Ansible patching order/inventory logic for productive environments

user_16834498 0
Viewed 240

It's less a technical question, but maybe Ansible has features that would help, here, that I don't know yet. I'm able to automate patching with Ansible, but choosing the right hosts/groups in the right order is complicated, I'll try to explain it.

Lets take this inventory for example:

---
all:
  dcs:
    hosts:
      domaincontroller1
      domaincontroller2
  dbs:
    hosts:
      sql1
      sql2
  webservers:
    hosts:
      websrv1 #has a mysql connection and services vars
      websrv2
      websrv3 #has a mysql connection and services vars
      websrv4

So what you do on a patch day? You want that at least one domain controller is running every time. You want that all webservers that connect to sql are down, or their services are stopped, after that you first patch the sql servers, wait until they are running again, patch the web servers and wait until they connect to sql.

At the moment, I split the host file into two groups. First group is one DC and all servers that don't connect to sql. The second group contains sql1, sql2, webserver1, ... and there is a different playbook that patches the first 2 in the row first, and all other after that. But when doing this, I have an ugly/unsorted hosts file and I'm unable to apply changes to all web servers for example.

---
all:
  patch1:
    hosts:
      domaincontroller1
      websrv2
      websrv4
  patch2:
    hosts:
      sql1
      sql2
      domaincontroller2
      websrv1 #has a mysql connection and services vars
      websrv3 #has a mysql connection and services vars

How others do that? Is there a way to split groups in half, so there is a patch1 group that contains 50% of the DCs and all web servers where no services are defined (probably with dynamic groups?). Otherwise I would need to create the perfect grouped inventory and add groups for patch day 1 and 2 underneath it, which results in having one server multiple times in the same inventory what makes changes more complicated.

Another idea would be the use of tags, like patchfirst, patchsecond, and create for any server a host_vars file which is again pretty much work for about 100 hosts. Anyone ideas or examples how to get the best looking, best working result without making more work as manual patching would need?

ansible
1 Answers

Here is a, probably naive, approach to achieve this.

To start with, you can create the groups in your inventory, but without hosts on them.

Notes:

  • I fixed some issues in your inventory, that was missing the children keyword, and where the hosts should have been YAML keys, so, ending with colon (:).
  • I added a sql_connection variable on some web servers, to craft the logic that elects web servers in the same patch day as the databases or not.
    Adapt it to the existing variables definied in your hosts.
  • I also added some extra hosts, just for demonstration purpose.

So, we have an inventory that looks like:

all:
  children:
    patch_day_1:
    patch_day_2:
    dcs:
      hosts:
        domaincontroller1:
        domaincontroller2:
    dbs:
      hosts:
        sql1:
        sql2:
        sql3:
        sql4:
        sql5:
    webservers:
      hosts:
        websrv1:
          sql_connection: mysql://some_connection_string
        websrv2:
        websrv3:
          sql_connection: mysql://some_connection_string
        websrv4:
        websrv5:

Then:

  • We elect the domain controllers, with the help of the array slicing notation.
  • We elect the web servers based on the fact that sql_connection is defined or not in the hostvars.
  • We elect all databases in the group containing the web servers defining a sql_connection.
  • Last, but not least, we run the playbook with the help of an extra-vars.

Here would be the playbook doing this:

- name: Logically split hosts
  hosts: localhost
  gather_facts: no

  tasks:
    - name: Elect half of the controllers in the first group
      add_host:
        name: "{{ item }}"
        groups: patch_day_1
      loop: "{{ groups.dcs[0:(groups.dcs | length / 2) | int] }}"

    - name: Elect all the web servers that do not have a `sql_connection` in the first group
      add_host:
        name: "{{ item }}"
        groups: patch_day_1
      loop: "{{ groups.webservers }}"
      when: hostvars[item].sql_connection is not defined

    - name: Elect the other half of the controllers in the second group
      add_host:
        name: "{{ item }}"
        groups: patch_day_2
      loop: "{{ groups.dcs[(groups.dcs | length / 2) | int:groups.dcs | length] }}"

    - name: Elect all the databases in the second group
      add_host:
        name: "{{ item }}"
        groups: patch_day_2
      loop: "{{ groups.dbs }}"

    - name: Elect all the web servers that do have a `sql_connection` in the second group
      add_host:
        name: "{{ item }}"
        groups: patch_day_2
      loop: "{{ groups.webservers }}"
      when: hostvars[item].sql_connection is defined


