Running mongodb as a replicaSet in Docker (and adding a new SECONDARY and then upgrading from 3.0 to 3.4)

This is a continuation of the previous article on how to run mongodb in docker in a replica set

We start off by a mongodb cluster of two nodes, running in a docker setup like this:

docker-compose.yml

version: '3'
services:
  db01:
    image: mongo:3.0
    volumes:
    - datadb01:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30001:30001"
    command: ["mongod", "--config", "/etc/mongod.conf", "--port","30001"]
    container_name: db01

  db02:
    image: mongo:3.0
    volumes:
    - datadb02:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30002:30002"
    command: ["mongod", "--config", "/etc/mongod.conf", "--port","30002"]
    container_name: db02

volumes:
  datadb01:
  datadb02:

Step Three – Add another host to the replication set

Now adding a third one to the config seems straight forward:

version: '3'
services:
  db01:
    image: mongo:3.0
    volumes:
    - datadb01:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30001:30001"
    command: ["mongod", "--config", "/etc/mongod.conf", "--port","30001"]
    container_name: db01

  db02:
    image: mongo:3.0
    volumes:
    - datadb02:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30002:30002"
    command: ["mongod", "--config", "/etc/mongod.conf", "--port","30002"]
    container_name: db02

  db03:
    image: mongo:3.0
    volumes:
    - datadb03:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30003:30003"
    command: ["mongod", "--config", "/etc/mongod.conf", "--port","30003"]
    container_name: db03

volumes:
  datadb01:
  datadb02:
  datadb03:

To add this third machine to the replSet, reconfiguration in the mongo shell is required:

Seems to be as easy as rs.add("db03:30003") (with MongoDB version >=3.0)

A rs.status() check reveals that the third server is part of the cluster.

It stayed in startup states for a little bit of time (no transactions going on in this test environment)…


                {
                        "_id" : 3,
                        "name" : "db03:30003",
                        "health" : 1,
                        "state" : 5,
                        "stateStr" : "STARTUP2",
                        "uptime" : 12,
                }

… but finally managed to start up completely:

{
        "set" : "rs0",
        "date" : ISODate("2017-06-23T10:30:02.225Z"),
        "myState" : 1,
        "members" : [
                {
                        "_id" : 1,
                        "name" : "db01:30001",
                        "health" : 1,
                        "state" : 1,
                        "stateStr" : "PRIMARY",
...
                        "self" : true
                },
                {
                        "_id" : 2,
                        "name" : "db02:30002",
                        "health" : 1,
                        "state" : 2,
                        "stateStr" : "SECONDARY",
...
                        "pingMs" : 0,
                        "configVersion" : 139230
                },
                {
                        "_id" : 3,
                        "name" : "db03:30003",
                        "health" : 1,
                        "state" : 2,
                        "stateStr" : "SECONDARY",
...
                        "pingMs" : 0,
                        "configVersion" : 139230
                }
        ],
        "ok" : 1
}

We have a running cluster on mongodb version 3.0

So, next stop: Update the cluster from 3.0 to 3.4.

Step four and five – update 3.0 to 3.2 to 3.4

((WAIT, you might want to update your applications configuration now as well, see below))

According to the official mongo documentation, this needs to be done in two steps:

3.0 to 3.2: https://docs.mongodb.com/manual/release-notes/3.2-upgrade/#upgrade-a-replica-set-to-3-2

3.2 to 3.4: https://docs.mongodb.com/manual/release-notes/3.4-upgrade-replica-set/#upgrade-replica-set

In both cases, the steps seem to be the same and quite straight forward:

  1. Upgrade secondary members of the replica set
  2. Step down the replica set primary to secondary, so an upgraded one becomes primary
  3. Upgrade the previous primary so all are on the same version

In this case, using docker, the upgrades of the instances should be as easy as changing the version tag in the docker-compose.yml.
So, one at a time:
As my current primary is db01, I’ll start with db02. The change is just a version number in the file, so I’m not pasting the whole file here:

  db02:
    image: mongo:3.2

A docker-compose up -d brought db02 down, replacing it with an updated mongod 3.2 and repeating and watching rs.status(), I could see the machine disapear and the re-sync.
NICE
Repeat it for db03
NICE again

Next step – step down
Running rs.stepDown on the PRIMARY db01 makes db03 turn PRIMARY and leaves db01 a SECONDARY, ready to update to 3.2 as well…

BUT WAIT!

This made me aware of the fact that I forgot to update my application configuration. While I extended the cluster to a 3-host-system, I did not add db03 to the applications mongo server config and the application server’s /etc/host – which I quickly changed at this point.

Changing the db01’s image to 3.2 now and running docker-compose up -d did update the image/container and restart it – but rs.status() made me also aware that – according to their uptime – the other instances seem to have been restarted as well.

