Haproxy Proxy MongoDB

The MongoDB cluster uses the replication set mode, at least 3 nodes or more. One question is, how do I let ceilometer automatically connect to the re-elected master node when the MongoDB master node is down?

Here are two solutions, as for which to look at the demand.

1, using VIP to locate the MongoDB master node

Use keepalived to start VIP, here are two vip,vrrp_instance vi_1 is dedicated to MongoDB vip,vrrp_instance vi_2 is OpenStack ha vip

VRRP script Chk_mongo_primary scripts specifically to detect whether the node is MONGO master node, if it is the weight of +2, and Vrrp_instance vi_1 VIP can be preempted;

When the MongoDB master node is switched on, Vrrp_instance Vi_1 's VIP will also switch.

keepalived.conf configuration file

! configuration file for keepalivedvrrp_script chk_haproxy {     script  "Killall -0 haproxy"     interval 2     weight 2}# Define the script used to check if mongod  is runningvrrp_script chk_mongod {    script  "killall -0  Mongod "    interval 2 # every two seconds     weight 2}# define the script to see if the local node  is the primaryvrrp_script chk_mongo_primary {    script  "MONGO  --eval  ' (!! Db.runcommand ("IsMaster") ["IsMaster"])? Quit (0): Quit (1) ' "    interval 2 #  every two seconds    weight 2}# configuation for the  Virtual interfacevrrp_instance VI_1 {    interface br-ex     state node master        # slave on the  other nodes    priority 101              # 100 on other nodes    virtual_ router_id 55    authentication {         auth_type pass        auth_pass secret      # Set this to some secret phrase    }     # the virtual ip address shared between the two  nodes    virtual_ipaddress {         172.16.140.251/24   &nBSP;}     # Use the script above to check if we  should fail over    track_script {         chk_mongod        chk_mongo_primary     }}vrrp_instance VI_2 {    interface br-ex     virtual_router_id 51    state backup    priority 200#  if use it,the openstack api do not response normally#     use_vmac virtualmac#    advert_int 1     dont_track_primary    nopreempt    authentication {     auth_type PASS    auth_pass password    }     virtual_ipaddress {       172.16.140.250/24    }     track_script {      chk_haproxy    }     NOTIFY /USR/LOCAL/BIN/KEEPALIVEDNOTIFY.SH}

2, the use of Haproxy Tcp-check to detect

Haproyx MongoDB Configuration

Listen mongodb_cluster    bind openstack_vip:27017    option  tcpka    option tcplog    option tcp-check     # MongoDB Wire Protocol    tcp-check send-binary  3a000000 # message length  (    tcp-check send-binary ) eeeeeeee # request id  (Random value)     tcp-check  send-binary 00000000 # response to  (Nothing)     tcp-check  send-binary d4070000 # opcode  (Query)     tcp-check send-binary  00000000 # Query Flags    tcp-check send-binary  61646d696e2e # fullcollectionname  (admin. $cmd)     tcp-check  send-binary 24636d6400 # continued    tcp-check send-binary 00000000 # numtoskip     Tcp-check send-binary ffffffff # numtoreturn    # start of  Document    tcp-check send-binary 13000000 # Document  length      tcp-check send-binary 10 # Type  (Int32)      tcp-check send-binary 69736d617374657200 # ismaster:     tcp-check send-binary 01000000 # Value : 1     Tcp-check send-binary 00 # term        tcp-check  expect binary 69736d61737465720001  #ismaster  true    server  controller1 controller1:27017 check inter 2000 rise 2 fall 3     server controlLer2 controller2:27017 check inter 2000 rise 2 fall 3 

Here's the problem, the non-master nodes that you see on the Haproxy monitoring page are down, because of Tcp-check's sake.

Why did Send-binary write that ? (with Tcpdum grab bag out, get Wireshark analysis)

For details, see here: https://blog.danman.eu/mongodb-haproxy/

Http://serverfault.com/questions/625492/how-to-construct-a-mongodb-health-check-in-haproxy

This article is from the "The-way-to-cloud" blog, make sure to keep this source http://iceyao.blog.51cto.com/9426658/1836237

Haproxy Proxy MongoDB