- name: Patch execution
  hosts: "patch_day_{{ patch_day | default(1) }}"
  gather_facts: no

  tasks:
    - name: Display hosts targeted in the play
      debug:
        var: ansible_play_hosts_all
      run_once: true

This, run with no extra-vars would default to the patch_day_1 group, yielding the result:

PLAY [Logically split hosts] *************************************************************************************

TASK [Elect half of the controllers in the first group] **********************************************************
ok: [localhost] => (item=domaincontroller1)
ok: [localhost] => (item=domaincontroller2)

TASK [Elect all the web servers that do not have a `sql_connection` in the first group] **************************
skipping: [localhost] => (item=websrv1) 
ok: [localhost] => (item=websrv2)
skipping: [localhost] => (item=websrv3) 
ok: [localhost] => (item=websrv4)
ok: [localhost] => (item=websrv5)

TASK [Elect the other half of the controllers in the second group] ***********************************************
ok: [localhost] => (item=domaincontroller3)
ok: [localhost] => (item=domaincontroller4)
ok: [localhost] => (item=domaincontroller5)

TASK [Elect all the databases in the second group] ***************************************************************
ok: [localhost] => (item=sql1)
ok: [localhost] => (item=sql2)
ok: [localhost] => (item=sql3)
ok: [localhost] => (item=sql4)
ok: [localhost] => (item=sql5)

TASK [Elect all the web servers that do have a `sql_connection` in the second group] *****************************
ok: [localhost] => (item=websrv1)
skipping: [localhost] => (item=websrv2) 
ok: [localhost] => (item=websrv3)
skipping: [localhost] => (item=websrv4) 
skipping: [localhost] => (item=websrv5) 

PLAY [Patch execution] *******************************************************************************************

TASK [Display hosts targeted in the play] ************************************************************************
ok: [domaincontroller1] => 
  ansible_play_hosts_all:
  - domaincontroller1
  - domaincontroller2
  - websrv2
  - websrv4
  - websrv5

PLAY RECAP *******************************************************************************************************
domaincontroller1          : ok=1    changed=0    unreachable=0    failed=0    skipped=0    rescued=0    ignored=0   
localhost                  : ok=6    changed=0    unreachable=0    failed=0    skipped=0    rescued=0    ignored=0

But, run with an extra-vars, then we can target the patch_day_2.
So, running it with the command

ansible-playbook play.yml --extra-vars="patch_day=2"

This yields the recap:

PLAY [Logically split hosts] *************************************************************************************

TASK [Elect half of the controllers in the first group] **********************************************************
ok: [localhost] => (item=domaincontroller1)
ok: [localhost] => (item=domaincontroller2)

TASK [Elect all the web servers that do not have a `sql_connection` in the first group] **************************
skipping: [localhost] => (item=websrv1) 
ok: [localhost] => (item=websrv2)
skipping: [localhost] => (item=websrv3) 
ok: [localhost] => (item=websrv4)
ok: [localhost] => (item=websrv5)

TASK [Elect the other half of the controllers in the second group] ***********************************************
ok: [localhost] => (item=domaincontroller3)
ok: [localhost] => (item=domaincontroller4)
ok: [localhost] => (item=domaincontroller5)

TASK [Elect all the databases in the second group] ***************************************************************
ok: [localhost] => (item=sql1)
ok: [localhost] => (item=sql2)
ok: [localhost] => (item=sql3)
ok: [localhost] => (item=sql4)
ok: [localhost] => (item=sql5)

TASK [Elect all the web servers that do have a `sql_connection` in the second group] *****************************
ok: [localhost] => (item=websrv1)
skipping: [localhost] => (item=websrv2) 
ok: [localhost] => (item=websrv3)
skipping: [localhost] => (item=websrv4) 
skipping: [localhost] => (item=websrv5) 

PLAY [Patch execution] *******************************************************************************************

TASK [Display hosts targeted in the play] ************************************************************************
ok: [domaincontroller3] => 
  ansible_play_hosts_all:
  - domaincontroller3
  - domaincontroller4
  - domaincontroller5
  - sql1
  - sql2
  - sql3
  - sql4
  - sql5
  - websrv1
  - websrv3

PLAY RECAP *******************************************************************************************************
domaincontroller3          : ok=1    changed=0    unreachable=0    failed=0    skipped=0    rescued=0    ignored=0   
localhost                  : ok=6    changed=0    unreachable=0    failed=0    skipped=0    rescued=0    ignored=0
user_21235300
Related