So, there must be a way to update/restart single services of docker-compose, right? Let’s check during the upgrade from 3.2 to 3.4

Now that all 3 containers are running the 3.2 image, the SECONDARYs can be updated as well. The line changed in the docker-compose.yml:

version: '3'
services:
  db01:
    image: mongo:3.4
    ...

Now, instead of running a full docker-compose up -d, it seems the way to go is

docker-compose stop db02
docker-compose create db02
docker-compose start db02

A previous docker-compose up -d db01 had an effect on the other servers uptimes as well, so I verified with db02 that this works.

After connection with the mongo shell to the PRIMARY (db03) and sending it a rs.stepDown(), this one is ready to be upgraded as well.

With the stop, create, start sequence, the last container is upgraded to 3.4 as well and the exercise is finished.

Running mongodb as a replicaSet in Docker (and upgrading it from 3.0 to 3.4)

Goal:

This post is about two things in one go:

Prerequesits (or what I have and use for my case):

  •  a virtual machine running docker
  • an application conneting to and using mongodb
  • a db dump for the application

Step One – Configure and boot a single mongo server with docker

This is influenced by the very good article Creating a MongoDB replica set using Docker.

It gives details about the basics (set up containers, start a replica set). My goal is to go a little bit furhter, though. In addtion to what the article suggests, I’d like to have the data of each container in a data volume. And I’d like to use docker-compose to keep the whole setup in order.

Using the version 3 syntax of docker-compose, I come up with a very basic initial file to start from:

version: '3'
services:
  db01:
    image: mongo:3.0
    volumes:
    - datadb01:/data/db
    ports:
    - "30001:27017"

volumes:
  datadb01:

What it does:

  • Use version 3 of the compose syntax
  • define a first db service, based on a mongo image with the version tag 3.0
  • expose the image’s port 27017 on the host as port 30001
  • mount a named data volume datadb01 into the container at /data/db (the default path of MongoDBs data storage)

This can be run with docker-compose up -d and new we have a single instance mongodb running on port 30001, accessible from the outside.

Step Two – Extend the single mongodb instance to become a multi-host replica set

Adding a second host to the configuration is straight forward and requires some copy & paste so the docker-compose.yml file looks like this

version: '3'
services:
  db01:
    image: mongo:3.0
    volumes:
    - datadb01:/data/db
    ports:
    - "30001:27017"
  db02:
    image: mongo:3.0
    volumes:
    - datadb02:/data/db
    ports:
    - "30002:27017"

volumes:
  datadb01:
  datadb02:

To check if the machine is up, I connect to the second mongod instance from another machine with mongo --port 30002. Of course, this is – as of right now – only a separate single instance of mongod and not a replicaSet, as confirmed by a quick check of the replication status:


> rs.status()
{ "ok" : 0, "errmsg" : "not running with --replSet", "code" : 76 }

At this point, I decided to make this another mongo exercise and start the replicaSet with only two servers, import my data, and only later add the third machine.

So, to get this dual-setup running, we need to tell the machines what replicaSet they are part of. This can be done with a command line option on mongod (--replSet), but I wanted to make it more versatile and put some options into a config file for mongo and start the daemon by telling it where to pull the config from.

So, in a subfolder etc, the simple config file etc/mongod.conf is created:

replication:
   oplogSizeMB: 400
   replSetName: rs0

(the oplog size is a random number here and should be correctly adjusted in production environments)

Now we need to map this file into the containers and tell mongod to read it during startup:

version: '3'
services:
  db01:
    image: mongo:3.0
    volumes:
    - datadb01:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30001:27017"
    command: ["mongod", "--config", "/etc/mongod.conf"]

  db02:
    image: mongo:3.0
    volumes:
    - datadb02:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30002:27017"
    command: ["mongod", "--config", "/etc/mongod.conf"]

volumes:
  datadb01:
  datadb02:

What we have here now in addition:

  • copy the file /etc/mongod.conf to /etc/mongod.conf in the container
  • start the container with the additional options, resulting in mongod --conf /etc/mongod.conf

Until now, I was under the impression I could spin up a mongo cluster just like that, but some research and this question on stack overflow made me aware that it won’t work withough a little bit of shell command line.

