## What changes were proposed in this pull request?
Hadoop 3 introduces HDFS federation. This means that multiple namespaces are allowed on the same HDFS cluster. In Spark, we need to ask the delegation token for all the namenodes (for each namespace), otherwise accessing any other namespace different from the default one (for which we already fetch the delegation token) fails.
The PR adds the automatic discovery of all the namenodes related to all the namespaces available according to the configs in hdfs-site.xml.
## How was this patch tested?
manual tests in dockerized env
Author: Marco Gaido <marcogaido91@gmail.com>
Closes#21216 from mgaido91/SPARK-24149.
Instead of always throwing a generic exception when the AM fails,
print a generic error and throw the exception with the YARN
diagnostics containing the reason for the failure.
There was an issue with YARN sometimes providing a generic diagnostic
message, even though the AM provides a failure reason when
unregistering. That was happening because the AM was registering
too late, and if errors happened before the registration, YARN would
just create a generic "ExitCodeException" which wasn't very helpful.
Since most errors in this path are a result of not being able to
connect to the driver, this change modifies the AM registration
a bit so that the AM is registered before the connection to the
driver is established. That way, errors are properly propagated
through YARN back to the driver.
As part of that, I also removed the code that retried connections
to the driver from the client AM. At that point, the driver should
already be up and waiting for connections, so it's unlikely that
retrying would help - and in case it does, that means a flaky
network, which would mean problems would probably show up again.
The effect of that is that connection-related errors are reported
back to the driver much faster now (through the YARN report).
One thing to note is that there seems to be a race on the YARN
side that causes a report to be sent to the client without the
corresponding diagnostics string from the AM; the diagnostics are
available later from the RM web page. For that reason, the generic
error messages are kept in the Spark scheduler code, to help
guide users to a way of debugging their failure.
Also of note is that if YARN's max attempts configuration is lower
than Spark's, Spark will not unregister the AM with a proper
diagnostics message. Unfortunately there seems to be no way to
unregister the AM and still allow further re-attempts to happen.
Testing:
- existing unit tests
- some of our integration tests
- hardcoded an invalid driver address in the code and verified
the error in the shell. e.g.
```
scala> 18/05/04 15:09:34 ERROR cluster.YarnClientSchedulerBackend: YARN application has exited unexpectedly with state FAILED! Check the YARN application logs for more details.
18/05/04 15:09:34 ERROR cluster.YarnClientSchedulerBackend: Diagnostics message: Uncaught exception: org.apache.spark.SparkException: Exception thrown in awaitResult:
<AM stack trace>
Caused by: java.io.IOException: Failed to connect to localhost/127.0.0.1:1234
<More stack trace>
```
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#21243 from vanzin/SPARK-24182.
## What changes were proposed in this pull request?
Drastically improves performance and won't cause Spark applications to fail because they write too much data to the Docker image's specific file system. The file system's directories that back emptydir volumes are generally larger and more performant.
## How was this patch tested?
Has been in use via the prototype version of Kubernetes support, but lost in the transition to here.
Author: mcheah <mcheah@palantir.com>
Closes#21238 from mccheah/mount-local-dirs.
## What changes were proposed in this pull request?
Shell escaped the name passed to spark-submit and change how conf attributes are shell escaped.
## How was this patch tested?
This test has been tested manually with Hive-on-spark with mesos or with the use case described in the issue with the sparkPi application with a custom name which contains illegal shell characters.
With this PR, hive-on-spark on mesos works like a charm with hive 3.0.0-SNAPSHOT.
I state that this contribution is my original work and that I license the work to the project under the project’s open source license
Author: Bounkong Khamphousone <bounkong.khamphousone@ebiznext.com>
Closes#21014 from tiboun/fix/SPARK-23941.
## What changes were proposed in this pull request?
Added support to specify the 'spark.executor.extraJavaOptions' value in terms of the `{{APP_ID}}` and/or `{{EXECUTOR_ID}}`, `{{APP_ID}}` will be replaced by Application Id and `{{EXECUTOR_ID}}` will be replaced by Executor Id while starting the executor.
## How was this patch tested?
I have verified this by checking the executor process command and gc logs. I verified the same in different deployment modes(Standalone, YARN, Mesos) client and cluster modes.
Author: Devaraj K <devaraj@apache.org>
Closes#21088 from devaraj-kavali/SPARK-24003.
## What changes were proposed in this pull request?
When a user specifies the wrong class -- or, in fact, a class instead of an object -- Spark throws an NPE which is not useful for debugging. This was reported in [SPARK-23830](https://issues.apache.org/jira/browse/SPARK-23830). This PR adds a check to ensure the main method was found and logs a useful error in the event that it's null.
## How was this patch tested?
* Unit tests + Manual testing
* The scope of the changes is very limited
Author: eric-maynard <emaynard@cloudera.com>
Author: Eric Maynard <emaynard@cloudera.com>
Closes#21168 from eric-maynard/feature/SPARK-23830.
## What changes were proposed in this pull request?
Log stacktrace for uncaught exception
## How was this patch tested?
UT and manually test
Author: zhoukang <zhoukang199191@gmail.com>
Closes#21151 from caneGuy/zhoukang/log-stacktrace.
