1
0
Fork 0

cleanup & docs

This commit is contained in:
Claudio Atzori 2023-10-12 12:23:20 +02:00
parent 1902728f7e
commit 76447958bb
6 changed files with 127 additions and 190 deletions

128
README.md
View File

@ -1,2 +1,128 @@
# dnet-hadoop
Dnet-hadoop is the project that defined all the OOZIE workflows for the OpenAIRE Graph construction, processing, provisioning.
Dnet-hadoop is the project that defined all the [OOZIE workflows](https://oozie.apache.org/) for the OpenAIRE Graph construction, processing, provisioning.
How to build, package and run oozie workflows
====================
Oozie-installer is a utility allowing building, uploading and running oozie workflows. In practice, it creates a `*.tar.gz`
package that contains resources that define a workflow and some helper scripts.
This module is automatically executed when running:
`mvn package -Poozie-package -Dworkflow.source.dir=classpath/to/parent/directory/of/oozie_app`
on module having set:
```
<parent>
<groupId>eu.dnetlib.dhp</groupId>
<artifactId>dhp-workflows</artifactId>
</parent>
```
in `pom.xml` file. `oozie-package` profile initializes oozie workflow packaging, `workflow.source.dir` property points to
a workflow (notice: this is not a relative path but a classpath to directory usually holding `oozie_app` subdirectory).
The outcome of this packaging is `oozie-package.tar.gz` file containing inside all the resources required to run Oozie workflow:
- jar packages
- workflow definitions
- job properties
- maintenance scripts
Required properties
====================
In order to include proper workflow within package, `workflow.source.dir` property has to be set. It could be provided
by setting `-Dworkflow.source.dir=some/job/dir` maven parameter.
In oder to define full set of cluster environment properties one should create `~/.dhp/application.properties` file with
the following properties:
- `dhp.hadoop.frontend.user.name` - your user name on hadoop cluster and frontend machine
- `dhp.hadoop.frontend.host.name` - frontend host name
- `dhp.hadoop.frontend.temp.dir` - frontend directory for temporary files
- `dhp.hadoop.frontend.port.ssh` - frontend machine ssh port
- `oozieServiceLoc` - oozie service location required by run_workflow.sh script executing oozie job
- `nameNode` - name node address
- `jobTracker` - job tracker address
- `oozie.execution.log.file.location` - location of file that will be created when executing oozie job, it contains output
produced by `run_workflow.sh` script (needed to obtain oozie job id)
- `maven.executable` - mvn command location, requires parameterization due to a different setup of CI cluster
- `sparkDriverMemory` - amount of memory assigned to spark jobs driver
- `sparkExecutorMemory` - amount of memory assigned to spark jobs executors
- `sparkExecutorCores` - number of cores assigned to spark jobs executors
All values will be overriden with the ones from `job.properties` and eventually `job-override.properties` stored in module's
main folder.
When overriding properties from `job.properties`, `job-override.properties` file can be created in main module directory
(the one containing `pom.xml` file) and define all new properties which will override existing properties.
One can provide those properties one by one as command line `-D` arguments.
Properties overriding order is the following:
1. `pom.xml` defined properties (located in the project root dir)
2. `~/.dhp/application.properties` defined properties
3. `${workflow.source.dir}/job.properties`
4. `job-override.properties` (located in the project root dir)
5. `maven -Dparam=value`
where the maven `-Dparam` property is overriding all the other ones.
Workflow definition requirements
====================
`workflow.source.dir` property should point to the following directory structure:
[${workflow.source.dir}]
|
|-job.properties (optional)
|
\-[oozie_app]
|
\-workflow.xml
This property can be set using maven `-D` switch.
`[oozie_app]` is the default directory name however it can be set to any value as soon as `oozieAppDir` property is
provided with directory name as value.
Sub-workflows are supported as well and sub-workflow directories should be nested within `[oozie_app]` directory.
Creating oozie installer step-by-step
=====================================
Automated oozie-installer steps are the following:
1. creating jar packages: `*.jar` and `*tests.jar` along with copying all dependencies in `target/dependencies`
2. reading properties from maven, `~/.dhp/application.properties`, `job.properties`, `job-override.properties`
3. invoking priming mechanism linking resources from import.txt file (currently resolving subworkflow resources)
4. assembling shell scripts for preparing Hadoop filesystem, uploading Oozie application and starting workflow
5. copying whole `${workflow.source.dir}` content to `target/${oozie.package.file.name}`
6. generating updated `job.properties` file in `target/${oozie.package.file.name}` based on maven,
`~/.dhp/application.properties`, `job.properties` and `job-override.properties`
7. creating `lib` directory (or multiple directories for sub-workflows for each nested directory) and copying jar packages
created at step (1) to each one of them
8. bundling whole `${oozie.package.file.name}` directory into single tar.gz package
Uploading oozie package and running workflow on cluster
=======================================================
In order to simplify deployment and execution process two dedicated profiles were introduced:
- `deploy`
- `run`
to be used along with `oozie-package` profile e.g. by providing `-Poozie-package,deploy,run` maven parameters.
The `deploy` profile supplements packaging process with:
1) uploading oozie-package via scp to `/home/${user.name}/oozie-packages` directory on `${dhp.hadoop.frontend.host.name}` machine
2) extracting uploaded package
3) uploading oozie content to hadoop cluster HDFS location defined in `oozie.wf.application.path` property (generated dynamically by maven build process, based on `${dhp.hadoop.frontend.user.name}` and `workflow.source.dir` properties)
The `run` profile introduces:
1) executing oozie application uploaded to HDFS cluster using `deploy` command. Triggers `run_workflow.sh` script providing runtime properties defined in `job.properties` file.
Notice: ssh access to frontend machine has to be configured on system level and it is preferable to set key-based authentication in order to simplify remote operations.

