You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@flink.apache.org by zentol <gi...@git.apache.org> on 2018/02/12 14:30:10 UTC

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

GitHub user zentol opened a pull request:

    https://github.com/apache/flink/pull/5463

    [FLINK-8475][config][docs] Integrate YARN options

    ## What is the purpose of the change
    
    This PR integrates the YARN `ConfigOptions` into the configuration docs generator.
    
    ## Brief change log
    
    * Add missing descriptions to config options (derived from existing description/javadocs)
    * integrate YARN  configuration table into `config.md`

You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/zentol/flink 8475_yarn

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/flink/pull/5463.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #5463
    
----
commit 7199209e6f6ed68589ab841dbbc781802e608e55
Author: zentol <ch...@...>
Date:   2018-01-23T13:04:36Z

    [FLINK-8475][config][docs] Integrate YARN options

----


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by aljoscha <gi...@git.apache.org>.

Github user aljoscha commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5463#discussion_r167607077
  
    --- Diff: flink-yarn/src/main/java/org/apache/flink/yarn/configuration/YarnConfigOptions.java ---
    @@ -79,14 +87,18 @@
     	 */
     	public static final ConfigOption<String> APPLICATION_ATTEMPTS =
     		key("yarn.application-attempts")
    -		.noDefaultValue();
    +		.noDefaultValue()
    --- End diff --
    
    We don't really recommend to leave this at `1` anymore, do we?


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by zentol <gi...@git.apache.org>.

Github user zentol commented on a diff in the pull request:

https://github.com/apache/flink/pull/5463#discussion_r167613260

--- Diff: docs/ops/config.md ---
@@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use

### YARN

-- `containerized.heap-cutoff-ratio`: (Default 0.25) Percentage of heap space to remove from containers started by YARN. When a user requests a certain amount of memory for each TaskManager container (for example 4 GB), we can not pass this amount as the maximum heap space for the JVM (`-Xmx` argument) because the JVM is also allocating memory outside the heap. YARN is very strict with killing containers which are using more memory than requested. Therefore, we remove this fraction of the memory from the requested heap as a safety margin and add it to the memory used off-heap.
-
-- `containerized.heap-cutoff-min`: (Default 600 MB) Minimum amount of memory to cut off the requested heap size.
-
-- `yarn.maximum-failed-containers` (Default: number of requested containers). Maximum number of containers the system is going to reallocate in case of a failure.
-
-- `yarn.application-attempts` (Default: 1). Number of ApplicationMaster restarts. Note that that the entire Flink cluster will restart and the YARN Client will loose the connection. Also, the JobManager address will change and you'll need to set the JM host:port manually. It is recommended to leave this option at 1.
-
-- `yarn.heartbeat-delay` (Default: 5 seconds). Time between heartbeats with the ResourceManager.
-
-- `yarn.properties-file.location` (Default: temp directory). When a Flink job is submitted to YARN, the JobManager's host and the number of available processing slots is written into a properties file, so that the Flink client is able to pick those details up. This configuration parameter allows changing the default location of that file (for example for environments sharing a Flink installation between users)
-
-- `yarn.containers.vcores` The number of virtual cores (vcores) per YARN container. By default, the number of `vcores` is set to the number of slots per TaskManager, if set, or to 1, otherwise.
-
-- `containerized.master.env.`*ENV_VAR1=value* Configuration values prefixed with `containerized.master.env.` will be passed as environment variables to the ApplicationMaster/JobManager process. For example for passing `LD_LIBRARY_PATH` as an env variable to the ApplicationMaster, set:
-
- `containerized.master.env.LD_LIBRARY_PATH: "/usr/lib/native"`
-
-- `containerized.taskmanager.env.` Similar to the configuration prefix about, this prefix allows setting custom environment variables for the TaskManager processes.
-
-- `yarn.container-start-command-template`: Flink uses the following template when starting on YARN:
--- End diff --

looks like you still can. `ConfigConstangs#YARN_CONTAINER_START_COMMAND_TEMPLATE` is read by the yarn cluster descriptor and BootStrapTools. Looks like we missed porting that option.

