Sahara Advanced Configuration Guide

This guide addresses specific aspects of Sahara configuration that pertain to advanced usage. It is divided into sections about various features that can be utilized, and their related configurations.

Custom network topologies

Sahara accesses instances at several stages of cluster spawning through SSH and HTTP. Floating IPs and network namespaces will be automatically used for access when present. When floating IPs are not assigned to instances and namespaces are not being used, sahara will need an alternative method to reach them.

The proxy_command parameter of the configuration file can be used to give sahara a command to access instances. This command is run on the sahara host and must open a netcat socket to the instance destination port. The {host} and {port} keywords should be used to describe the destination, they will be substituted at runtime. Other keywords that can be used are: {tenant_id}, {network_id} and {router_id}.

Additionally, if proxy_command_use_internal_ip is set to True, then the internal IP will be substituted for {host} in the command. Otherwise (if False, by default) the management IP will be used: this corresponds to floating IP if present in the relevant node group, else the internal IP. The option is ignored if proxy_command is not also set.

For example, the following parameter in the sahara configuration file would be used if instances are accessed through a relay machine:

[DEFAULT]
proxy_command='ssh relay-machine-{tenant_id} nc {host} {port}'

Whereas the following shows an example of accessing instances though a custom network namespace:

[DEFAULT]
proxy_command='ip netns exec ns_for_{network_id} nc {host} {port}'

DNS Hostname Resolution

Sahara can resolve hostnames of cluster instances by using DNS. For this Sahara uses Designate. With this feature, for each instance of the cluster Sahara will create two A records (for internal and external ips) under one hostname and one PTR record. Also all links in the Sahara dashboard will be displayed as hostnames instead of just ip addresses.

You should configure DNS server with Designate. Designate service should be properly installed and registered in Keystone catalog. The detailed instructions about Designate configuration can be found here: Designate manual installation and here: Configuring OpenStack Networking with Designate. Also if you use devstack you can just enable the Designate devstack plugin.

When Designate is configured you should create domain(s) for hostname resolution. This can be done by using the Designate dashboard or by CLI. Also you have to create in-addr.arpa. domain for reverse hostname resolution because some plugins (e.g. HDP) determine hostname by ip.

Sahara also should be properly configured. In sahara.conf you must specify two config properties:

[DEFAULT]
# Use Designate for internal and external hostnames resolution:
use_designate=true
# IP addresses of Designate nameservers:
nameservers=1.1.1.1,2.2.2.2

An OpenStack operator should properly configure the network. It must enable DHCP and specify DNS server ip addresses (e.g. 1.1.1.1 and 2.2.2.2) in DNS Name Servers field in the Subnet Details. If the subnet already exists and changing it or creating new one is impossible then Sahara will manually change /etc/resolv.conf file on every instance of the cluster (if nameservers list has been specified in sahara.conf). In this case, though, Sahara cannot guarantee that these changes will not be overwritten by DHCP or other services of the existing network. Sahara has a health check for track this situation (and if it occurs the health status will be red).

In order to resolve hostnames from your local machine you should properly change your /etc/resolv.conf file by adding appropriate ip addresses of DNS servers (e.g. 1.1.1.1 and 2.2.2.2). Also the VMs with DNS servers should be available from your local machine.

Data-locality configuration

Hadoop provides the data-locality feature to enable task tracker and data nodes the capability of spawning on the same rack, Compute node, or virtual machine. Sahara exposes this functionality to the user through a few configuration parameters and user defined topology files.

To enable data-locality, set the enable_data_locality parameter to true in the sahara configuration file

[DEFAULT]
enable_data_locality=true

With data locality enabled, you must now specify the topology files for the Compute and Object Storage services. These files are specified in the sahara configuration file as follows:

[DEFAULT]
compute_topology_file=/etc/sahara/compute.topology
swift_topology_file=/etc/sahara/swift.topology

The compute_topology_file should contain mappings between Compute nodes and racks in the following format:

compute1 /rack1
compute2 /rack2
compute3 /rack2

Note that the Compute node names must be exactly the same as configured in OpenStack (host column in admin list for instances).

The swift_topology_file should contain mappings between Object Storage nodes and racks in the following format:

node1 /rack1
node2 /rack2
node3 /rack2

Note that the Object Storage node names must be exactly the same as configured in the object ring. Also, you should ensure that instances with the task tracker process have direct access to the Object Storage nodes.

Hadoop versions after 1.2.0 support four-layer topology (for more detail please see HADOOP-8468 JIRA issue). To enable this feature set the enable_hypervisor_awareness parameter to true in the configuration file. In this case sahara will add the Compute node ID as a second level of topology for virtual machines.

