supervisor hosts.
- Create unique group and make it the primary group for the headless user on the supervisor nodes.
- The set following properties on storm for these supervisor nodes.
To support multi-tenancy better we have written a new scheduler. To enable this scheduler set.
storm.scheduler: "backtype.storm.scheduler.multitenant.MultitenantScheduler"Be aware that many of the features of this scheduler rely on storm authentication. Without them the scheduler will not know what the user is and will not isolate topologies properly.
The goal of the multi-tenant scheduler is to provide a way to isolate topologies from one another, but to also limit the resources that an individual user can have in the cluster.
The scheduler currently has one config that can be set either through =storm.yaml= or through a separate config file called =multitenant-scheduler.yaml= that should be placed in the same directory as =storm.yaml=. It is preferable to use =multitenant-scheduler.yaml= because it can be updated without needing to restart nimbus.
There is currently only one config in =multitenant-scheduler.yaml=, =multitenant.scheduler.user.pools= is a map from the user name, to the maximum number of nodes that user is guaranteed to be able to use for their topologies.
For example:
multitenant.scheduler.user.pools:
"evans": 10
"derek": 10By default storm runs workers as the user that is running the supervisor. This is not ideal for security. To make storm run the topologies as the user that launched them set.
supervisor.run.worker.as.user: trueThere are several files that go along with this that are needed to be configured properly to make storm secure.
The worker-launcher executable is a special program that allows the supervisor to launch workers as different users. For this to work it needs to be owned by root, but with the group set to be a group that only teh supervisor headless user is a part of. It also needs to have 6550 permissions. There is also a worker-launcher.cfg file, usually located under /etc/ that should look something like the following
storm.worker-launcher.group=$(worker_launcher_group)
min.user.id=$(min_user_id)
where worker_launcher_group is the same group the supervisor is a part of, and min.user.id is set to the first real user id on the system. This config file also needs to be owned by root and not have world or group write permissions.
Individual topologies have the ability to push credentials (tickets and tokens) to workers so that they can access secure services. Exposing this to all of the users can be a pain for them.
To hide this from them in the common case plugins can be used to populate the credentials, unpack them on the other side into a java Subject, and also allow Nimbus to renew the credentials if needed.
These are controlled by the following configs. topology.auto-credentials is a list of java plugins, all of which must implement IAutoCredentials interface, that populate the credentials on gateway
and unpack them on the worker side. On a kerberos secure cluster they should be set by default to point to backtype.storm.security.auth.kerberos.AutoTGT.
nimbus.credential.renewers.classes should also be set to this value so that nimbus can periodically renew the TGT on behalf of the user.
nimbus.credential.renewers.freq.secs controls how often the renewer will poll to see if anything needs to be renewed, but the default should be fine.
In addition Nimbus itself can be used to get credentials on behalf of the user submitting topologies. This can be configures using nimbus.autocredential.plugins.classes which is a list of fully qualified class names ,all of which must implement INimbusCredentialPlugin. Nimbus will invoke the populateCredentials method of all the configured implementation as part of topology submission. You should use this config with topology.auto-credentials and nimbus.credential.renewers.classes so the credentials can be populated on worker side and nimbus can automatically renew them. Currently there are 2 examples of using this config, AutoHDFS and AutoHBase which auto populates hdfs and hbase delegation tokens for topology submitter so they don't have to distribute keytabs on all possible worker hosts.
By default storm allows any sized topology to be submitted. But ZK and others have limitations on how big a topology can actually be. The following configs allow you to limit the maximum size a topology can be.
| YAML Setting | Description |
|---|---|
| nimbus.slots.perTopology | The maximum number of slots/workers a topology can use. |
| nimbus.executors.perTopology | The maximum number of executors/threads a topology can use. |
The Logviewer daemon now is also responsible for cleaning up old log files for dead topologies.
| YAML Setting | Description |
|---|---|
| logviewer.cleanup.age.mins | How old (by last modification time) must a worker's log be before that log is considered for clean-up. (Living workers' logs are never cleaned up by the logviewer: Their logs are rolled via logback.) |
| logviewer.cleanup.interval.secs | Interval of time in seconds that the logviewer cleans up worker logs. |
Hopefully more on this soon