Skip to content

Latest commit

 

History

History
500 lines (395 loc) · 18.6 KB

README.md

File metadata and controls

500 lines (395 loc) · 18.6 KB

Build Status Maintainability Test Coverage

aws-ops

Utils to provision services on AWS

Incomplete - work in progress.

Goals

  • use ruby AWS API client to provision AWS resources
  • use capistrano for service installation/configuration
  • use bash scripts and/or puppet provision tools
  • use convenient capistrano/rake tasks
  • manage settings with the config gem
    • default settings are in config/settings/test.yml
  • use AWS discovery services to find everything dynamically

Background information and resources

Install

git clone https://github.com/darrenleeweber/aws-ops.git
cd aws-ops
bundle install
bundle exec cap -T

Additional Documentation

Getting Started

AWS resources and services are defined by YAML settings and managed by capistrano tasks. This is not exactly a quick-start, turn-key solution; there are settings to change and things to think about; read all the details below and use the online resources to provision systems.

  • review config/settings and config/deploy content
  • create a new {STAGE}, let's call it 'dev'
cp config/settings/test.yml config/settings/dev.yml
sed -i s/test/dev/ config/settings/dev.yml
sed -i s/TEST/DEV/ config/settings/dev.yml

cp config/deploy/dev.rb config/deploy/dev.rb
sed -i s/test/dev/ config/deploy/dev.rb
  • make some decisions about the settings in those files
  • establish AWS-API credentials and get a PEM file
    • use the PEM file in config/settings/dev.yml
export AWS_PROFILE=profileX
export AWS_SECRET_ACCESS_KEY=secret_key
export AWS_ACCESS_KEY_ID=secret_key
export AWS_DEFAULT_REGION="regionX"
  • check settings
export CLUSTER_ENV='dev' # or whatever {env} in config/settings/{env}.yml
export STAGE='dev' # or whatever {stage} in config/deploy/{stage}.rb
bundle exec cap ${STAGE} ops:aws:check_settings
  • run capistrano tasks to manage AWS nodes
  • run capistrano tasks to provision services on AWS nodes
  • bundle exec cap -T to list node and service tasks
  • read all the details below

Configure

AWS Access Keys

Use the AWS console to create any authorized user/group accounts for access to any or all of the AWS services. For a user, get their access-key-id and the security-access-key. Details can be found at AWS documentation, e.g.

Use the aws configure --profile <username> to set these access credentials for the AWS CLI utility. It saves the values to ~/.aws/credentials.

I also like to set environment variables, e.g.

export AWS_SECRET_ACCESS_KEY=<YourKey>
export AWS_ACCESS_KEY_ID=<YourKeyID>
export AWS_DEFAULT_REGION=<YourRegion>

AWS Key Pairs

Use the AWS console to create and save any Key Pairs to be used for access to any or all of the AWS EC2 instances. Add the name of a key pair to the /config/settings/{stage}.rb instances.

# Setup your AWS credentials using ENV values or config/setting.yml
bundle exec cap dev ops:aws:check_credentials

Once any instance is created and assigned a public DNS, it can be added to your /etc/hosts file and a corresponding entry in ~/.ssh/config file, along with the required AWS Key Pair used to access it. It's recommended that a Host value matches the instance name in config/settings/{stage}.yml and it's essential that the Host value is used in the Capistrano stage file to define a "server" entry by that name (e.g. config/deploy/{stage}.rb settings). (Note that it is easier to use the same CLUSTER_SETTINGS or CLUSTER_ENV name as the capistrano {stage} name, but this is not essential, they can be different.)

An example of an /etc/hosts/ and matching ~/.ssh/config entry:

# /etc/hosts entry
{AWS_PUBLIC_IP} {AWS_PUBLIC_DNS} {YourHostAlias}

# ~/.ssh/config entry
Host <Instance name in your config/settings/{stage}.yml>
  Hostname {YourHostAlias}
  user {ubuntu or something}
  IdentityFile ~/.ssh/{AWS_Key_Pair_Name}.pem
  Port 22

To test the ssh access to any AWS EC2 public DNS, try something like:

ssh -i ~/.ssh/{AWS_Key_Pair_Name}.pem {user}@{AWS_PUBLIC_DNS}

If that works and the /etc/hosts and ~/.ssh/config entries exist, try:

ssh {YourHostAlias}

There are some capistrano tasks to provide details for the ~/.ssh/config and related /etc/hosts values, e.g.

cap zookeeper:nodes:find                 # Find and describe all nodes
cap zookeeper:nodes:etc_hosts_private    # Compose entries for /etc/hosts using private IPs
cap zookeeper:nodes:etc_hosts_public     # Compose entries for /etc/hosts using public IPs
cap zookeeper:nodes:ssh_config_private   # Compose private entries for ~/.ssh/config for nodes
cap zookeeper:nodes:ssh_config_public    # Compose public entries for ~/.ssh/config for nodes

WARNING: the /etc/hosts entries must be updated whenever an instance is stopped and restarted, because the public network interface can change. Also, there might be problems with the ~/.ssh/known_hosts file when a system is restarted and assigned a new host fingerprint.