---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by zentol <gi...@git.apache.org>.

Github user zentol commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5463#discussion_r167613681
  
    --- Diff: docs/ops/config.md ---
    @@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use
     
     ### YARN
     
    -- `containerized.heap-cutoff-ratio`: (Default 0.25) Percentage of heap space to remove from containers started by YARN. When a user requests a certain amount of memory for each TaskManager container (for example 4 GB), we can not pass this amount as the maximum heap space for the JVM (`-Xmx` argument) because the JVM is also allocating memory outside the heap. YARN is very strict with killing containers which are using more memory than requested. Therefore, we remove this fraction of the memory from the requested heap as a safety margin and add it to the memory used off-heap.
    -
    -- `containerized.heap-cutoff-min`: (Default 600 MB) Minimum amount of memory to cut off the requested heap size.
    -
    -- `yarn.maximum-failed-containers` (Default: number of requested containers). Maximum number of containers the system is going to reallocate in case of a failure.
    -
    -- `yarn.application-attempts` (Default: 1). Number of ApplicationMaster restarts. Note that that the entire Flink cluster will restart and the YARN Client will loose the connection. Also, the JobManager address will change and you'll need to set the JM host:port manually. It is recommended to leave this option at 1.
    -
    -- `yarn.heartbeat-delay` (Default: 5 seconds). Time between heartbeats with the ResourceManager.
    -
    -- `yarn.properties-file.location` (Default: temp directory). When a Flink job is submitted to YARN, the JobManager's host and the number of available processing slots is written into a properties file, so that the Flink client is able to pick those details up. This configuration parameter allows changing the default location of that file (for example for environments sharing a Flink installation between users)
    -
    -- `yarn.containers.vcores` The number of virtual cores (vcores) per YARN container. By default, the number of `vcores` is set to the number of slots per TaskManager, if set, or to 1, otherwise.
    -
    -- `containerized.master.env.`*ENV_VAR1=value* Configuration values prefixed with `containerized.master.env.` will be passed as environment variables to the ApplicationMaster/JobManager process. For example for passing `LD_LIBRARY_PATH` as an env variable to the ApplicationMaster, set:
    -
    -    `containerized.master.env.LD_LIBRARY_PATH: "/usr/lib/native"`
    -
    -- `containerized.taskmanager.env.` Similar to the configuration prefix about, this prefix allows setting custom environment variables for the TaskManager processes.
    -
    -- `yarn.container-start-command-template`: Flink uses the following template when starting on YARN:
    -`%java% %jvmmem% %jvmopts% %logging% %class% %args% %redirects%`. This configuration parameter allows users
    -to pass custom settings (such as JVM paths, arguments etc.). Note that in most cases, it is sufficient to
    -use the `env.java.opts` setting, which is the `%jvmopts%` variable in the String.
    -
    -- `yarn.application-master.port` (Default: 0, which lets the OS choose an ephemeral port) With this configuration option, users can specify a port, a range of ports or a list of ports for the  Application Master (and JobManager) RPC port. By default we recommend using the default value (0) to let the operating system choose an appropriate port. In particular when multiple AMs are running on the  same physical host, fixed port assignments prevent the AM from starting.
    -
    -  For example when running Flink on YARN on an environment with a restrictive firewall, this option allows specifying a range of allowed ports.
    -
    -- `yarn.tags` A comma-separated list of tags to apply to the Flink YARN application.
    -
    -- `yarn.per-job-cluster.include-user-jar` (Default: ORDER) Control whether and how the user-jar is included in the system class path for per-job clusters. Setting this parameter to `DISABLED` causes the jar to be included in the user class path instead. Setting this parameter to one of `FIRST`, `LAST` or `ORDER` causes the jar to be included in the system class path, with the jar either being placed at the beginning of the class path (`FIRST`), at the end (`LAST`), or based on the lexicographic order (`ORDER`).
    --- End diff --
    
    you can, I'll update the description. Another case where enum ConfigOptions would be useful.


---

[GitHub] flink issue #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by aljoscha <gi...@git.apache.org>.