## What changes were proposed in this pull request?
We propose not to hard-code the RPC port in the AM registration.
## How was this patch tested?
Tested application reports from a pseudo-distributed cluster
```
18/04/10 14:56:21 INFO Client:
client token: N/A
diagnostics: N/A
ApplicationMaster host: localhost
ApplicationMaster RPC port: 58338
queue: default
start time: 1523397373659
final status: UNDEFINED
tracking URL: http://localhost:8088/proxy/application_1523370127531_0016/
```
Author: Gera Shegalov <gera@apache.org>
Closes#21047 from gerashegalov/gera/am-to-rm-nmhost.
## What changes were proposed in this pull request?
Breaks down the construction of driver pods and executor pods in a way that uses a common abstraction for both spark-submit creating the driver and KubernetesClusterSchedulerBackend creating the executor. Encourages more code reuse and is more legible than the older approach.
The high-level design is discussed in more detail on the JIRA ticket. This pull request is the implementation of that design with some minor changes in the implementation details.
No user-facing behavior should break as a result of this change.
## How was this patch tested?
Migrated all unit tests from the old submission steps architecture to the new architecture. Integration tests should not have to change and pass given that this shouldn't change any outward behavior.
Author: mcheah <mcheah@palantir.com>
Closes#20910 from mccheah/spark-22839-incremental.
The current in-process launcher implementation just calls the SparkSubmit
object, which, in case of errors, will more often than not exit the JVM.
This is not desirable since this launcher is meant to be used inside other
applications, and that would kill the application.
The change turns SparkSubmit into a class, and abstracts aways some of
the functionality used to print error messages and abort the submission
process. The default implementation uses the logging system for messages,
and throws exceptions for errors. As part of that I also moved some code
that doesn't really belong in SparkSubmit to a better location.
The command line invocation of spark-submit now uses a special implementation
of the SparkSubmit class that overrides those behaviors to do what is expected
from the command line version (print to the terminal, exit the JVM, etc).
A lot of the changes are to replace calls to methods such as "printErrorAndExit"
with the new API.
As part of adding tests for this, I had to fix some small things in the
launcher option parser so that things like "--version" can work when
used in the launcher library.
There is still code that prints directly to the terminal, like all the
Ivy-related code in SparkSubmitUtils, and other areas where some re-factoring
would help, like the CommandLineUtils class, but I chose to leave those
alone to keep this change more focused.
Aside from existing and added unit tests, I ran command line tools with
a bunch of different arguments to make sure messages and errors behave
like before.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#20925 from vanzin/SPARK-22941.
## What changes were proposed in this pull request?
`YarnAllocator` uses `numExecutorsRunning` to track the number of running executor. `numExecutorsRunning` is used to check if there're executors missing and need to allocate more.
In current code, `numExecutorsRunning` can be negative when driver asks to kill a same idle executor multiple times.
## How was this patch tested?
UT added
Author: jinxing <jinxing6042@126.com>
Closes#20781 from jinxing64/SPARK-23637.
## What changes were proposed in this pull request?
Pass through the `imagePullSecrets` option to the k8s pod in order to allow user to access private image registries.
See https://kubernetes.io/docs/tasks/configure-pod-container/pull-image-private-registry/
## How was this patch tested?
Unit tests + manual testing.
Manual testing procedure:
1. Have private image registry.
2. Spark-submit application with no `spark.kubernetes.imagePullSecret` set. Do `kubectl describe pod ...`. See the error message:
```
Error syncing pod, skipping: failed to "StartContainer" for "spark-kubernetes-driver" with ErrImagePull: "rpc error: code = 2 desc = Error: Status 400 trying to pull repository ...: \"{\\n \\\"errors\\\" : [ {\\n \\\"status\\\" : 400,\\n \\\"message\\\" : \\\"Unsupported docker v1 repository request for '...'\\\"\\n } ]\\n}\""
```
3. Create secret `kubectl create secret docker-registry ...`
4. Spark-submit with `spark.kubernetes.imagePullSecret` set to the new secret. See that deployment was successful.
Author: Andrew Korzhuev <andrew.korzhuev@klarna.com>
Author: Andrew Korzhuev <korzhuev@andrusha.me>
Closes#20811 from andrusha/spark-23668-image-pull-secrets.
## What changes were proposed in this pull request?
As mentioned in SPARK-23285, this PR introduces a new configuration property `spark.kubernetes.executor.cores` for specifying the physical CPU cores requested for each executor pod. This is to avoid changing the semantics of `spark.executor.cores` and `spark.task.cpus` and their role in task scheduling, task parallelism, dynamic resource allocation, etc. The new configuration property only determines the physical CPU cores available to an executor. An executor can still run multiple tasks simultaneously by using appropriate values for `spark.executor.cores` and `spark.task.cpus`.
## How was this patch tested?
Unit tests.
felixcheung srowen jiangxb1987 jerryshao mccheah foxish
Author: Yinan Li <ynli@google.com>
Author: Yinan Li <liyinan926@gmail.com>
Closes#20553 from liyinan926/master.
