release-4.0: Support Inject cluster topology information for am and g…

…rafana to etcd in pd (#1196)

* Support Inject cluster topology information for am and grafana to etcd in pd (#1124)

* [Init] Initialize pr and create scripts for sending request to pd

* [ADD] adding scripts to yml

* [Replace] replacing with shell arguments

* [Try] trying to testing ansible

* [Update] Updating script

* update

* [FIX] fix wrong api

* Add deploy path

* [ADD] adding another logic loading

* [ADD] adding net/deploy

* add all pd info

* [Update] Change directory name and update etcd write path.

Signed-off-by: mapleFU <1506118561@qq.com>

* change init-pd-topo to dashboart topo

* [Update] Support `port` field.

Signed-off-by: mapleFU <1506118561@qq.com>

* [Fix] Fixing errors on scripts.

Signed-off-by: mapleFU <1506118561@qq.com>

* [Final] Trying to fix all the problems

Signed-off-by: mapleFU <1506118561@qq.com>

* [Update] update port type.

Signed-off-by: mapleFU <1506118561@qq.com>

* [FIX] fix bugs for tidb-anisble deploy dir

Signed-off-by: mapleFU <1506118561@qq.com>

* [Update] Wait until testing.

Signed-off-by: mapleFU <1506118561@qq.com>

* [Revert] Revert unnescessary changes.

Signed-off-by: mapleFU <1506118561@qq.com>

* [DONE] Finish dev.

Signed-off-by: mapleFU <1506118561@qq.com>

* Update start.yml

* Update start.yml

Co-authored-by: liubo <liubo0127@qq.com>

* Support Prometheus in for dashboard topology (#1195)

* [Support] change tidb-ansible script in tidb-ansible.

Signed-off-by: mapleFU <1506118561@qq.com>

* [ADD] Adding supports in roles.

Signed-off-by: mapleFU <1506118561@qq.com>

* [Multiple] pass the first when there is multiple arguments.

Co-authored-by: mwish <1506118561@qq.com>

deploy.yml

@@ -186,3 +186,10 @@
     - always
   roles:
     - perf_tools
+
+- name: deploying init pd topo scripts
+  hosts: localhost
+  tags:
+    - always
+  roles:
+    - dashboard_topo

excessive_rolling_update.yml

@@ -511,6 +511,12 @@
       delay: 5
       when: enable_tls|default(false)
 
+- hosts: localhost
+  tags:
+    - always
+  tasks:
+    - name: init pd topo
+      shell: "cd {{ playbook_dir }}/scripts && ./dashboard_topo.sh"
 
 - name: rolling update TiFlash cluster
   hosts: tiflash_servers

roles/dashboard_topo/tasks/main.yml

@@ -0,0 +1,7 @@
+---
+
+- name: generate init_dashboard_topo script
+  template:
+    src: "init_dashboard_topo.sh.j2"
+    dest: "{{ playbook_dir }}/scripts/dashboard_topo.sh"
+    mode: "0755"

roles/dashboard_topo/templates/init_dashboard_topo.sh.j2

@@ -0,0 +1,47 @@
+#!/bin/bash
+set -e
+
+{% set all_pd = [] -%}
+{% set pd_hosts = groups.pd_servers %}
+{% for host in pd_hosts -%}
+  {% set pd_ip = hostvars[host].ansible_host | default(hostvars[host].inventory_hostname) -%}
+  {% set pd_port = hostvars[host].pd_client_port -%}
+  {% set pd_path = hostvars[host].deploy_dir -%}
+  {% set _ = all_pd.append("%s:%s%s" % (pd_ip, pd_port, pd_path)) -%}
+{% endfor -%}
+{% set all_grafana = [] -%}
+{% set grafana_hosts = groups.grafana_servers %}
+{% for host in grafana_hosts -%}
+  {% set grafana_ip = hostvars[host].ansible_host | default(hostvars[host].inventory_hostname) -%}
+  {% set grafana_port = hostvars[host].grafana_port -%}
+  {% set grafana_path = hostvars[host].deploy_dir -%}
+  {% set _ = all_grafana.append("%s:%s%s" % (grafana_ip, grafana_port, grafana_path)) -%}
+{% endfor -%}
+{% set all_alertmanager = [] -%}
+{% set alertmanager_hosts = groups.alertmanager_servers %}
+{% for host in alertmanager_hosts -%}
+  {% set alertmanager_ip = hostvars[host].ansible_host | default(hostvars[host].inventory_hostname) -%}
+  {% set alertmanager_port = hostvars[host].alertmanager_port -%}
+  {% set alertmanager_path = hostvars[host].deploy_dir -%}
+  {% set _ = all_alertmanager.append("%s:%s%s" % (alertmanager_ip, alertmanager_port, alertmanager_path)) -%}
+{% endfor -%}
+{% set all_prometheus = [] -%}
+{% set prometheus_hosts = groups.monitoring_servers %}
+{% for host in prometheus_hosts -%}
+  {% set prometheus_ip = hostvars[host].ansible_host | default(hostvars[host].inventory_hostname) -%}
+  {% set prometheus_port = hostvars[host].prometheus_port -%}
+  {% set prometheus_path = hostvars[host].deploy_dir -%}
+  {% set _ = all_prometheus.append("%s:%s%s" % (prometheus_ip, prometheus_port, prometheus_path)) -%}
+{% endfor -%}
+{% set flag = "" %}
+{% if all_grafana -%}
+  {% set flag = flag + " --grafana " + ','.join(all_grafana) -%}
+{% endif -%}
+{% if all_alertmanager -%}
+  {% set flag = flag + " --alertmanager " + ','.join(all_alertmanager) -%}
+{% endif -%}
+{% if all_prometheus %}
+  {% set flag = flag + " --prometheus " + ','.join(all_prometheus) -%}
+{% endif -%}
+
+python2 dashboard_topo.py --pd {{ all_pd | join(',') }} {{ flag }}