Github user aljoscha commented on the issue:

    https://github.com/apache/flink/pull/5463
  
    Ok, please go ahead, then. 👍 


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by aljoscha <gi...@git.apache.org>.

Github user aljoscha commented on a diff in the pull request:

https://github.com/apache/flink/pull/5463#discussion_r167607866

--- Diff: docs/ops/config.md ---
@@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use

### YARN

I see this in the doc and it's referenced twice in comments in tests but you can't actually set this anymore, can you?

---

[GitHub] flink issue #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by zentol <gi...@git.apache.org>.

Github user zentol commented on the issue:

    https://github.com/apache/flink/pull/5463
  
    merging.


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by zentol <gi...@git.apache.org>.

Github user zentol commented on a diff in the pull request:

https://github.com/apache/flink/pull/5463#discussion_r167612255

--- Diff: docs/ops/config.md ---
@@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use

### YARN

yes

---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by aljoscha <gi...@git.apache.org>.

Github user aljoscha commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5463#discussion_r167608886
  
    --- Diff: docs/ops/config.md ---
    @@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use
     
     ### YARN
     
    -- `containerized.heap-cutoff-ratio`: (Default 0.25) Percentage of heap space to remove from containers started by YARN. When a user requests a certain amount of memory for each TaskManager container (for example 4 GB), we can not pass this amount as the maximum heap space for the JVM (`-Xmx` argument) because the JVM is also allocating memory outside the heap. YARN is very strict with killing containers which are using more memory than requested. Therefore, we remove this fraction of the memory from the requested heap as a safety margin and add it to the memory used off-heap.
    -
    -- `containerized.heap-cutoff-min`: (Default 600 MB) Minimum amount of memory to cut off the requested heap size.
    -
    -- `yarn.maximum-failed-containers` (Default: number of requested containers). Maximum number of containers the system is going to reallocate in case of a failure.
    -
    -- `yarn.application-attempts` (Default: 1). Number of ApplicationMaster restarts. Note that that the entire Flink cluster will restart and the YARN Client will loose the connection. Also, the JobManager address will change and you'll need to set the JM host:port manually. It is recommended to leave this option at 1.
    -
    -- `yarn.heartbeat-delay` (Default: 5 seconds). Time between heartbeats with the ResourceManager.
    -
    -- `yarn.properties-file.location` (Default: temp directory). When a Flink job is submitted to YARN, the JobManager's host and the number of available processing slots is written into a properties file, so that the Flink client is able to pick those details up. This configuration parameter allows changing the default location of that file (for example for environments sharing a Flink installation between users)
    -
    -- `yarn.containers.vcores` The number of virtual cores (vcores) per YARN container. By default, the number of `vcores` is set to the number of slots per TaskManager, if set, or to 1, otherwise.
    -
    -- `containerized.master.env.`*ENV_VAR1=value* Configuration values prefixed with `containerized.master.env.` will be passed as environment variables to the ApplicationMaster/JobManager process. For example for passing `LD_LIBRARY_PATH` as an env variable to the ApplicationMaster, set:
    -
    -    `containerized.master.env.LD_LIBRARY_PATH: "/usr/lib/native"`
    -
    -- `containerized.taskmanager.env.` Similar to the configuration prefix about, this prefix allows setting custom environment variables for the TaskManager processes.
    -
    -- `yarn.container-start-command-template`: Flink uses the following template when starting on YARN:
    -`%java% %jvmmem% %jvmopts% %logging% %class% %args% %redirects%`. This configuration parameter allows users
    -to pass custom settings (such as JVM paths, arguments etc.). Note that in most cases, it is sufficient to
    -use the `env.java.opts` setting, which is the `%jvmopts%` variable in the String.
    -
    -- `yarn.application-master.port` (Default: 0, which lets the OS choose an ephemeral port) With this configuration option, users can specify a port, a range of ports or a list of ports for the  Application Master (and JobManager) RPC port. By default we recommend using the default value (0) to let the operating system choose an appropriate port. In particular when multiple AMs are running on the  same physical host, fixed port assignments prevent the AM from starting.
    -
    -  For example when running Flink on YARN on an environment with a restrictive firewall, this option allows specifying a range of allowed ports.
    -
    -- `yarn.tags` A comma-separated list of tags to apply to the Flink YARN application.
    -
    -- `yarn.per-job-cluster.include-user-jar` (Default: ORDER) Control whether and how the user-jar is included in the system class path for per-job clusters. Setting this parameter to `DISABLED` causes the jar to be included in the user class path instead. Setting this parameter to one of `FIRST`, `LAST` or `ORDER` causes the jar to be included in the system class path, with the jar either being placed at the beginning of the class path (`FIRST`), at the end (`LAST`), or based on the lexicographic order (`ORDER`).
    --- End diff --
    
