This post looks like I am jumping on the bandwagon of IT orchestration like a lot of people are doing. Maybe I should say ‘except for (die hard) Oracle DBA’s’. Or maybe not, it up to you to decide.
Most people who are interested in IT in general will have noticed IT orchestration has gotten attention, especially in the form of Puppet and/or Chef. I _think_ IT orchestration has gotten important with the rise of “web scale” (scaling up and down applications by adding virtual machines to horizontal scale resource intensive tasks), in order to provision/configure the newly added machines without manual intervention, and people start picking it up now to use it for more tasks than provisioning of virtual machines for web applications.
I am surprised by that. I am not surprised that people want boring tasks like making settings in configuration files and restarting daemons, installing software with all the correct options, etc. being automated. Instead, I am surprised that people are now picking this up after it has been around for so long.
A little history.
As far as I know, IT orchestration started with cfengine, which was really a configuration engine (hence the name). Despite having a little different purpose (configuration vs. orchestration), this tool is the parent of all the orchestration/configuration tools which exist nowaday. cfengine started off as a study in 1993 by Mark Burgess at the university of Oslo, with the creation of the cfengine software as a result. As far as I can see, it has been available as open source software since the beginning.
Now back to what I am surprised at: with cfengine, there has been a way to configure linux/unix systems in a structured way, and install and configure software on systems since the mid-nineties. Yet, this wasn’t picked up (of course with a few exceptions). Fast forward to today, we see it is being adopted. And that is a good thing.
I created a setup with cfengine for a client a long time ago, which had the ability to install the Oracle software, different PSU’s in different home’s, and remove it by adding or removing machines to groups in a cfengine configuration file. It wasn’t picked up by the client, it’s way more fun running X to install the software, and make the choices by hand, and redo this over and over on every machine, right?
I almost forgotten about my work with cfengine, until I spoke with Alex Gorbatchev at a conference, at which he pointed me to Ansible. At first I didn’t do a lot with it, but lately I’ve given it a go, and I am very happy with it.
Another redo of cfengine?
From what I read, most of the configuration/orchestration engines created after cfengine are created to circumvent all kinds of difficulties with cfengine. I can understand that. It took me a while to learn cfengine, and indeed it forces you to think in a different way.
The Ansible project decided to radically do it different than all the other engines. It is different in the sense that it advertises itself as simple, agentless and powerful.
Simple.
Simple is a terrific goal. For those of you that have worked with configuration/orchestration engines, there is a steep learning curve. It is just hard to get the basic principles in your head. To be honest, also Ansible took me a while too, to grasp the basic principles, and get the picture correctly in my head. Yet, having worked with cfengine comparing it with Ansible’s playbooks, which are the scripts to do things on the targets, it is a breath of fresh air. Playbooks are so clean they (almost) can be read and understood as plain english.
Agentless.
This is where Ansible is truly different than any of the other configuration/orchestration tools. Ansible does not require any agent installation on the targets. The obvious next question then is: how can this work? Well, quite simple: Ansible uses ssh to connect to the host, and executes commands via the shell. Having that said, it requires a little more detail; Ansible uses python on the remote host for it’s normal execution. However, you can use it without python, for example to setup the host up for the Ansible normal usage mode Which requires python and the simple-json module.
This is truly important, and makes it an excellent fit for my daily work as an IT consultant.
Powerful.
Ansible is powerful in the way that you can do the configuration and orchestration tasks in a simple clean way.
Summary on the introduction.
Above was a brief personal history, and some of the “marketed” features of Ansible. I think being agentless is the true “killer feature” here. All the other configuration/orchestration engines require you to setup and configure a fixed client-server connection, and install a deamon and a central server process. In case you wondered, yes, authentication is important, and it’s simply brilliant that the ssh password authentication or public key infrastructure can be used.
Because there’s no daemon to install, you can run your created play books everywhere. So instead of a fixed client configuration, you can create play books to do routine tasks, and repeat it at multiple sites.
Okay, how does this work?
Installation: add EPEL and install ansible.
If you are on one of the clones of RedHat Enterprise Linux (I use Oracle Linux), you simply need to add the EPEL repository to your yum source list, and run:
# yum install ansible
First steps.