## What changes were proposed in this pull request?
Kubernetes driver and executor pods should request `memory + memoryOverhead` as their resources instead of just `memory`, see https://issues.apache.org/jira/browse/SPARK-23825
## How was this patch tested?
Existing unit tests were adapted.
Author: David Vogelbacher <dvogelbacher@palantir.com>
Closes#20943 from dvogelbacher/spark-23825.
Currently, the Spark AM relies on the initial set of tokens created by
the submission client to be able to talk to HDFS and other services that
require delegation tokens. This means that after those tokens expire, a
new AM will fail to start (e.g. when there is an application failure and
re-attempts are enabled).
This PR makes it so that the first thing the AM does when the user provides
a principal and keytab is to create new delegation tokens for use. This
makes sure that the AM can be started irrespective of how old the original
token set is. It also allows all of the token management to be done by the
AM - there is no need for the submission client to set configuration values
to tell the AM when to renew tokens.
Note that even though in this case the AM will not be using the delegation
tokens created by the submission client, those tokens still need to be provided
to YARN, since they are used to do log aggregation.
To be able to re-use the code in the AMCredentialRenewal for the above
purposes, I refactored that class a bit so that it can fetch tokens into
a pre-defined UGI, insted of always logging in.
Another issue with re-attempts is that, after the fix that allows the AM
to restart correctly, new executors would get confused about when to
update credentials, because the credential updater used the update time
initially set up by the submission code. This could make the executor
fail to update credentials in time, since that value would be very out
of date in the situation described in the bug.
To fix that, I changed the YARN code to use the new RPC-based mechanism
for distributing tokens to executors. This allowed the old credential
updater code to be removed, and a lot of code in the renewer to be
simplified.
I also made two currently hardcoded values (the renewal time ratio, and
the retry wait) configurable; while this probably never needs to be set
by anyone in a production environment, it helps with testing; that's also
why they're not documented.
Tested on real cluster with a specially crafted application to test this
functionality: checked proper access to HDFS, Hive and HBase in cluster
mode with token renewal on and AM restarts. Tested things still work in
client mode too.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#20657 from vanzin/SPARK-23361.
## What changes were proposed in this pull request?
Yarn throws the following exception in cluster mode when the application is really small:
```
18/03/07 23:34:22 WARN netty.NettyRpcEnv: Ignored failure: java.util.concurrent.RejectedExecutionException: Task java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask7c974942 rejected from java.util.concurrent.ScheduledThreadPoolExecutor1eea9d2d[Terminated, pool size = 0, active threads = 0, queued tasks = 0, completed tasks = 0]
18/03/07 23:34:22 ERROR yarn.ApplicationMaster: Uncaught exception:
org.apache.spark.SparkException: Exception thrown in awaitResult:
at org.apache.spark.util.ThreadUtils$.awaitResult(ThreadUtils.scala:205)
at org.apache.spark.rpc.RpcTimeout.awaitResult(RpcTimeout.scala:75)
at org.apache.spark.rpc.RpcEndpointRef.askSync(RpcEndpointRef.scala:92)
at org.apache.spark.rpc.RpcEndpointRef.askSync(RpcEndpointRef.scala:76)
at org.apache.spark.deploy.yarn.YarnAllocator.<init>(YarnAllocator.scala:102)
at org.apache.spark.deploy.yarn.YarnRMClient.register(YarnRMClient.scala:77)
at org.apache.spark.deploy.yarn.ApplicationMaster.registerAM(ApplicationMaster.scala:450)
at org.apache.spark.deploy.yarn.ApplicationMaster.runDriver(ApplicationMaster.scala:493)
at org.apache.spark.deploy.yarn.ApplicationMaster.org$apache$spark$deploy$yarn$ApplicationMaster$$runImpl(ApplicationMaster.scala:345)
at org.apache.spark.deploy.yarn.ApplicationMaster$$anonfun$run$2.apply$mcV$sp(ApplicationMaster.scala:260)
at org.apache.spark.deploy.yarn.ApplicationMaster$$anonfun$run$2.apply(ApplicationMaster.scala:260)
at org.apache.spark.deploy.yarn.ApplicationMaster$$anonfun$run$2.apply(ApplicationMaster.scala:260)
at org.apache.spark.deploy.yarn.ApplicationMaster$$anon$5.run(ApplicationMaster.scala:810)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:422)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1920)
at org.apache.spark.deploy.yarn.ApplicationMaster.doAsUser(ApplicationMaster.scala:809)
at org.apache.spark.deploy.yarn.ApplicationMaster.run(ApplicationMaster.scala:259)
at org.apache.spark.deploy.yarn.ApplicationMaster$.main(ApplicationMaster.scala:834)
at org.apache.spark.deploy.yarn.ApplicationMaster.main(ApplicationMaster.scala)
Caused by: org.apache.spark.rpc.RpcEnvStoppedException: RpcEnv already stopped.