View File

@ -1,13 +0,0 @@
<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
<parent>
<artifactId>dhp-workflows</artifactId>
<groupId>eu.dnetlib.dhp</groupId>
<version>1.2.5-SNAPSHOT</version>
</parent>
<modelVersion>4.0.0</modelVersion>
<artifactId>dhp-distcp</artifactId>
</project>

View File

@ -1,18 +0,0 @@
<configuration>
<property>
<name>jobTracker</name>
<value>yarnRM</value>
</property>
<property>
<name>nameNode</name>
<value>hdfs://nameservice1</value>
</property>
<property>
<name>sourceNN</name>
<value>webhdfs://namenode2.hadoop.dm.openaire.eu:50071</value>
</property>
<property>
<name>oozie.use.system.libpath</name>
<value>true</value>
</property>
</configuration>

View File

@ -1,46 +0,0 @@
<workflow-app name="distcp" xmlns="uri:oozie:workflow:0.5">
<parameters>
<property>
<name>sourceNN</name>
<description>the source name node</description>
</property>
<property>
<name>sourcePath</name>
<description>the source path</description>
</property>
<property>
<name>targetPath</name>
<description>the target path</description>
</property>
<property>
<name>hbase_dump_distcp_memory_mb</name>
<value>6144</value>
<description>memory for distcp action copying InfoSpace dump from remote cluster</description>
</property>
<property>
<name>hbase_dump_distcp_num_maps</name>
<value>1</value>
<description>maximum number of simultaneous copies of InfoSpace dump from remote location</description>
</property>
</parameters>
<start to="distcp"/>
<kill name="Kill">
<message>Action failed, error message[${wf:errorMessage(wf:lastErrorNode())}]</message>
</kill>
<action name="distcp">
<distcp xmlns="uri:oozie:distcp-action:0.2">
<arg>-Dmapreduce.map.memory.mb=${hbase_dump_distcp_memory_mb}</arg>
<arg>-pb</arg>
<arg>-m ${hbase_dump_distcp_num_maps}</arg>
<arg>${sourceNN}/${sourcePath}</arg>
<arg>${nameNode}/${targetPath}</arg>
</distcp>
<ok to="End" />
<error to="Kill" />
</action>
<end name="End"/>
</workflow-app>