AWS Switching Accounts

  • reset the AWS environment variables with access credentials
AWS_DEFAULT_REGION={region}
AWS_SECRET_ACCESS_KEY={key}
AWS_ACCESS_KEY_ID={id}
  • update the config/settings/{stage}.yml
    • update the instance default details for the {key-name}.pem, i.e.
      • key_name: {key-name}
    • update the VPC settings for security groups, i.e.
      • vpc_id: {vpc-id}
      • the "Default VPC" is displayed on the EC2 dashboard, "Account Attributes"

Settings

Check details of config/settings.yml and subdirectories; modify the settings as required, esp. AWS details in the instance defaults, like: AMI, AWS region, instance types and tags.

export CLUSTER_ENV='test' # or whatever {env} in config/settings/{env}.yml
export STAGE='test' # or whatever {stage} in config/deploy/{stage}.rb
bundle exec cap ${STAGE} ops:aws:check_settings

Note, the CLUSTER_ENV and STAGE here assume the default capistrano paths are used for configs. Below there are options for using custom paths for configs that will override these settings.

Adding a Stage

Copy test settings into a few new {stage} files, i.e.:

  • config/deploy/{stage}.rb
  • config/settings/{stage}.yml

Review and modify all the values in those files. Make sure you have the access token details and the PEM access file.

Alternatively, try the custom config paths described below.

Using Custom Config Paths

It's possible to use config paths that are outside the paths of this aws-ops project.

Custom settings: the default settings are in config/settings.yml merged with config/settings/test.yml. These can be replaced by a single custom settings file. Use an absolute path to a settings file in the environment variable CLUSTER_SETTINGS, e.g.

export CLUSTER_SETTINGS="${HOME}/my_cluster/settings/test.yml"

Custom deploy: the default deploy file is in config/deploy.rb. This path can be replaced with a custom deploy file, by giving an absolute path to a deploy file in the environment variable CLUSTER_DEPLOY_PATH, e.g.

export CLUSTER_DEPLOY_PATH="${HOME}/my_cluster/deploy/deploy.rb"

Custom stage: the default stage file is in config/deploy/test.rb. This path can be replaced with a custom stage file, by giving an absolute path to a stage file in the environment variable CLUSTER_STAGE_PATH, e.g.

export CLUSTER_STAGE_PATH="${HOME}/my_cluster/stage/stage.rb"

Custom tasks: the default tasks are in lib/**/*.rake. These can be supplemented (or overriden) by adding a custom tasks path. Add an absolute path to the environment variable CLUSTER_TASKS_PATH; this should be the root path of all the tasks and it gets expanded to find all the *.rake files anywhere below that path, e.g.

export CLUSTER_TASKS_PATH="${HOME}/my_cluster/lib"

WARNING: these are experimental features that have not been fully tested.

General Use

Try to create new instances for the services to provision (e.g., zookeeper and kafka).

# Always set the environment variables noted above to configure capistrano.
export STAGE={stage}
bundle exec cap ${STAGE} ops:aws:check_settings
bundle exec cap ${STAGE} zookeeper:nodes:create
bundle exec cap ${STAGE} zookeeper:nodes:find
bundle exec cap ${STAGE} kafka:nodes:create
bundle exec cap ${STAGE} kafka:nodes:find
# create nodes for additional services and then get the /etc/hosts details, e.g.
bundle exec cap ${STAGE} zookeeper:nodes:etc_hosts_public | sudo tee -a /etc/hosts
bundle exec cap ${STAGE} kafka:nodes:etc_hosts_public | sudo tee -a /etc/hosts
# ensure the hostnames in /etc/hosts match those in /config/deploy/{stage}.rb,
# then deploy the aws-ops code (be sure the code is pushed to github)
bundle exec cap ${STAGE} deploy:check
bundle exec cap ${STAGE} deploy

Login shell on remote servers

The capistrano-shell plugin can drop you into a shell on a remote server into the project deployment directory.

export STAGE={stage}
bundle exec cap ${STAGE} shell

Provision Software with Capistrano

At this point, the capistrano tasks coordinate software installation and configuration on the servers, identified by their roles in config/deploy/{stage}.yml.