at org.apache.spark.rpc.netty.Dispatcher.postMessage(Dispatcher.scala:158)
at org.apache.spark.rpc.netty.Dispatcher.postLocalMessage(Dispatcher.scala:135)
at org.apache.spark.rpc.netty.NettyRpcEnv.ask(NettyRpcEnv.scala:229)
at org.apache.spark.rpc.netty.NettyRpcEndpointRef.ask(NettyRpcEnv.scala:523)
at org.apache.spark.rpc.RpcEndpointRef.askSync(RpcEndpointRef.scala:91)
... 17 more
18/03/07 23:34:22 INFO yarn.ApplicationMaster: Final app status: FAILED, exitCode: 13, (reason: Uncaught exception: org.apache.spark.SparkException: Exception thrown in awaitResult: )
```
Example application:
```
object ExampleApp {
def main(args: Array[String]): Unit = {
val conf = new SparkConf().setAppName("ExampleApp")
val sc = new SparkContext(conf)
try {
// Do nothing
} finally {
sc.stop()
}
}
```
This PR pauses user class thread after `SparkContext` created and keeps it so until application master initialises properly.
## How was this patch tested?
Automated: Existing unit tests
Manual: Application submitted into small cluster
Author: Gabor Somogyi <gabor.g.somogyi@gmail.com>
Closes#20807 from gaborgsomogyi/SPARK-23660.
## What changes were proposed in this pull request?
Removal of the init-container for downloading remote dependencies. Built off of the work done by vanzin in an attempt to refactor driver/executor configuration elaborated in [this](https://issues.apache.org/jira/browse/SPARK-22839) ticket.
## How was this patch tested?
This patch was tested with unit and integration tests.
Author: Ilan Filonenko <if56@cornell.edu>
Closes#20669 from ifilonenko/remove-init-container.
## What changes were proposed in this pull request?
As described in SPARK-23680, entrypoint.sh returns an error code because of a command pipeline execution where it is expected in case of Openshift environments, where arbitrary UIDs are used to run containers
## How was this patch tested?
This patch was manually tested by using docker-image-toll.sh script to generate a Spark driver image and running an example against an OpenShift cluster.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Ricardo Martinelli de Oliveira <rmartine@rmartine.gru.redhat.com>
Closes#20822 from rimolive/rmartine-spark-23680.
## What changes were proposed in this pull request?
In the current Spark on YARN code, AM always will copy and overwrite its env variables to executors, so we cannot set different values for executors.
To reproduce issue, user could start spark-shell like:
```
./bin/spark-shell --master yarn-client --conf spark.executorEnv.SPARK_ABC=executor_val --conf spark.yarn.appMasterEnv.SPARK_ABC=am_val
```
Then check executor env variables by
```
sc.parallelize(1 to 1).flatMap \{ i => sys.env.toSeq }.collect.foreach(println)
```
We will always get `am_val` instead of `executor_val`. So we should not let AM to overwrite specifically set executor env variables.
## How was this patch tested?
Added UT and tested in local cluster.
Author: jerryshao <sshao@hortonworks.com>
Closes#20799 from jerryshao/SPARK-23635.
This change restores functionality that was inadvertently removed as part
of the fix for SPARK-22372.
Also modified an existing unit test to make sure the feature works as intended.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#20776 from vanzin/SPARK-23630.
For some JVM options, like `-XX:+UnlockExperimentalVMOptions` ordering is necessary.
## What changes were proposed in this pull request?
Keep original `extraJavaOptions` ordering, when passing them through environment variables inside the Docker container.
## How was this patch tested?
Ran base branch a couple of times and checked startup command in logs. Ordering differed every time. Added sorting, ordering was consistent to what user had in `extraJavaOptions`.
Author: Andrew Korzhuev <korzhuev@andrusha.me>
Closes#20628 from andrusha/patch-2.
This commit modifies the Mesos submission client to allow the principal
and secret to be provided indirectly via files. The path to these files
can be specified either via Spark configuration or via environment
variable.
Assuming these files are appropriately protected by FS/OS permissions
this means we don't ever leak the actual values in process info like ps
Environment variable specification is useful because it allows you to
interpolate the location of this file when using per-user Mesos
credentials.
For some background as to why we have taken this approach I will briefly describe our set up. On our systems we provide each authorised user account with their own Mesos credentials to provide certain security and audit guarantees to our customers. These credentials are managed by a central Secret management service. In our `spark-env.sh` we determine the appropriate secret and principal files to use depending on the user who is invoking Spark hence the need to inject these via environment variables as well as by configuration properties. So we set these environment variables appropriately and our Spark read in the contents of those files to authenticate itself with Mesos.
This is functionality we have been using it in production across multiple customer sites for some time. This has been in the field for around 18 months with no reported issues. These changes have been sufficient to meet our customer security and audit requirements.
We have been building and deploying custom builds of Apache Spark with various minor tweaks like this which we are now looking to contribute back into the community in order that we can rely upon stock Apache Spark builds and stop maintaining our own internal fork.
Author: Rob Vesse <rvesse@dotnetrdf.org>
Closes#20167 from rvesse/SPARK-16501.
## What changes were proposed in this pull request?
Include stacktrace in the diagnostics message upon abnormal unregister from RM
## How was this patch tested?
Tested with a failing job, and confirmed a stacktrace in the client output and YARN webUI.
Author: Gera Shegalov <gera@apache.org>
Closes#20470 from gerashegalov/gera/stacktrace-diagnostics.