rolling_update.yml

@@ -542,6 +542,12 @@
       delay: 5
       when: enable_tls|default(false)
 
+- hosts: localhost
+  tags:
+    - always
+  tasks:
+    - name: init pd topo
+      shell: "cd {{ playbook_dir }}/scripts && ./dashboard_topo.sh"
 
 - name: rolling update TiFlash cluster
   hosts: tiflash_servers

scripts/dashboard_topo.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python2
+
+from __future__ import print_function, \
+    unicode_literals
+
+import urllib
+import urllib2
+import base64
+import json
+import argparse
+
+ComponentToRegister = ('alertmanager', 'grafana', 'pd', 'prometheus')
+
+
+def parse_opts():
+    """
+    parse_opts parse the input of involved components and pd address.
+    """
+    parser = argparse.ArgumentParser(description="Parse output.")
+    # pd is involved because we need to send http request
+    for target in ComponentToRegister:
+        parser.add_argument("--{}".format(target),
+                            help="the address list of {}".format(target))
+    args, unknown = parser.parse_known_args()
+    return args
+
+
+def etcd_write(etcd_url, key, value):
+    encoded_key = base64.b64encode(key)
+    encoded_value = base64.b64encode(value)
+    data = json.dumps({
+        "key": encoded_key,
+        "value": encoded_value,
+    })
+    req = urllib2.Request('http://' + etcd_url + '/v3/kv/put',
+                          data=data,
+                          headers={'Content-Type': 'application/json'})
+    try:
+        resp = urllib2.urlopen(req)
+        data = json.load(resp)
+        return data
+    except urllib2.HTTPError as error:
+        raise error
+
+
+def parse_address(con):
+    """
+    con: str for argument like "127.0.0.1:2379/deploy"
+    return: Tuple[str, str] like ("127.0.0.1:2379", "/deploy")
+    """
+    pos = con.find('/')
+    return (con[:pos], con[pos:])
+
+
+def request_topo(comp, topo, etcd_target):
+    """
+    Sending request to etcd v3, and leave:
+    under {pd_target}:
+    write: /topology/{comp}: {ip: ip, address: address}
+
+    comp: str for component name, which will be like "tidb"
+    topo: str for topology address, like "127.0.0.1:4000"
+    pd_target: the place to send etcd request, like "127.0.0.1:2379"
+    """
+    if topo is None:
+        # if topo is None, do nothing
+        return
+    if ',' in topo:
+        topo = topo.split(',')[0]
+    ip, add = parse_address(topo)
+    ip, port = ip.split(':')
+
+    message = json.dumps({
+        'ip': ip,
+        'binary_path': add,
+        'port': int(port),
+    })
+    etcd_write(etcd_target, "/topology/" + comp, message)
+
+
+def concat_to_address(ip, port):
+    """
+    ip: str for address to concat, like "127.0.0.1"
+    port: str for port, like "2379"
+
+    return: str like "127.0.0.1:2379"
+        return None if ip or port is None
+    """
+    if ip is None or port is None:
+        return None
+    return ip.strip() + ":" + port.strip()
+
+
+if __name__ == '__main__':
+    args = parse_opts()
+
+    # parse from args
+    pd_address = args.pd
+    pd_address_zero, _ = parse_address(pd_address.split(',')[0])
+
+    alertmanager_address = args.alertmanager
+    grafana_address = args.grafana
+    prometheus_address = args.prometheus
+
+    mapping = {
+        'alertmanager': alertmanager_address,
+        'grafana': grafana_address,
+        'prometheus': prometheus_address,
+    }
+
+    for comp in ComponentToRegister:
+        if comp == 'pd':
+            continue
+        request_topo(comp, mapping[comp], pd_address_zero)

start.yml

@@ -559,3 +559,10 @@
       vars:
         - ansible_become: false
         - ansible_connection: local
+
+- hosts: localhost
+  tags:
+    - always
+  tasks:
+    - name: init dashboard topology
+      shell: "cd {{ playbook_dir }}/scripts && ./dashboard_topo.sh"