And probably not only the nodeselector, looks like bunch of ansible variables openshift_logging_es_ops_* are not picked up based on the component but always the openshift_logging_es_* ones.

https://github.com/openshift/openshift-ansible/blob/c515805a1ecfa980bc56af4eb58c5655fde6be44/roles/openshift_logging_elasticsearch/tasks/main.yaml#L271-L285

I will try to work on a fix

The only relevant config not honored is "openshift_logging_es_nodeselector".

Others not forwarded are "openshift_logging_es_ca_ext" and "openshift_logging_es_ops_storage_group" but they are not used anywhere. 

PR with a fix pending:
https://github.com/openshift/openshift-ansible/pull/5416

Commits pushed to master at https://github.com/openshift/openshift-ansible

https://github.com/openshift/openshift-ansible/commit/3566f8c0e8ec12995096b3ea9256266c252f3fb8
Bug 1491636 - honor openshift_logging_es_ops_nodeselector

https://github.com/openshift/openshift-ansible/commit/a01ee6767c673fe2ee144eb851a5488fcf40f2ec
Merge pull request #5416 from wozniakjan/bug1491636/honor_ops_nodeselector

Automatic merge from submit-queue

Bug 1491636 - honor openshift_logging_es_ops_nodeselector

https://bugzilla.redhat.com/show_bug.cgi?id=1491636

No nodeselector was configured in es-ops deploymentconfig. 

1) set variables
openshift_logging_es_nodeselector={'logging-node':'1'}
openshift_logging_es_ops_nodeselector={'logging-node':'2'}

2) check the nodeSelector
# oc get dc logging-es-data-master-8xc21lmr -o json |grep -A 2 nodeSelector
                "nodeSelector": {
                    "logging-node": "1"
                },
# oc get dc logging-es-ops-data-master-9iujvkxc -o json |grep -A 2 nodeSelector

I was using openshift-ansible-3.7.0-0.134.0.git.0.6f43fc3.el7.noarch for comment 5

The variable was not honored openshift_logging_es_ops_memory_limit=2Gi too.

There was another PR [1] to openshift-ansible setting certain variables as 'facts'  due to the deprecation of '*_hosted_*' vars. Among those is the openshift_logging_es_ops_nodeselector.

Due to ansible variable precedence, the only way to override facts is with passing --extra-var on the command line [2]. Setting openshift_logging_es_ops_nodeselector in the inventory no longer sets it sufficiently because it will be overwritten when ansible sets the fact.

Currently, for the testing purposes, you can run the playbook as follows:
$ ansible-playbook playbooks/byo/openshift-cluster/openshift-logging.yml --extra-vars "openshift_logging_es_ops_nodeselector={'ns_key':'ns_val'}"


Maybe ewolinetz or sdodson could explain what is the rationale behind the facts setting and what is the expected behavior there?



[1] https://github.com/openshift/openshift-ansible/pull/5176

[2] http://docs.ansible.com/ansible/latest/playbooks_variables.html#variable-precedence-where-should-i-put-a-variable

Moving this to on_qa.  Per comment #6 shouldnt we be using a v3.6 of openshift-ansible instead of 3.7?

Jeff, Yes, the issue was pull in due to the deprecation of '*_hosted_*' vars . The openshift_logging_es_ops_nodeselector works with the released openshift-ansible-v3.6.173.0.21-17.

It still won't work with latest available puddle pacakge openshift-ansible-3.6.173.0.45 even with --extra-vars  "openshift_logging_es_ops_nodeselector={'logging-node':'2'}


Test Result:

With latest v3.6 packages openshift-ansible-3.6.173.0.45. The nodeSelector still use values from openshift_logging_es_nodeselector.

ansible-playbook /usr/share/ansible/openshift-ansible/playbooks/byo/openshift-cluster/openshift-logging.yml --extra-vars  "openshift_logging_es_ops_nodeselector={'logging-node':'2'}"


$ oc get dc logging-es-ops-data-master-15qoianl -o json |grep -A 1 Select
                "nodeSelector": {
                    "logging-node": "1"

(In reply to Jeff Cantrill from comment #9)
> Moving this to on_qa.  Per comment #6 shouldnt we be using a v3.6 of
> openshift-ansible instead of 3.7?