Third time is the charm?
There was still a race that was left in previous attempts. If the handle
closes the connection, the close() implementation would clean up state
that would prevent the thread from waiting on the connection thread to
finish. That could cause the race causing the test flakiness reported
in the bug.
The fix is to move the "wait for connection thread" code to a separate
close method that is used by the handle; that also simplifies the code
a bit and makes it also easier to follow.
I included an unrelated, but correct, change to a YARN test so that
it triggers when the PR is built.
Tested by inserting a sleep in the connection thread to mimic the race;
test failed reliably with the sleep, passes now. (Sleep not included in
the patch.) Also ran YARN tests to make sure.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#20388 from vanzin/SPARK-23020.
The race in the code is because the handle might update
its state to the wrong state if the connection handling
thread is still processing incoming data; so the handle
needs to wait for the connection to finish up before
checking the final state.
The race in the test is because when waiting for a handle
to reach a final state, the waitFor() method needs to wait
until all handle state is updated (which also includes
waiting for the connection thread above to finish).
Otherwise, waitFor() may return too early, which would cause
a bunch of different races (like the listener not being yet
notified of the state change, or being in the middle of
being notified, or the handle not being properly disposed
and causing postChecks() to assert).
On top of that I found, by code inspection, a couple of
potential races that could make a handle end up in the
wrong state when being killed.
The original version of this fix introduced the flipped
version of the first race described above; the connection
closing might override the handle state before the
handle might have a chance to do cleanup. The fix there
is to only dispose of the handle from the connection
when there is an error, and let the handle dispose
itself in the normal case.
The fix also caused a bug in YarnClusterSuite to be surfaced;
the code was checking for a file in the classpath that was
not expected to be there in client mode. Because of the above
issues, the error was not propagating correctly and the (buggy)
test was incorrectly passing.
Tested by running the existing unit tests a lot (and not
seeing the errors I was seeing before).
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#20297 from vanzin/SPARK-23020.
## What changes were proposed in this pull request?
In the Kubernetes mode, fails fast in the submission process if any submission client local dependencies are used as the use case is not supported yet.
## How was this patch tested?
Unit tests, integration tests, and manual tests.
vanzin foxish
Author: Yinan Li <liyinan926@gmail.com>
Closes#20320 from liyinan926/master.
Pass through spark java options to the executor in context of docker image.
Closes#20296
andrusha: Deployed two version of containers to local k8s, checked that java options were present in the updated image on the running executor.
Manual test
Author: Andrew Korzhuev <korzhuev@andrusha.me>
Closes#20322 from foxish/patch-1.
## What changes were proposed in this pull request?
This patch bumps the master branch version to `2.4.0-SNAPSHOT`.
## How was this patch tested?
N/A
Author: gatorsmile <gatorsmile@gmail.com>
Closes#20222 from gatorsmile/bump24.
This change allows a user to submit a Spark application on kubernetes
having to provide a single image, instead of one image for each type
of container. The image's entry point now takes an extra argument that
identifies the process that is being started.
The configuration still allows the user to provide different images
for each container type if they so desire.
On top of that, the entry point was simplified a bit to share more
code; mainly, the same env variable is used to propagate the user-defined
classpath to the different containers.
Aside from being modified to match the new behavior, the
'build-push-docker-images.sh' script was renamed to 'docker-image-tool.sh'
to more closely match its purpose; the old name was a little awkward
and now also not entirely correct, since there is a single image. It
was also moved to 'bin' since it's not necessarily an admin tool.
Docs have been updated to match the new behavior.
Tested locally with minikube.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#20192 from vanzin/SPARK-22994.
## What changes were proposed in this pull request?
Two filesystems comparing does not consider the authority of URI. This is specific for
WASB file storage system, where userInfo is honored to differentiate filesystems.
For example: wasbs://user1xyz.net, wasbs://user2xyz.net would consider as two filesystem.
Therefore, we have to add the authority to compare two filesystem, and two filesystem with different authority can not be the same FS.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Mingjie Tang <mtang@hortonworks.com>
Closes#19885 from merlintang/EAR-7377.
## What changes were proposed in this pull request?
The environment variable `SPARK_MOUNTED_CLASSPATH` is referenced in the executor's Dockerfile, where its value is added to the classpath of the executor. However, the scheduler backend code missed setting it when creating the executor pods. This PR fixes it.
## How was this patch tested?
Unit tested.
vanzin Can you help take a look? Thanks!
foxish
Author: Yinan Li <liyinan926@gmail.com>
Closes#20193 from liyinan926/master.
## What changes were proposed in this pull request?
Remove the use of FQDN to access the driver because it assumes that it's set up in a DNS zone - `cluster.local` which is common but not ubiquitous
Note that we already access the in-cluster API server through `kubernetes.default.svc`, so, by extension, this should work as well.
The alternative is to introduce DNS zones for both of those addresses.
## How was this patch tested?
Unit tests
cc vanzin liyinan926 mridulm mccheah
Author: foxish <ramanathana@google.com>
Closes#20187 from foxish/cluster.local.
