edrusb/atlas repository overview

DATA_ROOT must be set and point to the path inside the container where will be mounted the persistent storage.
ATLAS_HOSTNAME must also be set to a hostname that resolves to the IP address that external atlas hook client will be able to connect at, in order to publish messages through kafka (port 9027).

Here is an example of call:

docker run -d -p 8983:8983 -p 9838:9838 -p 8838:8838 -p 9026:9026 -p 9027:9027 -p 61500:61500 -p 61510:61510 -p 61520:61520 -p 61530:61530 -p 21000:21000 -e DATA_ROOT=/opt/persist -e ATLAS_HOSTNAME=`hostname -A` -v /mapr/clustername/volume:/opt/persist edrusb/atlas:1.18

remove the corresponding -p directive to make the corresponding service unavailable outside the container

To avoid data corruption when terminating the container while a lot of data is managed by atlas, it may be wise to increase the exit timeout to 60 seconds or more for a graceful shutdown to stay possible:

docker stop -t 60 <container ID>

TCP ports used and the corresponding process listening behind:

• Atlas listens on port 21000
• solr on 9838 and 8838
• kafka.zookeeper.connect on port 9026
• kafka.zookeeper.server on 9027
• hbase master port is 61500
• hbase master info port is 61510
• hbase region-server port is 61520
• hbase region-server info port is 51530
• zookeeper on port 2181

Since container version 1.15 the atlas hook binaries (for Hive, falcon, impala, kafka, sqoop, storm, hbase) are extracted from the docker image to the directory $DATA_ROOT/hook.packages

Version 1.16 brings support for Persistent Storage Claims which are not settable by root user, need to change to non-privileged user to setup the room for atlas, hbase, solr and zookeeper

Version 1.17 fixes a minor bug about modification brought to /etc/hosts in container: adding the $ATLAS_HOSTNAME as alias of the existing IP not as a new line

Version 1.18 exports the atlas-application.properties file beside the hooks to the $DATA_ROOT/hook.packages directory