I had expected there is fix in v3.7. so I tested with v3.7 packages.

(In reply to Anping Li from comment #10)
> Jeff, Yes, the issue was pull in due to the deprecation of '*_hosted_*' vars
> . The openshift_logging_es_ops_nodeselector works with the released
> openshift-ansible-v3.6.173.0.21-17.

To clarify, the *_hosted_* vars are not actually deprecated yet, they are still used. However we will warn users if they are being set in v3.7

Also, we only warn about variables that would have 'hosted' in the variable name. So openshift_logging_es_ops_nodeselector should not be impacted by deprecated vars. 




How are you setting the openshift_logging_es_nodeselector? If it is with -e on the command line, that will overwrite the openshift_logging role trying to set it to openshift_logging_es_ops_nodeselector when calling openshift_logging_elasticsearch for an ops install.

Forgive me use v3.7 as example. The v3.7 have same problem. I set the inventory variable in inventory file as following. 


openshift_logging_install_logging=true.
openshift_logging_image_version=v3.7
openshift_logging_namespace=logging
openshift_logging_es_cluster_size=1
openshift_logging_es_memory_limit=2Gi
openshift_logging_es_nodeselector={'logging-node': '1'}

openshift_logging_use_ops=true
openshift_logging_es_ops_cluster_size=1
openshift_logging_es_ops_nodeselector={'logging-node':'2'}

Created attachment 1342209 [details]
The inventory file I used

created two more PRs to unify behavior between master and 3.6

1) https://github.com/openshift/openshift-ansible/pull/5857 - remove setting of nodeselector as 'fact' to allow overriding default value in the inventory

2) https://github.com/openshift/openshift-ansible/pull/5858 - adding missing statement forwarding nodeselector to the DC

Commits pushed to master at https://github.com/openshift/openshift-ansible

https://github.com/openshift/openshift-ansible/commit/1d86e3a156dcf7950d1b2c4877d82f3a183b2033
Bug 1491636 - honor node selectors

The deprecation of `*_hosted_*` vars made logging node selectors set in
the inventory to be ignored.

Node selectors were set as 'facts' and they have higher priority than
inventory variables. Setting logging node selectors could be achieved only
by using command line --extra-vars.

https://github.com/openshift/openshift-ansible/commit/3dea8bceeb98d5a30f0223f7109cc78caa3a6b8f
Merge pull request #5857 from wozniakjan/bz_1491636/logging/nodeselector

Automatic merge from submit-queue.

Bug 1491636 - honor node selectors

The deprecation of `*_hosted_*` vars made logging node selectors set in the inventory to be ignored.

Node selectors were set as 'facts' and they have higher priority than inventory variables. Setting logging node selectors could be achieved only by using command line --extra-vars.

It is related to the same BZ as https://github.com/openshift/openshift-ansible/pull/5858 but the issue is different.

The fix pass testing on openshift-ansible-3.6.173.0.65-1.git.0.63cbf23.el7.noarch


openshift_logging_install_logging=true
openshift_logging_image_version=v3.6
openshift_logging_es_nodeselector={'logging-node': '1'}
openshift_logging_use_ops=true
openshift_logging_es_ops_nodeselector={'logging-node':'2'}


# oc get dc  --selector="logging-infra"="elasticsearch" -o json |grep -A 2 Selector 
                        "nodeSelector": {
                            "logging-node": "1"
                        },
--
                        "nodeSelector": {
                            "logging-node": "2"
                        },

@Samuel, The fix in in openshift-ansible only. Could we move this bug to the installer errata?

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHBA-2017:3255