One of the first things I do, is create a directory for a typical ansible ‘project’. Project means a set of tasks you want to do to a set of hosts here. Next, I create a file called ‘hosts’ which is the list of hosts you want to use for executing tasks on. By default, Ansible looks in /etc/ansible/hosts. In this case, I put a single machine in it (a test VM), but it can be a list of machines (ip addresses or hostnames).
$ cat hosts
192.168.101.2
In fact, you can create groups in the hosts file in the “ini style”. But I just put one host in for this example.
The next thing is to check if Ansible reads the file correctly. This is done in the following way:
$ ansible all -i hosts --list-hosts
192.168.101.2
Okay, this means Ansible will operate on this one host if invoked. The next logical thing (typically done when you are in a new client environment to check if you can reach the hosts):
$ ansible all -i hosts -m ping
192.168.101.2 | FAILED => FAILED: Authentication failed.
Ping might be a bit misleading for some people. What ping does here (-m means module), is trying to connect to the host over ssh, and log in. Because I didn’t specify a user, it used the username of the current user on the machine, which is ‘ansible’. A user ‘ansible’ typically doesn’t exist on a normal server (and is not necessary or should be created), and also not on my test server. So it failed, as the message said, on authentication.
My test VM is a basic installed (OL) linux 6 server. This means there’s only one user: root.
So, let’s specify the user root as user:
$ ansible all -i hosts -m ping -u root
192.168.101.2 | FAILED => FAILED: Authentication failed.
The authentication failed again. And it should! What this is doing, is trying to log on as root, and we haven’t given any password, nor have I put my local user’s public key in the remote authorised_keys file. So there is no way this could work. This is typically also the state when you want to do stuff with a “fresh” client system. Let’s add the ‘-k’ option (ask ssh password), and run again:
$ ansible all -i hosts -m ping -u root -k
SSH password:
192.168.101.2 | success >> {
"changed": false,
"ping": "pong"
}
To walk you through the output: It now asks for a password, which I’ve filled out, then lists the host and the state: success. During this execution, there was nothing changed on the remote host, and the ping command resulted in a pong (alike the ICMP ping response).
With what we have learned now, we can do things like this:
$ ansible all -i hosts -u root -k -a "ifconfig"
SSH password:
192.168.101.2 | success | rc=0 >>
eth0 Link encap:Ethernet HWaddr 00:0C:29:14:65:ED
inet addr:192.168.39.145 Bcast:192.168.39.255 Mask:255.255.255.0
inet6 addr: fe80::20c:29ff:fe14:65ed/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:47 errors:0 dropped:0 overruns:0 frame:0
TX packets:25 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:6293 (6.1 KiB) TX bytes:2594 (2.5 KiB)
eth1 Link encap:Ethernet HWaddr 00:0C:29:14:65:F7
inet addr:192.168.101.2 Bcast:192.168.101.255 Mask:255.255.255.0
inet6 addr: fe80::20c:29ff:fe14:65f7/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:188 errors:0 dropped:0 overruns:0 frame:0
TX packets:112 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:142146 (138.8 KiB) TX bytes:15545 (15.1 KiB)
lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
inet6 addr: ::1/128 Scope:Host
UP LOOPBACK RUNNING MTU:65536 Metric:1
RX packets:0 errors:0 dropped:0 overruns:0 frame:0
TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:0 (0.0 b) TX bytes:0 (0.0 b)
Does this look familiar for you Exadata DBA’s? Yes, this replicates some of the functionality of dcli (although dcli is aimed at executing simple tasks to a group of hosts, whilst Ansible is aimed at enterprise configuration and orchestration).
One step beyond! Playbooks.
Now let’s progress to playbooks. An Ansible playbook is where the true strength lies of Ansible. It allows you to specify tasks to execute on the remote hosts, and create sequences of tasks and make decisions based on the outcome of a tasks for further execution. Let me show you a simple playbook, and guide you through it:
---
- hosts: all
gather_facts: no
remote_user: root
tasks:
- name: upgrade all packages
yum: name=* state=latest
- name: install python-selinux
yum: name=libselinux-python state=installed
- name: add public key to authorized_key file of root
authorized_key: user=root state=present key="{{ lookup('file','/home/ansible/.ssh/id_rsa.pub') }}"
As you can see, this is a playbook with three tasks: upgrade all packages, install libselinux-python and adding my (local) public key to the authorised key file of root (to allow passwordless access).
