summaryrefslogtreecommitdiffstats
path: root/roles/openshift_logging/README.md
diff options
context:
space:
mode:
authorScott Dodson <sdodson@redhat.com>2017-01-25 08:35:16 -0500
committerGitHub <noreply@github.com>2017-01-25 08:35:16 -0500
commit223f3e1d99470c53dc5fdc92702bb1c17d18837a (patch)
treed2e5d6b986bbd516461b24ef592aa44114eba188 /roles/openshift_logging/README.md
parent800d0d9872807a3173ae268166d211c40fb34283 (diff)
parentd86bf1c52ca9e92679f6ad3ab4cc1315f067ff26 (diff)
downloadopenshift-223f3e1d99470c53dc5fdc92702bb1c17d18837a.tar.gz
openshift-223f3e1d99470c53dc5fdc92702bb1c17d18837a.tar.bz2
openshift-223f3e1d99470c53dc5fdc92702bb1c17d18837a.tar.xz
openshift-223f3e1d99470c53dc5fdc92702bb1c17d18837a.zip
Merge pull request #3169 from jcantrill/logging_node_selector
allow openshift_logging role to specify nodeSelectors
Diffstat (limited to 'roles/openshift_logging/README.md')
-rw-r--r--roles/openshift_logging/README.md3
1 files changed, 3 insertions, 0 deletions
diff --git a/roles/openshift_logging/README.md b/roles/openshift_logging/README.md
index 9b71dc676..856cfa2b9 100644
--- a/roles/openshift_logging/README.md
+++ b/roles/openshift_logging/README.md
@@ -35,6 +35,7 @@ When both `openshift_logging_install_logging` and `openshift_logging_upgrade_log
- `openshift_logging_curator_log_level`: The log level for the Curator process. Defaults to 'ERROR'.
- `openshift_logging_curator_cpu_limit`: The amount of CPU to allocate to Curator. Default is '100m'.
- `openshift_logging_curator_memory_limit`: The amount of memory to allocate to Curator. Unset if not specified.
+- `openshift_logging_curator_nodeselector`: A map of labels (e.g. {"node":"infra","region":"west"} to select the nodes where the curator pod will land.
- `openshift_logging_kibana_hostname`: The Kibana hostname. Defaults to 'kibana.example.com'.
- `openshift_logging_kibana_cpu_limit`: The amount of CPU to allocate to Kibana or unset if not specified.
@@ -43,6 +44,7 @@ When both `openshift_logging_install_logging` and `openshift_logging_upgrade_log
- `openshift_logging_kibana_proxy_cpu_limit`: The amount of CPU to allocate to Kibana proxy or unset if not specified.
- `openshift_logging_kibana_proxy_memory_limit`: The amount of memory to allocate to Kibana proxy or unset if not specified.
- `openshift_logging_kibana_replica_count`: The number of replicas Kibana should be scaled up to. Defaults to 1.
+- `openshift_logging_kibana_nodeselector`: A map of labels (e.g. {"node":"infra","region":"west"} to select the nodes where the pod will land.
- `openshift_logging_fluentd_nodeselector`: The node selector that the Fluentd daemonset uses to determine where to deploy to. Defaults to '"logging-infra-fluentd": "true"'.
- `openshift_logging_fluentd_cpu_limit`: The CPU limit for Fluentd pods. Defaults to '100m'.
@@ -67,6 +69,7 @@ When both `openshift_logging_install_logging` and `openshift_logging_upgrade_log
- `openshift_logging_es_pvc_prefix`: The prefix for the generated PVCs. Defaults to 'logging-es'.
- `openshift_logging_es_recover_after_time`: The amount of time ES will wait before it tries to recover. Defaults to '5m'.
- `openshift_logging_es_storage_group`: The storage group used for ES. Defaults to '65534'.
+- `openshift_logging_es_nodeselector`: A map of labels (e.g. {"node":"infra","region":"west"} to select the nodes where the pod will land.
When `openshift_logging_use_ops` is `True`, there are some additional vars. These work the
same as above for their non-ops counterparts, but apply to the OPS cluster instance: