[
https://issues.apache.org/jira/browse/SPARK-24446?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Marcelo Vanzin updated SPARK-24446:
-----------------------------------
Description:
When YARN runs the application's main command, it does it like this:
{code}
bash -c "<your command goes here>"
{code}
The way Spark injects the library path into that command makes it look like
this:
{code}
bash -c "LD_LIBRARY_PATH="/foo:/bar:/baz:$LD_LIBRARY_PATH" <rest of the
command>"
{code}
So that works kinda out of luck, because the concatenation of the strings
creates a proper final command... except if you have something like a space or
an ampersand in the library path, in which case all containers will fail with a
cryptic message like the following:
{noformat}
WARN cluster.YarnSchedulerBackend$YarnSchedulerEndpoint: Executor for container
container_1475411358336_0010_01_000002 exited because of a YARN event (e.g.,
pre-emption) and not because of an error in the running job.
{noformat}
And no useful log output.
was:
When YARN runs the application's main command, it does it like this:
{code}
bash -c "<your command goes here>"
{code}
The way Spark injects the library path into that command makes it look like
this:
{code}
bash -c "LD_LIBRARY_PATH="/foo:/bar:/baz:$LD_LIBRARY_PATH" <rest of the
command>"
{code}
So that works kinda out of luck, because the concatenation of the strings
creates a proper final command... except if you have something like a space of
an ampersand in the library path, in which case all containers will fail with a
cryptic message like the following:
{noformat}
WARN cluster.YarnSchedulerBackend$YarnSchedulerEndpoint: Executor for container
container_1475411358336_0010_01_000002 exited because of a YARN event (e.g.,
pre-emption) and not because of an error in the running job.
{noformat}
And no useful log output.
> Library path with special characters breaks Spark on YARN
> ---------------------------------------------------------
>
> Key: SPARK-24446
> URL: https://issues.apache.org/jira/browse/SPARK-24446
> Project: Spark
> Issue Type: Bug
> Components: YARN
> Affects Versions: 2.3.0
> Reporter: Marcelo Vanzin
> Priority: Minor
>
> When YARN runs the application's main command, it does it like this:
> {code}
> bash -c "<your command goes here>"
> {code}
> The way Spark injects the library path into that command makes it look like
> this:
> {code}
> bash -c "LD_LIBRARY_PATH="/foo:/bar:/baz:$LD_LIBRARY_PATH" <rest of the
> command>"
> {code}
> So that works kinda out of luck, because the concatenation of the strings
> creates a proper final command... except if you have something like a space
> or an ampersand in the library path, in which case all containers will fail
> with a cryptic message like the following:
> {noformat}
> WARN cluster.YarnSchedulerBackend$YarnSchedulerEndpoint: Executor for
> container container_1475411358336_0010_01_000002 exited because of a YARN
> event (e.g., pre-emption) and not because of an error in the running job.
> {noformat}
> And no useful log output.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
