diff options
author | Hari Shreedharan <hshreedharan@apache.org> | 2015-04-30 13:03:23 -0500 |
---|---|---|
committer | Thomas Graves <tgraves@apache.org> | 2015-04-30 13:03:23 -0500 |
commit | 6c65da6bb7d1213e6a4a9f7fd1597d029d87d07c (patch) | |
tree | 662e09f980e9eeaff8bd572537f9c2549c193337 /docs/security.md | |
parent | 7dacc08ab36188991a001df23880167433844767 (diff) | |
download | spark-6c65da6bb7d1213e6a4a9f7fd1597d029d87d07c.tar.gz spark-6c65da6bb7d1213e6a4a9f7fd1597d029d87d07c.tar.bz2 spark-6c65da6bb7d1213e6a4a9f7fd1597d029d87d07c.zip |
[SPARK-5342] [YARN] Allow long running Spark apps to run on secure YARN/HDFS
Current Spark apps running on Secure YARN/HDFS would not be able to write data
to HDFS after 7 days, since delegation tokens cannot be renewed beyond that. This
means Spark Streaming apps will not be able to run on Secure YARN.
This commit adds basic functionality to fix this issue. In this patch:
- new parameters are added - principal and keytab, which can be used to login to a KDC
- the client logs in, and then get tokens to start the AM
- the keytab is copied to the staging directory
- the AM waits for 60% of the time till expiry of the tokens and then logs in using the keytab
- each time after 60% of the time, new tokens are created and sent to the executors
Currently, to avoid complicating the architecture, we set the keytab and principal in the
SparkHadoopUtil singleton, and schedule a login. Once the login is completed, a callback is scheduled.
This is being posted for feedback, so I can gather feedback on the general implementation.
There are currently a bunch of things to do:
- [x] logging
- [x] testing - I plan to manually test this soon. If you have ideas of how to add unit tests, comment.
- [x] add code to ensure that if these params are set in non-YARN cluster mode, we complain
- [x] documentation
- [x] Have the executors request for credentials from the AM, so that retries are possible.
Author: Hari Shreedharan <hshreedharan@apache.org>
Closes #4688 from harishreedharan/kerberos-longrunning and squashes the following commits:
36eb8a9 [Hari Shreedharan] Change the renewal interval config param. Fix a bunch of comments.
611923a [Hari Shreedharan] Make sure the namenodes are listed correctly for creating tokens.
09fe224 [Hari Shreedharan] Use token.renew to get token's renewal interval rather than using hdfs-site.xml
6963bbc [Hari Shreedharan] Schedule renewal in AM before starting user class. Else, a restarted AM cannot access HDFS if the user class tries to.
072659e [Hari Shreedharan] Fix build failure caused by thread factory getting moved to ThreadUtils.
f041dd3 [Hari Shreedharan] Merge branch 'master' into kerberos-longrunning
42eead4 [Hari Shreedharan] Remove RPC part. Refactor and move methods around, use renewal interval rather than max lifetime to create new tokens.
ebb36f5 [Hari Shreedharan] Merge branch 'master' into kerberos-longrunning
bc083e3 [Hari Shreedharan] Overload RegisteredExecutor to send tokens. Minor doc updates.
7b19643 [Hari Shreedharan] Merge branch 'master' into kerberos-longrunning
8a4f268 [Hari Shreedharan] Added docs in the security guide. Changed some code to ensure that the renewer objects are created only if required.
e800c8b [Hari Shreedharan] Restore original RegisteredExecutor message, and send new tokens via NewTokens message.
0e9507e [Hari Shreedharan] Merge branch 'master' into kerberos-longrunning
7f1bc58 [Hari Shreedharan] Minor fixes, cleanup.
bcd11f9 [Hari Shreedharan] Refactor AM and Executor token update code into separate classes, also send tokens via akka on executor startup.
f74303c [Hari Shreedharan] Move the new logic into specialized classes. Add cleanup for old credentials files.
2f9975c [Hari Shreedharan] Ensure new tokens are written out immediately on AM restart. Also, pikc up the latest suffix from HDFS if the AM is restarted.
61b2b27 [Hari Shreedharan] Account for AM restarts by making sure lastSuffix is read from the files on HDFS.
62c45ce [Hari Shreedharan] Relogin from keytab periodically.
fa233bd [Hari Shreedharan] Adding logging, fixing minor formatting and ordering issues.