    You can't set `DISABLED` anymore?


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by aljoscha <gi...@git.apache.org>.

Github user aljoscha commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5463#discussion_r167606495
  
    --- Diff: docs/ops/config.md ---
    @@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use
     
     ### YARN
     
    -- `containerized.heap-cutoff-ratio`: (Default 0.25) Percentage of heap space to remove from containers started by YARN. When a user requests a certain amount of memory for each TaskManager container (for example 4 GB), we can not pass this amount as the maximum heap space for the JVM (`-Xmx` argument) because the JVM is also allocating memory outside the heap. YARN is very strict with killing containers which are using more memory than requested. Therefore, we remove this fraction of the memory from the requested heap as a safety margin and add it to the memory used off-heap.
    --- End diff --
    
    This will be in the general resource manager options from now on, I'm guessing? (this option and the next)


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by aljoscha <gi...@git.apache.org>.

Github user aljoscha commented on a diff in the pull request:

https://github.com/apache/flink/pull/5463#discussion_r167607428

--- Diff: docs/ops/config.md ---
@@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use

### YARN

These next three will also be in the general RM section?

---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by asfgit <gi...@git.apache.org>.

Github user asfgit closed the pull request at:

    https://github.com/apache/flink/pull/5463


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by zentol <gi...@git.apache.org>.

Github user zentol commented on a diff in the pull request:

https://github.com/apache/flink/pull/5463#discussion_r167615886

--- Diff: docs/ops/config.md ---
@@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use

### YARN

and porting that one is tricky since it is used by flink-runtime but the yarn options are in flink-yarn...

---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by zentol <gi...@git.apache.org>.

Github user zentol commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5463#discussion_r167612224
  
    --- Diff: flink-yarn/src/main/java/org/apache/flink/yarn/configuration/YarnConfigOptions.java ---
    @@ -79,14 +87,18 @@
     	 */
     	public static final ConfigOption<String> APPLICATION_ATTEMPTS =
     		key("yarn.application-attempts")
    -		.noDefaultValue();
    +		.noDefaultValue()
    --- End diff --
    
    I haven't seen any recommendation in the [Yarn docs](https://ci.apache.org/projects/flink/flink-docs-master/ops/deployment/yarn_setup.html). In the [HA docs](https://ci.apache.org/projects/flink/flink-docs-master/ops/jobmanager_high_availability.html#yarn-cluster-high-availability) we set it to 10 but never explicitly call this a recommended value.


---

[GitHub] flink pull request #5463: [FLINK-8475][config][docs] Integrate YARN options

Posted by zentol <gi...@git.apache.org>.

Github user zentol commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5463#discussion_r167611073
  
    --- Diff: docs/ops/config.md ---
    @@ -408,38 +408,7 @@ of the JobManager, because the same ActorSystem is used. Its not possible to use
     
     ### YARN
     
    -- `containerized.heap-cutoff-ratio`: (Default 0.25) Percentage of heap space to remove from containers started by YARN. When a user requests a certain amount of memory for each TaskManager container (for example 4 GB), we can not pass this amount as the maximum heap space for the JVM (`-Xmx` argument) because the JVM is also allocating memory outside the heap. YARN is very strict with killing containers which are using more memory than requested. Therefore, we remove this fraction of the memory from the requested heap as a safety margin and add it to the memory used off-heap.
    --- End diff --
    
    correct.


---