So, let’s init the replSet

To get the set working, we need to define a config in the mongo shell, for exaple like this:

> rs.initiate({
  "_id": "rs0",
  "version": 1,
  "members" : [
   {"_id": 1, "host": "db01:27017"},
   {"_id": 2, "host": "db02:27017"}
  ]
 })

(Note: as the machines connect internally, the internal ports 27017 need to be used, not the exposed ones)

However, to make this work, the containers need to be known as db01 and db02. They autom automatically got a generated name by docker-compose. So the names have to be added in the docker-compose file to be manually set:

version: '3'
services:
  db01:
    image: mongo:3.0
    volumes:
    - datadb01:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30001:27017"
    command: ["mongod", "--config", "/etc/mongod.conf"]
    container_name: db01

  db02:
    image: mongo:3.0
    volumes:
    - datadb02:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30002:27017"
    command: ["mongod", "--config", "/etc/mongod.conf"]
    container_name: db02

volumes:
  datadb01:
  datadb02:

After another docker-compose up -d, the config above can be initialized and results in a happy replication set:

> rs.status()
{
        "set" : "rs0",
        "date" : ISODate("2017-06-21T14:38:13.720Z"),
        "myState" : 2,
        "members" : [
                {
                        "_id" : 1,
                        "name" : "db01:27017",
                        "health" : 1,
                        "state" : 1,
                        "stateStr" : "PRIMARY",
                        "uptime" : 161,
                        "optime" : Timestamp(1498055732, 1),
                        "optimeDate" : ISODate("2017-06-21T14:35:32Z"),
                        "lastHeartbeat" : ISODate("2017-06-21T14:38:12.384Z"),
                        "lastHeartbeatRecv" : ISODate("2017-06-21T14:38:12.384Z"),
                        "pingMs" : 0,
                        "electionTime" : Timestamp(1498055736, 1),
                        "electionDate" : ISODate("2017-06-21T14:35:36Z"),
                        "configVersion" : 1
                },
                {
                        "_id" : 2,
                        "name" : "db02:27017",
                        "health" : 1,
                        "state" : 2,
                        "stateStr" : "SECONDARY",
                        "uptime" : 216,
                        "optime" : Timestamp(1498055732, 1),
                        "optimeDate" : ISODate("2017-06-21T14:35:32Z"),
                        "configVersion" : 1,
                        "self" : true
                }
        ],
        "ok" : 1
}

Now it’s time to import some data into the cluster and try to connect my existing application to the new cluster.
It should be noted here that the application is NOT running as part of the docker setup but is intended to connect to the ports exposed.

quick break, have some coffee while we wait for mongoimport to finish

Importing the data to the cluster with a mongoimport shell script on the primary server is not a but problem, but my PHP and old \MongoClient based application seems to have a problem:

MongoConnectionException
No candidate servers found

MongoConnectionException
MongoClient::__construct(): php_network_getaddresses: getaddrinfo failed: Name or service not known

Looks like that fact that using different IPs and ports “on the outside” (the configuration exposed by docker) is not good enough for the php mongo driver.
To circumvent this, let’s try to match internal and external configurations:

First, match up internal and external mongod ports by changing the internal ones:

version: '3'
services:
  db01:
    image: mongo:3.0
    volumes:
    - datadb01:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30001:30001"
    command: ["mongod", "--config", "/etc/mongod.conf", "--port","30001"]
    container_name: db01

  db02:
    image: mongo:3.0
    volumes:
    - datadb02:/data/db
    - ./etc/mongod.conf:/etc/mongod.conf
    ports:
    - "30002:30002"
    command: ["mongod", "--config", "/etc/mongod.conf", "--port","30002"]
    container_name: db02

volumes:
  datadb01:
  datadb02:

The command is extended to start mongod internally on port 30001 (30002 for db02) while still exposing it on the same port.

Then the hostnames db01/db02 are added to the application server’s /etc/hosts so there is no problem resolving the name

192.168.10.20    db01
192.168.10.20    db02

After another docker-compose up -d, the changed configurations is applied; however, this breaks our cluster!! The primary and secondary have changed their internal ports, so the cluster connection is lost.

To tell the replSet about this, we need to reconfigure the cluster with the changes:

rs.reconfig({
  "_id": "rs0",
  "version": 1,
  "members" : [
   {"_id": 1, "host": "db01:30001"},
   {"_id": 2, "host": "db02:30002"}
  ]},
  {"force": true }
)

After that, my application is able to connect to the new cluster and everything seems to be fine.

A note on the php config, though:

The MongoClient configuration needs some help to know that there is a cluster and where to perform read/write operations, so the following additional information is necessary:

server-config:
  'mongodb://db01:30001,db02:30002'

client-options:
 readPreference: primary
 replicaSet: rs0

Where and how to put this depends in the individual application, in my case with doctrine_mongodb it looks something like this in the symfony’s config.yml:

doctrine_mongodb:
    connections:
        default:
            server: "mongodb://db01:30001,db02:30002"
            options:
                db: "%mongo_database%"
                readPreference: primary
                replicaSet: rs0

As this article got a little bit longer than expected, the rest will follow in another one