Line 1 shows three dashes, which means the start of a YAML document.
Line 2 starts with a single dash, which indicates a list. There is one dash at this indention level, so it’s a list of one. The fields of this member are hosts, gather_facts and tasks. Tasks got his own list (mind the indention level, that is important). The fields are key/value pairs, with the separation indicated by the colon (:). The first field is ‘hosts’, with the value ‘all’. This means that all hosts in the hosts file are used for this playbook. I don’t think it’s hard to imagine how useful it can be to specify a group/kind of servers the playbook can run on. The next one is ‘gather_facts’. A normal playbook execution first gathers a lot of information from all the hosts it is going to run on before execution. These can be used during playbook execution. Next ‘remote_user’. This indicates with which user ansible is going to logon, so we don’t have to specify it on the command line. Then we see ‘tasks’ to indicate the list of tasks to be executed on the hosts.
It’s easy to spot we got three tasks. What is extremely important, is the indention of this list (it’s got a dash, so it’s a list!). Name is not mandatory, but it makes it easy to read if you give the tasks useful names and these will be shown when the playbook is executed. The first task has the name ‘upgrade all packages’. The next field shows the key is ‘yum’ indicating it is making use of the yum module. This key got two values: name=*, which means all ‘all packages’, and state=latest, which means we want all packages to be at the latest version. This means this command is the equivalent of ‘yum update’.
The second task is called ‘install python-selinux’. It makes use of the yum module again, and is self explanatory, it installs the libselinux-python package. This packages is necessary to work on a host which has selinux enabled on things that are protected by selinux.
The next task is called ‘add public key to authorised_key file of root’. It is making use of the authorized_key module. This module requires a parameter ‘key’, for which we use the lookup function to look up the local (!) public key, of the user with which I execute ansible, which is ‘ansible’. ‘state=present’ means we want this key to be present; ‘present’ is the default value, so it wasn’t necessary to put this in. Next ‘user=root’: we want the public key to be added to the authorized_keys file of the user root.
Of course these tasks could be executed using the ‘ansible’ executable as single tasks. To show the importance of the installation of the libselinux-python module on a host with selinux enabled (which is the state of selinux on a fresh installed Oracle Linux machine), let’s execute the task using the authorized_key module:
$ ansible all -i hosts -k -u root -m authorized_key -a "user=root state=present key=\"{{ lookup('file','/home/ansible/.ssh/id_rsa.pub') }}\""
SSH password:
192.168.101.2 | FAILED >> {
"failed": true,
"msg": "Aborting, target uses selinux but python bindings (libselinux-python) aren't installed!"
}
Clear, right? The host is selinux protected. Now, let’s execute the installation of the libselinux package as single task, and then add our public key to the authorized_key file of root:
$ ansible all -i hosts -k -u root -m yum -a "name=libselinux-python state=installed"
SSH password:
192.168.101.2 | success >> {
"changed": true,
"msg": "",
"rc": 0,
"results": [
"Loaded plugins: security\nSetting up Install Process\nResolving Dependencies\n--> Running transaction check\n---> Package libselinux-python.x86_64 0:2.0.94-5.3.el6_4.1 will be installed\n--> Finished Dependency Resolution\n\nDependencies Resolved\n\n================================================================================\n Package Arch Version Repository Size\n================================================================================\nInstalling:\n libselinux-python x86_64 2.0.94-5.3.el6_4.1 public_ol6_latest 201 k\n\nTransaction Summary\n================================================================================\nInstall 1 Package(s)\n\nTotal download size: 201 k\nInstalled size: 653 k\nDownloading Packages:\nRunning rpm_check_debug\nRunning Transaction Test\nTransaction Test Succeeded\nRunning Transaction\n\r Installing : libselinux-python-2.0.94-5.3.el6_4.1.x86_64 1/1 \n\r Verifying : libselinux-python-2.0.94-5.3.el6_4.1.x86_64 1/1 \n\nInstalled:\n libselinux-python.x86_64 0:2.0.94-5.3.el6_4.1 \n\nComplete!\n"