Usually, the first things to provision are general OS utilities and build tools.

export STAGE={stage}
bundle exec cap -T | grep ubuntu
bundle exec cap ${STAGE} ubuntu:update
bundle exec cap ${STAGE} ubuntu:install:build_tools
bundle exec cap ${STAGE} ubuntu:install:java_8_oracle
bundle exec cap ${STAGE} ubuntu:install:network_tools
bundle exec cap ${STAGE} ubuntu:install:os_utils

Then provision services (see the ":service:install" and ":service:configure" tasks) and then run services (see the "*:service:start" tasks). Be careful to provision configure and start services in order (e.g., zookeeper before anything that depends on it).

ZooKeeper

  • double check the latest releases available, see
    • https://zookeeper.apache.org/releases.html
    • don't just get the latest release, get the release that your cluster services recommend
    • TODO: allow a custom version of zookeeper
      • currently uses the default package for Ubuntu
      • the settings file has no version options for zookeeper
export STAGE={stage}
bundle exec cap -T | grep zookeeper
bundle exec cap ${STAGE} zookeeper:nodes:find
bundle exec cap ${STAGE} zookeeper:service:install
bundle exec cap ${STAGE} zookeeper:service:configure
bundle exec cap ${STAGE} zookeeper:service:start
bundle exec cap ${STAGE} zookeeper:service:status
# if all goes well, the status should report 'imok'
# Also check the 'srvr' details and look for leader/follower 'Mode'.
bundle exec cap ${STAGE} zookeeper:service:command['srvr']
# If it's 'Mode: standalone', stop and restart the service until the
# 'Mode' shows the servers have formed a quorum and elected a leader.
# It usually resolves after one, maybe two, restarts
bundle exec cap ${STAGE} zookeeper:service:stop # wait 10-30 sec
bundle exec cap ${STAGE} zookeeper:service:start # wait 10-30 sec
bundle exec cap ${STAGE} zookeeper:service:command['srvr']

ZooNavigator

export STAGE={stage}
bundle exec cap ${STAGE} zoonavigator:service:install
# If it fails, try it again (it might require the user to reconnect to enable docker)
bundle exec cap ${STAGE} zoonavigator:service:connections
# {stage}_zookeeper1:2181,{stage}_zookeeper2:2181,{stage}_zookeeper3:2181

Visit http://{stage}_zookeeper1:8001 and copy and paste the "connections" information into the 'Connection string' on the login dialog. Unless authorization is enabled, leave those fields blank

Kafka

export STAGE={stage}
bundle exec cap ${STAGE} kafka:nodes:find
bundle exec cap ${STAGE} kafka:service:install
bundle exec cap ${STAGE} kafka:service:configure
bundle exec cap ${STAGE} kafka:service:start
bundle exec cap ${STAGE} kafka:service:status

If the start succeeds, it does not mean that Kafka is running. When the status indicates that Kafka is running, it's running. If not, then check the Kafka logs and trouble shoot the startup. Sometimes Kafka fails to startup when the Zookeeper /kafka node is first created; the logs will indicate that Kafka created it but then couldn't find it. Just wait a minute and try to start Kafka again. To view logs, try

export STAGE={stage}
bundle exec cap ${STAGE} kafka:service:tail_server_log["${STAGE}_kafka1"]

When Kafka is running, confirm that it has registered brokers with ZooKeeper. Use the ZooNavigator UI to refresh the zoo-nodes and check that the Kafka node tree contains some brokers. The /kafka/brokers/ids should list all the brokers configured in config/settings/{STAGE}.yml

  • See Kafka docs for more about testing Kafka

Kafka Manager

TODO - run it on any system (e.g. a laptop) and connect to the cluster on AWS

Explanation of enabling AWS hosts for Capistrano

AWS EC2 Instance Information

To get AWS EC2 instance connection details, e.g.

export STAGE={stage}
bundle exec cap ${STAGE} zookeeper:nodes:create
# Wait a while for the Public IP and Public DNS values to be available, then:
bundle exec cap ${STAGE} zookeeper:nodes:find
# {
#   "ID": "i-0fd060e5124453b11",
#   "Type": "t2.medium",
#   "AMI ID": "ami-6e1a0117",
#   "A. Zone": "us-west-2a",
#   "State": "running",
#   "Tags": "Group: test_zookeeper; Name: test_zookeeper1; Service: zookeeper; Manager: dlweber; Stage: test",
#   "Key Pair": "test",
#   "Public IP": "55.27.247.237",
#   "Private IP": "172.31.22.75",
#   "Public DNS": "ec2-55-27-247-237.us-west-2.compute.amazonaws.com",
#   "Private DNS": "ip-172-31-22-75.us-west-2.compute.internal"
# }
# etc.