## What changes were proposed in this pull request?
This PR reverts the `ARG base_image` before `FROM` in the images of driver, executor, and init-container, introduced in https://github.com/apache/spark/pull/20154. The reason is Docker versions before 17.06 do not support this use (`ARG` before `FROM`).
## How was this patch tested?
Tested manually.
vanzin foxish kimoonkim
Author: Yinan Li <liyinan926@gmail.com>
Closes#20170 from liyinan926/master.
## What changes were proposed in this pull request?
Register spark.history.ui.port as a known spark conf to be used in substitution expressions even if it's not set explicitly.
## How was this patch tested?
Added unit test to demonstrate the issue
Author: Gera Shegalov <gera@apache.org>
Author: Gera Shegalov <gshegalov@salesforce.com>
Closes#20098 from gerashegalov/gera/register-SHS-port-conf.
- Make it possible to build images from a git clone.
- Make it easy to use minikube to test things.
Also fixed what seemed like a bug: the base image wasn't getting the tag
provided in the command line. Adding the tag allows users to use multiple
Spark builds in the same kubernetes cluster.
Tested by deploying images on minikube and running spark-submit from a dev
environment; also by building the images with different tags and verifying
"docker images" in minikube.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#20154 from vanzin/SPARK-22960.
## What changes were proposed in this pull request?
User-specified secrets are mounted into both the main container and init-container (when it is used) in a Spark driver/executor pod, using the `MountSecretsBootstrap`. Because `MountSecretsBootstrap` always adds new secret volumes for the secrets to the pod, the same secret volumes get added twice, one when mounting the secrets to the main container, and the other when mounting the secrets to the init-container. This PR fixes the issue by separating `MountSecretsBootstrap.mountSecrets` out into two methods: `addSecretVolumes` for adding secret volumes to a pod and `mountSecrets` for mounting secret volumes to a container, respectively. `addSecretVolumes` is only called once for each pod, whereas `mountSecrets` is called individually for the main container and the init-container (if it is used).
Ref: https://github.com/apache-spark-on-k8s/spark/issues/594.
## How was this patch tested?
Unit tested and manually tested.
vanzin This replaces https://github.com/apache/spark/pull/20148.
hex108 foxish kimoonkim
Author: Yinan Li <liyinan926@gmail.com>
Closes#20159 from liyinan926/master.
This change adds a new launcher that allows applications to be run
in a separate thread in the same process as the calling code. To
achieve that, some code from the child process implementation was
moved to abstract classes that implement the common functionality,
and the new launcher inherits from those.
The new launcher was added as a new class, instead of implemented
as a new option to the existing SparkLauncher, to avoid ambigous
APIs. For example, SparkLauncher has ways to set the child app's
environment, modify SPARK_HOME, or control the logging of the
child process, none of which apply to in-process apps.
The in-process launcher has limitations: it needs Spark in the
context class loader of the calling thread, and it's bound by
Spark's current limitation of a single client-mode application
per JVM. It also relies on the recently added SparkApplication
trait to make sure different apps don't mess up each other's
configuration, so config isolation is currently limited to cluster mode.
I also chose to keep the same socket-based communication for in-process
apps, even though it might be possible to avoid it for in-process
mode. That helps both implementations share more code.
Tested with new and existing unit tests, and with a simple app that
uses the launcher; also made sure the app ran fine with older launcher
jar to check binary compatibility.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#19591 from vanzin/SPARK-11035.
## What changes were proposed in this pull request?
This PR expands the Kubernetes mode to be able to use remote dependencies on http/https endpoints, GCS, S3, etc. It adds steps for configuring and appending the Kubernetes init-container into the driver and executor pods for downloading remote dependencies.
[Init-containers](https://kubernetes.io/docs/concepts/workloads/pods/init-containers/), as the name suggests, are containers that are run to completion before the main containers start, and are often used to perform initialization tasks prior to starting the main containers. We use init-containers to localize remote application dependencies before the driver/executors start running. The code that the init-container runs is also included. This PR also adds a step to the driver and executors for mounting user-specified secrets that may store credentials for accessing data storage, e.g., S3 and Google Cloud Storage (GCS), into the driver and executors.
## How was this patch tested?
* The patch contains unit tests which are passing.
* Manual testing: `./build/mvn -Pkubernetes clean package` succeeded.
* Manual testing of the following cases:
* [x] Running SparkPi using container-local spark-example jar.
* [x] Running SparkPi using container-local spark-example jar with user-specific secret mounted.
* [x] Running SparkPi using spark-example jar hosted remotely on an https endpoint.
cc rxin felixcheung mateiz (shepherd)
k8s-big-data SIG members & contributors: mccheah foxish ash211 ssuchter varunkatta kimoonkim erikerlandson tnachen ifilonenko liyinan926
reviewers: vanzin felixcheung jiangxb1987 mridulm
Author: Yinan Li <liyinan926@gmail.com>
Closes#19954 from liyinan926/init-container.
## What changes were proposed in this pull request?
The path was recently changed in https://github.com/apache/spark/pull/19946, but the dockerfile was not updated.
This is a trivial 1 line fix.