]
}
$ ansible all -i hosts -k -u root -m authorized_key -a "user=root state=present key=\"{{ lookup('file','/home/ansible/.ssh/id_rsa.pub') }}\""
SSH password:
192.168.101.2 | success >> {
"changed": true,
"key": "ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAliR905hxLnsOCRlOGnmN0H9dGH4NPV88ySC6GMv0KNnU7FfCXYE51Bkk97p2IWFsPhYO9qJDyAFxRm/lia1IZRDpCFcKKKMh5eXmEJC5XSrHWFdmGZRlFcS3VQ3rpCIyU3qFM6xMazh3JHKKEtE1J6nvw/hW3slY9G/6VoJ8CzpfeQMLDOdVXUIcZXqtCPuIEDBQ7yjfMzTGz+hEmz7ImbLaUyB4MDGrDnl33L8mkBEVYu8RrwgBcagDQSiQKnIca/EL45eX/74NG1e/6vxZkHZJz/W0ak4KD+o9vF4ikz0bdrGPMZ5gRYXWoSSHrVA+Rqk8A93qBXNKUUkzGoQYTQ== ansible@ansiblevm.local",
"key_options": null,
"keyfile": "/root/.ssh/authorized_keys",
"manage_dir": true,
"path": null,
"state": "present",
"unique": false,
"user": "root"
}
Maybe your customer doesn’t want you to store your keys in their servers. It’s easy to do the reverse, and remove your key from the authorized_key file:
$ ansible all -i hosts -u root -m authorized_key -a "user=root state=absent key=\"{{ lookup('file','/home/ansible/.ssh/id_rsa.pub') }}\""
192.168.101.2 | success >> {
"changed": true,
"key": "ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAliR905hxLnsOCRlOGnmN0H9dGH4NPV88ySC6GMv0KNnU7FfCXYE51Bkk97p2IWFsPhYO9qJDyAFxRm/lia1IZRDpCFcKKKMh5eXmEJC5XSrHWFdmGZRlFcS3VQ3rpCIyU3qFM6xMazh3JHKKEtE1J6nvw/hW3slY9G/6VoJ8CzpfeQMLDOdVXUIcZXqtCPuIEDBQ7yjfMzTGz+hEmz7ImbLaUyB4MDGrDnl33L8mkBEVYu8RrwgBcagDQSiQKnIca/EL45eX/74NG1e/6vxZkHZJz/W0ak4KD+o9vF4ikz0bdrGPMZ5gRYXWoSSHrVA+Rqk8A93qBXNKUUkzGoQYTQ== ansible@ansiblevm.local",
"key_options": null,
"keyfile": "/root/.ssh/authorized_keys",
"manage_dir": true,
"path": null,
"state": "absent",
"unique": false,
"user": "root"
}
Please mind I didn’t specify ‘-k’ on the command line to send a password: in the previous step we added our key, so we can access our host using our public key. Another extremely important thing is ‘changed’. ‘changed’ indicates if the task did actually change something on the destination server.
I have ran single task until now, I changed the state of my test VM back to it’s state before I started changing it with ansible (by removing the libselinux package using ‘ansible all -i hosts -k -u root -m yum -a “name=libselinux-python state=absent”‘
Let’s run the above described playbook:
$ ansible-playbook -i hosts -k linux_setup_example.yml
[WARNING]: The version of gmp you have installed has a known issue regarding
timing vulnerabilities when used with pycrypto. If possible, you should update
it (ie. yum update gmp).
SSH password:
PLAY [all] ********************************************************************
TASK: [upgrade all packages] **************************************************
changed: [192.168.101.2]
TASK: [install python-selinux] ************************************************
changed: [192.168.101.2]
TASK: [add public key to authorized_key file of root] *************************
changed: [192.168.101.2]
PLAY RECAP ********************************************************************
192.168.101.2 : ok=3 changed=3 unreachable=0 failed=0
Now at this point you might think: I get it, but these are all pretty simple tasks, it’s not special at all. Well, let me show you an actual thing which totally shows what the importance of using this is, even on a single machine, but even more when you got a large group of servers you have to administer.
The next example is a playbook created to apply PSU3 to an Oracle 11.2.0.4 home. It’s still quite simple, it just applies PSU3 to the Oracle home. But totally automatic. The point I am trying to make is that this is already nice to have automated a lot of work for a single home, but it saves a lot of hours (read: a lot of money), and saves you from human error.
