Add HA configuration #23
Conversation
| add_if_non_empty high-availability.job.delay "${FLINK_HIGH_AVAILABILITY_JOB_DELAY}" | ||
| add_if_non_empty high-availability.zookeeper.path.root "${FLINK_HIGH_AVAILABILITY_ZOOKEEPER_PATH_ROOT}" | ||
| # used for the client at ha services initiated by the mesos app master runner | ||
| add_if_non_empty ghigh-availability.zookeeper.client.session-timeout \ |
| @@ -73,13 +73,37 @@ function add_kerberos_configurations() { | |||
| fi | |||
| } | |||
|
|
|||
| function add_ha_configurations() { | |||
| if [[ "${FLINK_HA_ENABLED}" == true ]]; then | |||
| export FLINK_JAVA_OPTS="$FLINK_JAVA_OPTS -Dhigh-availability=ZOOKEEPER" | |||
There was a problem hiding this comment.
Why did you need to set high-availability via FLINK_JAVA_OPTS as opposed to the config file?
There was a problem hiding this comment.
I follow the pattern there. This pre-existed just used previous work, didnt refactor it.
There was a problem hiding this comment.
Have a look at the add_if_non_empty function, it is basically a wrapper for appending to FLINK_JAVA_OPTS. For the sake of uniformity, please use it to set the high-availability flag.
I have a minor concern about the sheer number of settings being configured as dynamic properties, as opposed to being configured in flink-conf.yaml. Perhaps some of these HA settings can be set statically.
There was a problem hiding this comment.
Yeah sure for the first option I missed that. By the way it wont work for every property, there is one that causes problems:
2017-08-02 00:00:20,277 INFO org.apache.flink.mesos.runtime.clusterframework.MesosApplicationMasterRunner - -Dhigh-availability.job.delay="10
2017-08-02 00:00:20,277 INFO org.apache.flink.mesos.runtime.clusterframework.MesosApplicationMasterRunner - s"
The property Dhigh-availability.job.delay="10 s" is not preserved when passed due to spaces.
In on order to fix this you can either pass each arg explicitly with quotes or if you want to keep a global variable you need to use eval. The first option is not handy as we have many variables and the second one is not a good choice since it opens security problems if we don't sanitize every input. It gets ugly so I will try put HA options in the file.
skonto
force-pushed
the
ha_configuration
branch
from
1dc35aa to
2c0c70f
Compare
|
@EronWright I updated the PR. Will test and report back. |
skonto
force-pushed
the
ha_configuration
branch
from
2c0c70f to
ed7bb4f
Compare
|
@EronWright fixed. Here is the guide how to run an example with the default settings: |
|
@EronWright is this good to go? I would like include it in the Flink 1.3.2 release of the package. |
|
@EronWright @joerg84 how do you want me to proceed? |
|
@EronWright @joerg84 gentle ping... |
|
Are we planning to merge that? this will be useful for our team as well? |
|
@joerg84 gentle ping. |
|
@joerg84 gentle ping... |
|
closing this. |
Fixes #7, relates to d2iq-archive/universe#1163
The mesosphere image needs to be built and pushed. Don't have access.
Assumes hdfs is there for the default case (any proper setup should have HA enabled by default) and the storage dir (hdfs://hdfs/flink/recovery) to have been created.
In order to create the dir you need to download the appropriate hdfs distro within your dcos cluster and
execute:
dcos hdfs endpoints core-site.xmldcos hdfs endpoints hdfs-site.xmlto download the required files, to override the defaults within the distro so that finally you can issue commands like:
hadoop fs -mkdir -p ...In order to test it after you build the image under your own account, launch flink with HA, use the following command within a docker container of the flink docker image:
./bin/flink run -z /default ./examples/batch/WordCount.jar --input /etc/resolv.confAlso you need to add to the flink-conf.yaml under the /flink-1.2.0 folder within the container the following settings:
high-availability: zookeeper
high-availability.zookeeper.quorum: master.mesos:2181
high-availability.zookeeper.storageDir: hdfs://hdfs/flink/recovery