## How was this patch tested?
`./sbin/build-push-docker-images.sh -r spark-repo -t latest build`
cc/ vanzin mridulm rxin jiangxb1987 liyinan926
Author: Anirudh Ramanathan <ramanathana@google.com>
Author: foxish <ramanathana@google.com>
Closes#20051 from foxish/patch-1.
## What changes were proposed in this pull request?
Fixing configuration that was taking an int which should take time. Discussion in https://github.com/apache/spark/pull/19946#discussion_r156682354
Made the granularity milliseconds as opposed to seconds since there's a use-case for sub-second reactions to scale-up rapidly especially with dynamic allocation.
## How was this patch tested?
TODO: manual run of integration tests against this PR.
PTAL
cc/ mccheah liyinan926 kimoonkim vanzin mridulm jiangxb1987 ueshin
Author: foxish <ramanathana@google.com>
Closes#20032 from foxish/fix-time-conf.
# What changes were proposed in this pull request?
1. entrypoint.sh for Kubernetes spark-base image is marked as executable (644 -> 755)
2. make-distribution script will now create kubernetes/dockerfiles directory when Kubernetes support is compiled.
## How was this patch tested?
Manual testing
cc/ ueshin jiangxb1987 mridulm vanzin rxin liyinan926
Author: foxish <ramanathana@google.com>
Closes#20007 from foxish/fix-dockerfiles.
## What changes were proposed in this pull request?
Changes discussed in https://github.com/apache/spark/pull/19946#discussion_r157063535
docker -> container, since with CRI, we are not limited to running only docker images.
## How was this patch tested?
Manual testing
Author: foxish <ramanathana@google.com>
Closes#19995 from foxish/make-docker-container.
## What changes were proposed in this pull request?
This PR added the missing service metadata for `KubernetesClusterManager`. Without the metadata, the service loader couldn't load `KubernetesClusterManager`, and caused the driver to fail to create a `ExternalClusterManager`, as being reported in SPARK-22778. The PR also changed the `k8s:` prefix used to `k8s://`, which is what existing Spark on k8s users are familiar and used to.
## How was this patch tested?
Manual testing verified that the fix resolved the issue in SPARK-22778.
/cc vanzin felixcheung jiangxb1987
Author: Yinan Li <liyinan926@gmail.com>
Closes#19972 from liyinan926/fix-22778.
## What changes were proposed in this pull request?
PR closed with all the comments -> https://github.com/apache/spark/pull/19793
It solves the problem when submitting a wrong CreateSubmissionRequest to Spark Dispatcher was causing a bad state of Dispatcher and making it inactive as a Mesos framework.
https://issues.apache.org/jira/browse/SPARK-22574
## How was this patch tested?
All spark test passed successfully.
It was tested sending a wrong request (without appArgs) before and after the change. The point is easy, check if the value is null before being accessed.
This was before the change, leaving the dispatcher inactive:
```
Exception in thread "Thread-22" java.lang.NullPointerException
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.getDriverCommandValue(MesosClusterScheduler.scala:444)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.buildDriverCommand(MesosClusterScheduler.scala:451)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.org$apache$spark$scheduler$cluster$mesos$MesosClusterScheduler$$createTaskInfo(MesosClusterScheduler.scala:538)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler$$anonfun$scheduleTasks$1.apply(MesosClusterScheduler.scala:570)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler$$anonfun$scheduleTasks$1.apply(MesosClusterScheduler.scala:555)
at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.scheduleTasks(MesosClusterScheduler.scala:555)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.resourceOffers(MesosClusterScheduler.scala:621)
```
And after:
```
"message" : "Malformed request: org.apache.spark.deploy.rest.SubmitRestProtocolException: Validation of message CreateSubmissionRequest failed!\n\torg.apache.spark.deploy.rest.SubmitRestProtocolMessage.validate(SubmitRestProtocolMessage.scala:70)\n\torg.apache.spark.deploy.rest.SubmitRequestServlet.doPost(RestSubmissionServer.scala:272)\n\tjavax.servlet.http.HttpServlet.service(HttpServlet.java:707)\n\tjavax.servlet.http.HttpServlet.service(HttpServlet.java:790)\n\torg.spark_project.jetty.servlet.ServletHolder.handle(ServletHolder.java:845)\n\torg.spark_project.jetty.servlet.ServletHandler.doHandle(ServletHandler.java:583)\n\torg.spark_project.jetty.server.handler.ContextHandler.doHandle(ContextHandler.java:1180)\n\torg.spark_project.jetty.servlet.ServletHandler.doScope(ServletHandler.java:511)\n\torg.spark_project.jetty.server.handler.ContextHandler.doScope(ContextHandler.java:1112)\n\torg.spark_project.jetty.server.handler.ScopedHandler.handle(ScopedHandler.java:141)\n\torg.spark_project.jetty.server.handler.HandlerWrapper.handle(HandlerWrapper.java:134)\n\torg.spark_project.jetty.server.Server.handle(Server.java:524)\n\torg.spark_project.jetty.server.HttpChannel.handle(HttpChannel.java:319)\n\torg.spark_project.jetty.server.HttpConnection.onFillable(HttpConnection.java:253)\n\torg.spark_project.jetty.io.AbstractConnection$ReadCallback.succeeded(AbstractConnection.java:273)\n\torg.spark_project.jetty.io.FillInterest.fillable(FillInterest.java:95)\n\torg.spark_project.jetty.io.SelectChannelEndPoint$2.run(SelectChannelEndPoint.java:93)\n\torg.spark_project.jetty.util.thread.strategy.ExecuteProduceConsume.executeProduceConsume(ExecuteProduceConsume.java:303)\n\torg.spark_project.jetty.util.thread.strategy.ExecuteProduceConsume.produceConsume(ExecuteProduceConsume.java:148)\n\torg.spark_project.jetty.util.thread.strategy.ExecuteProduceConsume.run(ExecuteProduceConsume.java:136)\n\torg.spark_project.jetty.util.thread.QueuedThreadPool.runJob(QueuedThreadPool.java:671)\n\torg.spark_project.jetty.util.thread.QueuedThreadPool$2.run(QueuedThreadPool.java:589)\n\tjava.lang.Thread.run(Thread.java:745)"
```
Author: German Schiavon <germanschiavon@gmail.com>
Closes#19966 from Gschiavon/fix-submission-request.