Distributed mode configuration

Sahara can be configured to run in a distributed mode that creates a separation between the API and engine processes. This allows the API process to remain relatively free to handle requests while offloading intensive tasks to the engine processes.

The sahara-api application works as a front-end and serves user requests. It offloads ‘heavy’ tasks to the sahara-engine process via RPC mechanisms. While the sahara-engine process could be loaded with tasks, sahara-api stays free and hence may quickly respond to user queries.

If sahara runs on several hosts, the API requests could be balanced between several sahara-api hosts using a load balancer. It is not required to balance load between different sahara-engine hosts as this will be automatically done via the message broker.

If a single host becomes unavailable, other hosts will continue serving user requests. Hence, a better scalability is achieved and some fault tolerance as well. Note that distributed mode is not a true high availability. While the failure of a single host does not affect the work of the others, all of the operations running on the failed host will stop. For example, if a cluster scaling is interrupted, the cluster will be stuck in a half-scaled state. The cluster might continue working, but it will be impossible to scale it further or run jobs on it via EDP.

To run sahara in distributed mode pick several hosts on which you want to run sahara services and follow these steps:

  • On each host install and configure sahara using the installation guide except:

    • Do not run sahara-db-manage or launch sahara with sahara-all

    • Ensure that each configuration file provides a database connection string to a single database for all hosts.

  • Run sahara-db-manage as described in the installation guide, but only on a single (arbitrarily picked) host.

  • The sahara-api and sahara-engine processes use oslo.messaging to communicate with each other. You will need to configure it properly on each host (see below).

  • Run sahara-api and sahara-engine on the desired hosts. You may run both processes on the same or separate hosts as long as they are configured to use the same message broker and database.

To configure oslo.messaging, first you need to choose a message broker driver. The recommended driver is RabbitMQ. For the RabbitMQ drivers please see the Notifications configuration documentation for an explanation of common configuration options; the entire list of configuration options is found in the oslo_messaging_rabbit documentation.

These options will also be present in the generated sample configuration file. For instructions on creating the configuration file please see the Sahara Configuration Guide.

Distributed periodic tasks configuration

If sahara is configured to run in distributed mode (see Distributed mode configuration), periodic tasks can also be launched in distributed mode. In this case tasks will be split across all sahara-engine processes. This will reduce overall load.

Distributed periodic tasks are based on Hash Ring implementation and the Tooz library that provides group membership support for a set of backends. In order to use periodic tasks distribution, the following steps are required:

  • One of the supported backends should be configured and started.

  • Backend URL should be set in the sahara configuration file with the periodic_coordinator_backend_url parameter. For example, if the ZooKeeper backend is being used:

    [DEFAULT]
    periodic_coordinator_backend_url=kazoo://IP:PORT
    
  • Tooz extras should be installed. When using Zookeeper as coordination backend, kazoo library should be installed. It can be done with pip:

    pip install tooz[zookeeper]
    
  • Periodic tasks can be performed in parallel. Number of threads to run periodic tasks on a single engine can be set with periodic_workers_number parameter (only 1 thread will be launched by default). Example:

    [DEFAULT]
    periodic_workers_number=2
    
  • coordinator_heartbeat_interval can be set to change the interval between heartbeat execution (1 second by default). Heartbeats are needed to make sure that connection to the coordination backend is active. Example:

    [DEFAULT]
    coordinator_heartbeat_interval=2
    
  • hash_ring_replicas_count can be set to change the number of replicas for each engine on a Hash Ring. Each replica is a point on a Hash Ring that belongs to a particular engine. A larger number of replicas leads to better task distribution across the set of engines. (40 by default). Example:

    [DEFAULT]
     hash_ring_replicas_count=100
    

External key manager usage

Sahara generates and stores several passwords during the course of operation. To harden sahara’s usage of passwords it can be instructed to use an external key manager for storage and retrieval of these secrets. To enable this feature there must first be an OpenStack Key Manager service deployed within the stack.

With a Key Manager service deployed on the stack, sahara must be configured to enable the external storage of secrets. Sahara uses the castellan library to interface with the OpenStack Key Manager service. This library provides configurable access to a key manager. To configure sahara to use barbican as the key manager, edit the sahara configuration file as follows:

[DEFAULT]
use_barbican_key_manager=true

Enabling the use_barbican_key_manager option will configure castellan to use barbican as its key management implementation. By default it will attempt to find barbican in the Identity service’s service catalog.

For added control of the barbican server location, optional configuration values may be added to specify the URL for the barbican API server.

[castellan]
barbican_api_endpoint=http://{barbican controller IP:PORT}/
barbican_api_version=v1

The specific values for the barbican endpoint will be dictated by the IP address of the controller for your installation.

With all of these values configured and the Key Manager service deployed, sahara will begin storing its secrets in the external manager.

Indirect instance access through proxy nodes

Warning

The indirect VMs access feature is in alpha state. We do not recommend using it in a production environment.

Sahara needs to access instances through SSH during cluster setup. This access can be obtained a number of different ways (see Floating IP management,:ref:custom_network_topologies).Sometimes it is impossible to provide access to all nodes (because of limited numbers of floating IPs or security policies). In these cases access can be gained using other nodes of the cluster as proxy gateways. To enable this set is_proxy_gateway=true for the node group you want to use as proxy. Sahara will communicate with all other cluster instances through the instances of this node group.

Note, if use_floating_ips=true and the cluster contains a node group with is_proxy_gateway=true, the requirement to have floating_ip_pool specified is applied only to the proxy node group. Other instances will be accessed through proxy instances using the standard private network.

Note, the Cloudera Hadoop plugin doesn’t support access to Cloudera manager through a proxy node. This means that for CDH clusters only nodes with the Cloudera manager can be designated as proxy gateway nodes.

Multi region deployment

Sahara supports multi region deployment. To enable this option each instance of sahara should have the os_region_name=<region> parameter set in the configuration file. The following example demonstrates configuring sahara to use the RegionOne region:

[DEFAULT]
os_region_name=RegionOne

Non-root users

In cases where a proxy command is being used to access cluster instances (for example, when using namespaces or when specifying a custom proxy command), rootwrap functionality is provided to allow users other than root access to the needed operating system facilities. To use rootwrap the following configuration parameter is required to be set:

[DEFAULT]
use_rootwrap=true

Assuming you elect to leverage the default rootwrap command (sahara-rootwrap), you will need to perform the following additional setup steps:

  • Copy the provided sudoers configuration file from the local project file etc/sudoers.d/sahara-rootwrap to the system specific location, usually /etc/sudoers.d. This file is setup to allow a user named sahara access to the rootwrap script. It contains the following:

sahara ALL = (root) NOPASSWD: /usr/bin/sahara-rootwrap /etc/sahara/rootwrap.conf *

When using devstack to deploy sahara, please pay attention that you need to change user in script from sahara to stack.

  • Copy the provided rootwrap configuration file from the local project file etc/sahara/rootwrap.conf to the system specific location, usually /etc/sahara. This file contains the default configuration for rootwrap.

  • Copy the provided rootwrap filters file from the local project file etc/sahara/rootwrap.d/sahara.filters to the location specified in the rootwrap configuration file, usually /etc/sahara/rootwrap.d. This file contains the filters that will allow the sahara user to access the ip netns exec, nc, and kill commands through the rootwrap (depending on proxy_command you may need to set additional filters). It should look similar to the followings:

[Filters]
ip: IpNetnsExecFilter, ip, root
nc: CommandFilter, nc, root
kill: CommandFilter, kill, root

If you wish to use a rootwrap command other than sahara-rootwrap you can set the following parameter in your sahara configuration file:

[DEFAULT]
rootwrap_command='sudo sahara-rootwrap /etc/sahara/rootwrap.conf'

For more information on rootwrap please refer to the official Rootwrap documentation

Object Storage access using proxy users

To improve security for clusters accessing files in Object Storage, sahara can be configured to use proxy users and delegated trusts for access. This behavior has been implemented to reduce the need for storing and distributing user credentials.

The use of proxy users involves creating an Identity domain that will be designated as the home for these users. Proxy users will be created on demand by sahara and will only exist during a job execution which requires Object Storage access. The domain created for the proxy users must be backed by a driver that allows sahara’s admin user to create new user accounts. This new domain should contain no roles, to limit the potential access of a proxy user.

Once the domain has been created, sahara must be configured to use it by adding the domain name and any potential delegated roles that must be used for Object Storage access to the sahara configuration file. With the domain enabled in sahara, users will no longer be required to enter credentials for their data sources and job binaries referenced in Object Storage.

Detailed instructions

First a domain must be created in the Identity service to hold proxy users created by sahara. This domain must have an identity backend driver that allows for sahara to create new users. The default SQL engine is sufficient but if your keystone identity is backed by LDAP or similar then domain specific configurations should be used to ensure sahara’s access. Please see the Keystone documentation for more information.

With the domain created, sahara’s configuration file should be updated to include the new domain name and any potential roles that will be needed. For this example let’s assume that the name of the proxy domain is sahara_proxy and the roles needed by proxy users will be member and SwiftUser.

