Skip to content

mhubig/atlassian

Repository files navigation

Atlassian services

Version: 1.1.0

This repository holds a dockerized orchestration of the Atlassian web apps Jira, Stash and Confluence. To simplify the usermangement Crowd is also included. For more information on the apps please refere to the offical Atlassian websites:

Prerequisites

In order to run this apps you need to make sure you're running at least docker 1.6.0 and docker-compose 1.2.0. For detailed installation instructions please refere to the origin websites:

Start the images

You can start all images as a orchestration from the root folder. To only use a particularly image change into a subfolder. You better use the docker-compose-dev.yml file if you're not in production. Here are some examples:

# build all the images
$ docker-compose build

# build only the stash image
$ cd atlassian-stash && docker-compose build

# start all the docker images (in development mode)
$ docker-compose -f docker-compose-dev.yml up

# start the stash image (in production mode)
$ cd atlassian-stash && docker-compose -f docker-compose-dev.yml up

# inspect the logs
$ docker-compose logs

If you deploy the apps for the first time you may need to restore the databases from a backup and adapt the database connection settings!

Develop Mode / Debug an image

# use the development compose file
$ docker-compose -f docker-compose-dev.yml up

# execute a bash shell inside a running container
$ docker exec -it atlassian_stash_1 bash

# add the following entrys to your `/etc/hosts`
$ boot2docker ip -> 192.168.59.103
$ cat /etc/hosts
192.168.59.103  boot2docker.local boot2docker
192.168.59.103  stash.boot2docker.local stash
192.168.59.103  jira.boot2docker.local jira
192.168.59.103  confluence.boot2docker.local confluence
192.168.59.103  crowd.boot2docker.local crowd
192.168.59.103  bamboo.boot2docker.local bamboo

First run

If you start this orchestration for the first time, a handy feature is to import your old data. If you're e.g. moving everything to another server you can put your database backups into the tmp folder and the db initscript will pick them up automagically on the first run.

# move your jira db backup file to tmp (filename is important).
$ mv jira_backup.sql tmp/jira.dump

# unpack your jira-home backup archive
$ tar xzf jira-home.tar.gz --strip=1 -C atlassian-jira/home

Backup the home folders

$ mkdir -p backup/$(date +%F)
$ for i in crowd confluence stash jira bamboo; do \
  tar czf backup/$(date +%F)/$i-home.tgz atlassian-$i/home; done

Backup the PostgreSQL data

# backup the confluence database
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_dump -U confluence -h "$DB_PORT_5432_TCP_ADDR" \
    -w confluence > /tmp/confluence.dump'

# backup the stash database
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_dump -U stash -h "$DB_PORT_5432_TCP_ADDR" \
    -w stash > /tmp/stash.dump'

# backup the jira database
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_dump -U jira -h "$DB_PORT_5432_TCP_ADDR" \
    -w jira > /tmp/jira.dump'

# backup the crowd database
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_dump -U crowd -h "$DB_PORT_5432_TCP_ADDR" \
    -w crowd > /tmp/crowd.dump'

# backup the bamboo database
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_dump -U bamboo -h "$DB_PORT_5432_TCP_ADDR" \
    -w bamboo > /tmp/bamboo.dump'

Restore the PostgreSQL data

# restore the confluence database backup
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_restore -U confluence -h "$DB_PORT_5432_TCP_ADDR" \
    -n public -w -d confluence /tmp/confluence.dump'

# restore the stash database backup
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_restore -U stash -h "$DB_PORT_5432_TCP_ADDR" \
    -n public -w -d stash /tmp/stash.dump'

# restore the jira database backup
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_restore -U jira -h "$DB_PORT_5432_TCP_ADDR" \
    -n public -w -d jira /tmp/jira.dump'

# restore the crowd database backup
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_restore -U crowd -h "$DB_PORT_5432_TCP_ADDR" \
    -n public -w -d crowd /tmp/crowd.dump'

# restore the bamboo database backup
$ docker run -it --rm --link atlassian_database_1:db -v $(pwd)/tmp:/tmp \
    postgres sh -c 'pg_restore -U bamboo -h "$DB_PORT_5432_TCP_ADDR" \
    -n public -w -d bamboo /tmp/bamboo.dump'