As noted above, there are also tasks to provide all the details for ~/.ssh/config and /etc/hosts for each service, e.g.

cap zookeeper:nodes:find                 # Find and describe all nodes
cap zookeeper:nodes:etc_hosts_private    # Compose entries for /etc/hosts using private IPs
cap zookeeper:nodes:etc_hosts_public     # Compose entries for /etc/hosts using public IPs
cap zookeeper:nodes:ssh_config_private   # Compose private entries for ~/.ssh/config for nodes
cap zookeeper:nodes:ssh_config_public    # Compose public entries for ~/.ssh/config for nodes

Capistrano configuration

Once AWS EC2 instances are running, add their connection details to the config/deploy/*.rb files and assign roles to each instance; e.g.

server 'ec2-52-32-121-252.us-west-2.compute.amazonaws.com', user: 'ubuntu', roles: %w{zookeeper}

If any instance serves multiple roles, just add them to the roles. This can allow installation of multiple services on one instance. (The general assumption of this project, however, is that an instance will host one service.)

If you want to use a ~/.ssh/config file to rename and manage the public DNS entries for your AWS instances, the connection details can be simplified to something like:

server 'test_zookeeper1', user: 'ubuntu', roles: %w{zookeeper}

In this case, the ~/.ssh/config file contains:

Host test_zookeeper1
  Hostname ec2-52-32-121-252.us-west-2.compute.amazonaws.com
  user ubuntu
  IdentityFile ~/.ssh/an_aws_key.pem
  Port 22

Their connection details can be found using various aws:ops tasks or more specific service tasks, like zookeeper:nodes:find.

WARNING: if the systems are stopped and restarted, AWS can reassign the public DNS entries, which must be updated again. (There are solutions to this problem that require additional IP management for instances.)

WARNING: Some service installation set environment variables in /etc/profile.d/*.sh, like KAFKA_HOME, and add some entries to the PATH, like $KAFKA_HOME/bin to the PATH. While these are available to a login shell, capistrano explicitly ignores them, see http://capistranorb.com/documentation/faq/why-does-something-work-in-my-ssh-session-but-not-in-capistrano/# To work around this, the values in config/settings/{stage}.yml that are used to set the environment variables are also used explicitly in some capistrano tasks. (Note, to avoid confusion in this project, the settings/variable names for these are lower case.)

Capistrano Deployment and Connections

The config/deploy.rb and the config/deploy/{stage}.rb files contain the connection information for all the systems. By default, cap {stage} deploy will deploy this code repository to the ~/aws-ops path (using conventional capistrano deployment directories). Some of the capistrano tasks require access to the files in this project on the remote hosts, so it's best to deploy this project and then run some system/service package installation and configuration tasks.

To test the deployment, use

export STAGE={stage}
bundle exec cap ${STAGE} deploy:check

To run the deployment, use

export STAGE={stage}
bundle exec cap ${STAGE} deploy

To connect to a remote host, this project includes the capistrano-shell gem, e.g.

export STAGE={stage}
bundle exec cap ${STAGE} shell
# when multiple hosts are configured, it prompts for a specific host to connect to.

Capistrano Namespaces and Services Available

See bundle exec cap -T for a complete listing. Additional README docs might be added to lib/{service}/README.md.

  • ops:aws
  • hdfs (TODO)
  • kafka
  • kafka_manager (TODO)
  • mesos (TODO)
  • spark (TODO)
  • zookeeper
  • zoonavigator
  • others include many common capistrano tasks, like deploy
  • see find . -name '*.rake' for details
  • task helpers are in lib/** and included in Capfile

Licence

Copyright 2017 Darren L. Weber, Ph.D. under the Apache 2 license

This project was inspired by and adapted from: https://github.com/adobe-research/spark-cluster-deployment

  • copyright 2014 Adobe Systems Incorporated under the Apache 2 license
  • initial adaptations are conversion of fabric (python) to capistrano (ruby)
  • initial project code here does not use the puppet code upstream
  • this project adopts the AWS API client for ruby (aws-sdk gem)
  • the adaptations are too substantial to constitute a fork