feat: decommission playbooks

playbooks/meta/hadoop-decommission.yml

@@ -0,0 +1,11 @@
+# Copyright 2022 TOSIT.IO
+# SPDX-License-Identifier: Apache-2.0
+
+---
+# Add variable of the decommissioned node FQDN example:
+# ansible-playbook tdp/playbooks/meta/decommission.yml -e "excluded_node_fqdn=worker-01.novalocal"
+- ansible.builtin.import_playbook: ../utils/yarn_capacity_scheduler.yml
+- ansible.builtin.import_playbook: ../utils/yarn_resourcemanager_decomm_nodemanager.yml
+# Decommission Yarn nodemanager
+- ansible.builtin.import_playbook: ../utils/hdfs_namenode_decomm_datanode.yml
+# Decommission HDFS namenode

playbooks/meta/hbase-decommission.yaml

@@ -0,0 +1,8 @@
+# Copyright 2022 TOSIT.IO
+# SPDX-License-Identifier: Apache-2.0
+
+---
+# Add the host of the regionserver which should be decommissioned example:
+# ansible-playbook tdpplaybooks/meta/hbase-decommission.yaml -l worker-01
+- ansible.builtin.import_playbook: ../hbase_regionserver_stop.yml
+# Decommission HBase regionserver

playbooks/utils/hdfs_namenode_decomm_datanode.yml

@@ -0,0 +1,14 @@
+# Copyright 2022 TOSIT.IO
+# SPDX-License-Identifier: Apache-2.0
+
+---
+- name: Hadoop HDFS datanode Decommission
+  hosts: hdfs_nn
+  tasks:
+    - tosit.tdp.resolve: # noqa unnamed-task
+        node_name: hdfs_namenode
+    - name: Decommission HDFS datanode
+      ansible.builtin.import_role:
+        name: tosit.tdp.hdfs.namenode
+        tasks_from: decommission
+    - ansible.builtin.meta: clear_facts # noqa unnamed-task

playbooks/utils/yarn_resourcemanager_decomm_nodemanager.yml

@@ -0,0 +1,14 @@
+# Copyright 2022 TOSIT.IO
+# SPDX-License-Identifier: Apache-2.0
+
+---
+- name: Hadoop Yarn resourcemanager decommission
+  hosts: yarn_rm
+  tasks:
+    - tosit.tdp.resolve: # noqa unnamed-task
+        node_name: yarn_resourcemanager
+    - name: Decommision YARN NM
+      ansible.builtin.import_role:
+        name: tosit.tdp.yarn.resourcemanager
+        tasks_from: decommission
+    - ansible.builtin.meta: clear_facts # noqa unnamed-task

roles/hdfs/namenode/tasks/decommission.yml

@@ -0,0 +1,35 @@
+# Copyright 2022 TOSIT.IO
+# SPDX-License-Identifier: Apache-2.0
+
+---
+- name: Render dfs.exclude file
+  ansible.builtin.template:
+    src: dfs.exclude.j2
+    dest: "{{ hdfs_site['dfs.hosts.exclude'] }}"
+    owner: root
+    group: root
+    mode: "644"
+
+- name: Update exlude nodes file
+  ansible.builtin.lineinfile:
+    path: /etc/hadoop/conf.nn/dfs.exclude
+    line: "{{ item }}"
+    state: present
+  loop: "{{ [excluded_node_fqdn] }}"
+
+- name: kinit hdfs NN
+  ansible.builtin.command: kinit -kt /etc/security/keytabs/nn.service.keytab nn/{{ ansible_fqdn }}@{{ realm }}
+  become_user: hdfs
+
+- name: RefreshNodes
+  ansible.builtin.command: /usr/bin/hdfs dfsadmin -refreshNodes
+  become_user: hdfs
+
+- name: Check node status
+  ansible.builtin.command: hdfs dfsadmin -report -decommissioning
+  register: hdfs_output
+  become_user: hdfs
+
+- name: Print output of node status
+  ansible.builtin.debug:
+    var: hdfs_output.stdout

roles/yarn/resourcemanager/tasks/decommission.yml

@@ -0,0 +1,35 @@
+# Copyright 2022 TOSIT.IO
+# SPDX-License-Identifier: Apache-2.0
+
+---
+- name: Render yarn.exclude file
+  ansible.builtin.template:
+    src: yarn.exclude.j2
+    dest: "{{ yarn_site['yarn.resourcemanager.nodes.exclude-path'] }}"
+    owner: root
+    group: root
+    mode: "644"
+
+- name: Update exlude nodes file
+  ansible.builtin.lineinfile:
+    path: /etc/hadoop/conf.rm/yarn.exclude
+    line: "{{ item }}"
+    state: present
+  loop: "{{ [excluded_node_fqdn] }}"
+
+- name: kinit yarn RM
+  ansible.builtin.command: kinit -kt /etc/security/keytabs/rm.service.keytab rm/{{ ansible_fqdn }}@{{ realm }}
+  become_user: yarn
+
+- name: RefreshNodes
+  ansible.builtin.command: /usr/bin/yarn rmadmin -refreshNodes
+  become_user: yarn
+
+- name: Check node status
+  ansible.builtin.command: yarn node -list -all
+  register: yarn_output
+  become_user: yarn
+
+- name: Print output of node status
+  ansible.builtin.debug:
+    var: yarn_output.stdout