## What changes were proposed in this pull request?
It solves the problem when submitting a wrong CreateSubmissionRequest to Spark Dispatcher was causing a bad state of Dispatcher and making it inactive as a Mesos framework.
https://issues.apache.org/jira/browse/SPARK-22574
## How was this patch tested?
All spark test passed successfully.
It was tested sending a wrong request (without appArgs) before and after the change. The point is easy, check if the value is null before being accessed.
This was before the change, leaving the dispatcher inactive:
```
Exception in thread "Thread-22" java.lang.NullPointerException
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.getDriverCommandValue(MesosClusterScheduler.scala:444)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.buildDriverCommand(MesosClusterScheduler.scala:451)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.org$apache$spark$scheduler$cluster$mesos$MesosClusterScheduler$$createTaskInfo(MesosClusterScheduler.scala:538)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler$$anonfun$scheduleTasks$1.apply(MesosClusterScheduler.scala:570)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler$$anonfun$scheduleTasks$1.apply(MesosClusterScheduler.scala:555)
at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.scheduleTasks(MesosClusterScheduler.scala:555)
at org.apache.spark.scheduler.cluster.mesos.MesosClusterScheduler.resourceOffers(MesosClusterScheduler.scala:621)
```
And after:
```
"message" : "Malformed request: org.apache.spark.deploy.rest.SubmitRestProtocolException: Validation of message CreateSubmissionRequest failed!\n\torg.apache.spark.deploy.rest.SubmitRestProtocolMessage.validate(SubmitRestProtocolMessage.scala:70)\n\torg.apache.spark.deploy.rest.SubmitRequestServlet.doPost(RestSubmissionServer.scala:272)\n\tjavax.servlet.http.HttpServlet.service(HttpServlet.java:707)\n\tjavax.servlet.http.HttpServlet.service(HttpServlet.java:790)\n\torg.spark_project.jetty.servlet.ServletHolder.handle(ServletHolder.java:845)\n\torg.spark_project.jetty.servlet.ServletHandler.doHandle(ServletHandler.java:583)\n\torg.spark_project.jetty.server.handler.ContextHandler.doHandle(ContextHandler.java:1180)\n\torg.spark_project.jetty.servlet.ServletHandler.doScope(ServletHandler.java:511)\n\torg.spark_project.jetty.server.handler.ContextHandler.doScope(ContextHandler.java:1112)\n\torg.spark_project.jetty.server.handler.ScopedHandler.handle(ScopedHandler.java:141)\n\torg.spark_project.jetty.server.handler.HandlerWrapper.handle(HandlerWrapper.java:134)\n\torg.spark_project.jetty.server.Server.handle(Server.java:524)\n\torg.spark_project.jetty.server.HttpChannel.handle(HttpChannel.java:319)\n\torg.spark_project.jetty.server.HttpConnection.onFillable(HttpConnection.java:253)\n\torg.spark_project.jetty.io.AbstractConnection$ReadCallback.succeeded(AbstractConnection.java:273)\n\torg.spark_project.jetty.io.FillInterest.fillable(FillInterest.java:95)\n\torg.spark_project.jetty.io.SelectChannelEndPoint$2.run(SelectChannelEndPoint.java:93)\n\torg.spark_project.jetty.util.thread.strategy.ExecuteProduceConsume.executeProduceConsume(ExecuteProduceConsume.java:303)\n\torg.spark_project.jetty.util.thread.strategy.ExecuteProduceConsume.produceConsume(ExecuteProduceConsume.java:148)\n\torg.spark_project.jetty.util.thread.strategy.ExecuteProduceConsume.run(ExecuteProduceConsume.java:136)\n\torg.spark_project.jetty.util.thread.QueuedThreadPool.runJob(QueuedThreadPool.java:671)\n\torg.spark_project.jetty.util.thread.QueuedThreadPool$2.run(QueuedThreadPool.java:589)\n\tjava.lang.Thread.run(Thread.java:745)"
```
Author: German Schiavon <germanschiavon@gmail.com>
Closes#19793 from Gschiavon/fix-submission-request.