42813b4 [Hari Shreedharan] Remove utils.sh, which was re-added due to merge with master.
0de27ee [Hari Shreedharan] Merge branch 'master' into kerberos-longrunning
55522e3 [Hari Shreedharan] Fix failure caused by Preconditions ambiguity.
9ef5f1b [Hari Shreedharan] Added explanation of how the credentials refresh works, some other minor fixes.
f4fd711 [Hari Shreedharan] Fix SparkConf usage.
2debcea [Hari Shreedharan] Change the file structure for credentials files. I will push a followup patch which adds a cleanup mechanism for old credentials files. The credentials files are small and few enough for it to cause issues on HDFS.
af6d5f0 [Hari Shreedharan] Cleaning up files where changes weren't required.
f0f54cb [Hari Shreedharan] Be more defensive when updating the credentials file.
f6954da [Hari Shreedharan] Got rid of Akka communication to renew, instead the executors check a known file's modification time to read the credentials.
5c11c3e [Hari Shreedharan] Move tests to YarnSparkHadoopUtil to fix compile issues.
b4cb917 [Hari Shreedharan] Send keytab to AM via DistributedCache rather than directly via HDFS
0985b4e [Hari Shreedharan] Write tokens to HDFS and read them back when required, rather than sending them over the wire.
d79b2b9 [Hari Shreedharan] Make sure correct credentials are passed to FileSystem#addDelegationTokens()
8c6928a [Hari Shreedharan] Fix issue caused by direct creation of Actor object.
fb27f46 [Hari Shreedharan] Make sure principal and keytab are set before CoarseGrainedSchedulerBackend is started. Also schedule re-logins in CoarseGrainedSchedulerBackend#start()
41efde0 [Hari Shreedharan] Merge branch 'master' into kerberos-longrunning
d282d7a [Hari Shreedharan] Fix ClientSuite to set YARN mode, so that the correct class is used in tests.
bcfc374 [Hari Shreedharan] Fix Hadoop-1 build by adding no-op methods in SparkHadoopUtil, with impl in YarnSparkHadoopUtil.
f8fe694 [Hari Shreedharan] Handle None if keytab-login is not scheduled.
2b0d745 [Hari Shreedharan] [SPARK-5342][YARN] Allow long running Spark apps to run on secure YARN/HDFS.
ccba5bc [Hari Shreedharan] WIP: More changes wrt kerberos
77914dd [Hari Shreedharan] WIP: Add kerberos principal and keytab to YARN client.
Diffstat (limited to 'docs/security.md')
-rw-r--r-- | docs/security.md | 2 |
1 files changed, 2 insertions, 0 deletions
diff --git a/docs/security.md b/docs/security.md index c034ba12ff..d4ffa60e59 100644 --- a/docs/security.md +++ b/docs/security.md @@ -32,6 +32,8 @@ SSL must be configured on each node and configured for each component involved i ### YARN mode The key-store can be prepared on the client side and then distributed and used by the executors as the part of the application. It is possible because the user is able to deploy files before the application is started in YARN by using `spark.yarn.dist.files` or `spark.yarn.dist.archives` configuration settings. The responsibility for encryption of transferring these files is on YARN side and has nothing to do with Spark. +For long-running apps like Spark Streaming apps to be able to write to HDFS, it is possible to pass a principal and keytab to `spark-submit` via the `--principal` and `--keytab` parameters respectively. The keytab passed in will be copied over to the machine running the Application Master via the Hadoop Distributed Cache (securely - if YARN is configured with SSL and HDFS encryption is enabled). The Kerberos login will be periodically renewed using this principal and keytab and the delegation tokens required for HDFS will be generated periodically so the application can continue writing to HDFS. + ### Standalone mode The user needs to provide key-stores and configuration options for master and workers. They have to be set by attaching appropriate Java system properties in `SPARK_MASTER_OPTS` and in `SPARK_WORKER_OPTS` environment variables, or just in `SPARK_DAEMON_JAVA_OPTS`. In this mode, the user may allow the executors to use the SSL settings inherited from the worker which spawned that executor. It can be accomplished by setting `spark.ssl.useNodeLocalConf` to `true`. If that parameter is set, the settings provided by user on the client side, are not used by the executors. |