[
https://issues.apache.org/jira/browse/YARN-8569?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16556348#comment-16556348
]
Eric Yang commented on YARN-8569:
---------------------------------
[[email protected]] Your approach works fine for application that carry
hadoop client or zookeeper client. This proposed interface is to lower the bar
of entry to obtain cluster information for non-Hadoop native applications.
This is the main reason to offer a file based interface for nodes.
The high level view of the design looks like this:
# Application Master received YARN service JSON from yarn cli.
# Application Master write the hostname information to YARN service JSON
resides in /user/${USER}/.yarn/services/[service]/[service].json
# The file is added to distributed cache and localized during container launch.
# The file is bind-mount into docker container for consumption at a predefined
location.
# Flex operation will trigger update of [service].json and repopulate
distributed cache when nodes involved in the cluster has changed.
User application can poll file changes from docker container to be notified of
cluster information changes.
> Create an interface to provide cluster information to application
> -----------------------------------------------------------------
>
> Key: YARN-8569
> URL: https://issues.apache.org/jira/browse/YARN-8569
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Eric Yang
> Priority: Major
> Labels: Docker
>
> Some program requires container hostnames to be known for application to run.
> For example, distributed tensorflow requires launch_command that looks like:
> {code}
> # On ps0.example.com:
> $ python trainer.py \
> --ps_hosts=ps0.example.com:2222,ps1.example.com:2222 \
> --worker_hosts=worker0.example.com:2222,worker1.example.com:2222 \
> --job_name=ps --task_index=0
> # On ps1.example.com:
> $ python trainer.py \
> --ps_hosts=ps0.example.com:2222,ps1.example.com:2222 \
> --worker_hosts=worker0.example.com:2222,worker1.example.com:2222 \
> --job_name=ps --task_index=1
> # On worker0.example.com:
> $ python trainer.py \
> --ps_hosts=ps0.example.com:2222,ps1.example.com:2222 \
> --worker_hosts=worker0.example.com:2222,worker1.example.com:2222 \
> --job_name=worker --task_index=0
> # On worker1.example.com:
> $ python trainer.py \
> --ps_hosts=ps0.example.com:2222,ps1.example.com:2222 \
> --worker_hosts=worker0.example.com:2222,worker1.example.com:2222 \
> --job_name=worker --task_index=1
> {code}
> This is a bit cumbersome to orchestrate via Distributed Shell, or YARN
> services launch_command. In addition, the dynamic parameters do not work
> with YARN flex command. This is the classic pain point for application
> developer attempt to automate system environment settings as parameter to end
> user application.
> It would be great if YARN Docker integration can provide a simple option to
> expose hostnames of the yarn service via a mounted file. The file content
> gets updated when flex command is performed. This allows application
> developer to consume system environment settings via a standard interface.
> It is like /proc/devices for Linux, but for Hadoop. This may involve
> updating a file in distributed cache, and allow mounting of the file via
> container-executor.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