---
- hosts: all
vars:
u01_size_gb: 1
tmp_size_gb: 1
oracle_base: /u01/app/oracle
oracle_home: /u01/app/oracle/product/11.2.0.4/dbhome_1
patch_dir: /u01/install
remote_user: oracle
tasks:
- name: check u01 free disk space
action: shell df -P /u01 | awk 'END { print $4 }'
register: u01size
failed_when: u01size.stdout|int < {{ u01_size_gb }} * 1024 * 1024
- name: check tmp free disk space
action: shell df -P /tmp | awk 'END { print $4 }'
register: tmpsize
failed_when: tmpsize.stdout|int < {{ tmp_size_gb }} * 1024 * 1024
- name: create directory for installation files
action: file dest={{ patch_dir }} state=directory owner=oracle group=oinstall
- name: copy opatch and psu
copy: src=files/{{ item }} dest={{ patch_dir }} owner=oracle group=oinstall mode=0644
with_items:
- p6880880_112000_Linux-x86-64.zip
- p18522509_112040_Linux-x86-64.zip
- ocm.rsp
- name: install opatch in database home
action: shell unzip -oq {{ patch_dir }}/p6880880_112000_Linux-x86-64.zip -d {{ oracle_home }}
- name: unzip psu patch
action: shell unzip -oq {{ patch_dir }}/p18522509_112040_Linux-x86-64.zip -d {{ patch_dir }}
- name: patch conflict detection
action: shell export ORACLE_HOME={{ oracle_home }}; cd {{ patch_dir }}/18522509; $ORACLE_HOME/OPatch/opatch prereq CheckConflictAgainstOHWithDetail -ph ./
register: conflict_detection
failed_when: "'Prereq \"checkConflictAgainstOHWithDetail\" passed.' not in conflict_detection.stdout"
- name: apply psu
action: shell export ORACLE_HOME={{ oracle_home}}; cd {{ patch_dir }}/18522509; $ORACLE_HOME/OPatch/opatch apply -silent -ocmrf {{ patch_dir }}/ocm.rsp
register: apply_psu
failed_when: "'Composite patch 18522509 successfully applied.' not in apply_psu.stdout"
- name: clean up install directory
file: path={{ patch_dir }} state=absent
Let me run you through this playbook! It starts off with the indication of a YAML document: ‘—‘. Next hosts: all again. I just put all the hosts in the hosts file, I did not create all kinds of groups of hosts (which would be fitting when you use it at a fixed environment, but I use it for various customers). Then vars, with a list of variables. As you can see, I can use the variables, which are shown in the playbook as {{ variable }}. Then remote_user: oracle and tasks.
The first and second task use variables, and use the argument ‘register’ to save all response into a named variable. I also use ‘failed_when’ to make the playbook stop executing when the argument after ‘failed_when’ is true. Arguments of ‘failed_when’ is the named variable, for which the output of the standard out is used (.stdout). Then a filter is used to cast the output to integer, and is compared with a calculation of the variable.
The third task is using the files module to create a directory. The fourth task is using the copy module. The copy module means a file or files (in this case) are copied from the machine from which the playbook is run, onto the destination host or hosts. Here is also another trick used, to process the task with a list of items. As you can see, the copy line contains a variable {{ items }}, and the task is executed for all the items in the list ‘with_items’. I found this is fine for smaller files (up to a few hundred of megabytes), but too slow for bigger files. I use http (the get_url module) to speed up file transfer.
The fifth and sixth tasks execute a shell command, unzip, to extract the contents of a zip file into a specific place.
The seventh task is executing a small list of shell commands, in order to be able to run the conflict detection option of opatch. The same trick as with the first two tasks is used, register a name for the output of the conflict detection. Here I check if the stdout contains what I would manually check for when I would run it. The eighth task is the main task of the whole playbook: the actual patch. However, it uses the same technique as task seven. The last task simply removes a directory, in order to remove the files we used for this patch.
Summary
I hope this shows what a tremendous help Ansible can be for a consultant. This kind of tool is simply mandatory if you got an environment with more than approximately ten to twenty servers to administer. Ansible can be used even if the organisation does not want to spend time on the implementation of a configuration tool.
Like this:
Like Loading...
Frits Hoogland Weblog 