[DEFAULT]
use_domain_for_proxy_users=true
proxy_user_domain_name=sahara_proxy
proxy_user_role_names=member,SwiftUser

A note on the use of roles. In the context of the proxy user, any roles specified here are roles intended to be delegated to the proxy user from the user with access to Object Storage. More specifically, any roles that are required for Object Storage access by the project owning the object store must be delegated to the proxy user for authentication to be successful.

Finally, the stack administrator must ensure that images registered with sahara have the latest version of the Hadoop swift filesystem plugin installed. The sources for this plugin can be found in the sahara extra repository. For more information on images or swift integration see the sahara documentation sections Building guest images and Swift Integration.

Volume instance locality configuration

The Block Storage service provides the ability to define volume instance locality to ensure that instance volumes are created on the same host as the hypervisor. The InstanceLocalityFilter provides the mechanism for the selection of a storage provider located on the same physical host as an instance.

To enable this functionality for instances of a specific node group, the volume_local_to_instance field in the node group template should be set to true and some extra configurations are needed:

  • The cinder-volume service should be launched on every physical host and at least one physical host should run both cinder-scheduler and cinder-volume services.

  • InstanceLocalityFilter should be added to the list of default filters (scheduler_default_filters in cinder) for the Block Storage configuration.

  • The Extended Server Attributes extension needs to be active in the Compute service (this is true by default in nova), so that the OS-EXT-SRV-ATTR:host property is returned when requesting instance info.

  • The user making the call needs to have sufficient rights for the property to be returned by the Compute service. This can be done by:

    • by changing nova’s policy.yaml to allow the user access to the extended_server_attributes option.

    • by designating an account with privileged rights in the cinder configuration:

      os_privileged_user_name =
      os_privileged_user_password =
      os_privileged_user_tenant =
      

It should be noted that in a situation when the host has no space for volume creation, the created volume will have an Error state and can not be used.

Autoconfiguration for templates

Autoconfiguring templates

NTP service configuration

By default sahara will enable the NTP service on all cluster instances if the NTP package is included in the image (the sahara disk image builder will include NTP in all images it generates). The default NTP server will be pool.ntp.org; this can be overridden using the default_ntp_server setting in the DEFAULT section of the sahara configuration file.

If you are creating cluster templates using the sahara UI and would like to specify a different NTP server for a particular cluster template, use the URL of NTP server setting in the General Parameters section when you create the template. If you would like to disable NTP for a particular cluster template, deselect the Enable NTP service checkbox in the General Parameters section when you create the template.

If you are creating clusters using the sahara CLI, you can specify another NTP server or disable NTP service using the examples below.

If you want to enable configuring the NTP service, you should specify the following configs for the cluster:

{
    "cluster_configs": {
        "general": {
            "URL of NTP server": "your_server.net"
        }
    }
}

If you want to disable configuring NTP service, you should specify following configs for the cluster:

{
    "cluster_configs": {
        "general": {
            "Enable NTP service": false
        }
    }
}

CORS (Cross Origin Resource Sharing) Configuration

Sahara provides direct API access to user-agents (browsers) via the HTTP CORS protocol. Detailed documentation, as well as troubleshooting examples, may be found in the documentation of the oslo.db cross-project features.

To get started quickly, use the example configuration block below, replacing the allowed origin field with the host(s) from which your API expects access.

[cors]
allowed_origin=https://we.example.com:443
max_age=3600
allow_credentials=true

[cors.additional_domain_1]
allowed_origin=https://additional_domain_1.example.com:443

[cors.additional_domain_2]
allowed_origin=https://additional_domain_2.example.com:443

For more information on Cross Origin Resource Sharing, please review the W3C CORS specification.

Cleanup time for incomplete clusters

Sahara provides maximal time (in hours) for clusters allowed to be in states other than “Active”, “Deleting” or “Error”. If a cluster is not in “Active”, “Deleting” or “Error” state and last update of it was longer than cleanup_time_for_incomplete_clusters hours ago then it will be deleted automatically. You can enable this feature by adding appropriate config property in the DEFAULT section (by default it set up to 0 value which means that automatic clean up is disabled). For example, if you want cluster to be deleted after 3 hours if it didn’t leave “Starting” state then you should specify:

[DEFAULT]
cleanup_time_for_incomplete_clusters = 3

Security Group Rules Configuration

When auto_security_group is used, the amount of created security group rules may be bigger than the default values configured in neutron.conf. Then the default limit should be raised up to some bigger value which is proportional to the number of cluster node groups. You can change it in neutron.conf file:

[quotas]
quota_security_group = 1000
quota_security_group_rule = 10000

Or you can execute openstack CLI command:

openstack quota set --secgroups 1000 --secgroup-rules 10000 $PROJECT_ID