View File

@ -1,111 +0,0 @@
General notes
====================
Oozie-installer is a utility allowing building, uploading and running oozie workflows. In practice, it creates a `*.tar.gz` package that contains resouces that define a workflow and some helper scripts.
This module is automatically executed when running:
`mvn package -Poozie-package -Dworkflow.source.dir=classpath/to/parent/directory/of/oozie_app`
on module having set:
<parent>
<groupId>eu.dnetlib.dhp</groupId>
<artifactId>dhp-workflows</artifactId>
</parent>
in `pom.xml` file. `oozie-package` profile initializes oozie workflow packaging, `workflow.source.dir` property points to a workflow (notice: this is not a relative path but a classpath to directory usually holding `oozie_app` subdirectory).
The outcome of this packaging is `oozie-package.tar.gz` file containing inside all the resources required to run Oozie workflow:
- jar packages
- workflow definitions
- job properties
- maintenance scripts
Required properties
====================
In order to include proper workflow within package, `workflow.source.dir` property has to be set. It could be provided by setting `-Dworkflow.source.dir=some/job/dir` maven parameter.
In oder to define full set of cluster environment properties one should create `~/.dhp/application.properties` file with the following properties:
- `dhp.hadoop.frontend.user.name` - your user name on hadoop cluster and frontend machine
- `dhp.hadoop.frontend.host.name` - frontend host name
- `dhp.hadoop.frontend.temp.dir` - frontend directory for temporary files
- `dhp.hadoop.frontend.port.ssh` - frontend machine ssh port
- `oozieServiceLoc` - oozie service location required by run_workflow.sh script executing oozie job
- `nameNode` - name node address
- `jobTracker` - job tracker address
- `oozie.execution.log.file.location` - location of file that will be created when executing oozie job, it contains output produced by `run_workflow.sh` script (needed to obtain oozie job id)
- `maven.executable` - mvn command location, requires parameterization due to a different setup of CI cluster
- `sparkDriverMemory` - amount of memory assigned to spark jobs driver
- `sparkExecutorMemory` - amount of memory assigned to spark jobs executors
- `sparkExecutorCores` - number of cores assigned to spark jobs executors
All values will be overriden with the ones from `job.properties` and eventually `job-override.properties` stored in module's main folder.
When overriding properties from `job.properties`, `job-override.properties` file can be created in main module directory (the one containing `pom.xml` file) and define all new properties which will override existing properties. One can provide those properties one by one as command line -D arguments.
Properties overriding order is the following:
1. `pom.xml` defined properties (located in the project root dir)
2. `~/.dhp/application.properties` defined properties
3. `${workflow.source.dir}/job.properties`
4. `job-override.properties` (located in the project root dir)
5. `maven -Dparam=value`
where the maven `-Dparam` property is overriding all the other ones.
Workflow definition requirements
====================
`workflow.source.dir` property should point to the following directory structure:
[${workflow.source.dir}]
|
|-job.properties (optional)
|
\-[oozie_app]
|
\-workflow.xml
This property can be set using maven `-D` switch.
`[oozie_app]` is the default directory name however it can be set to any value as soon as `oozieAppDir` property is provided with directory name as value.
Subworkflows are supported as well and subworkflow directories should be nested within `[oozie_app]` directory.
Creating oozie installer step-by-step
=====================================
Automated oozie-installer steps are the following:
1. creating jar packages: `*.jar` and `*tests.jar` along with copying all dependancies in `target/dependencies`
2. reading properties from maven, `~/.dhp/application.properties`, `job.properties`, `job-override.properties`
3. invoking priming mechanism linking resources from import.txt file (currently resolving subworkflow resources)
4. assembling shell scripts for preparing Hadoop filesystem, uploading Oozie application and starting workflow
5. copying whole `${workflow.source.dir}` content to `target/${oozie.package.file.name}`
6. generating updated `job.properties` file in `target/${oozie.package.file.name}` based on maven, `~/.dhp/application.properties`, `job.properties` and `job-override.properties`
7. creating `lib` directory (or multiple directories for subworkflows for each nested directory) and copying jar packages created at step (1) to each one of them
8. bundling whole `${oozie.package.file.name}` directory into single tar.gz package
Uploading oozie package and running workflow on cluster
=======================================================
In order to simplify deployment and execution process two dedicated profiles were introduced:
- `deploy`
- `run`
to be used along with `oozie-package` profile e.g. by providing `-Poozie-package,deploy,run` maven parameters.
`deploy` profile supplements packaging process with:
1) uploading oozie-package via scp to `/home/${user.name}/oozie-packages` directory on `${dhp.hadoop.frontend.host.name}` machine
2) extracting uploaded package
3) uploading oozie content to hadoop cluster HDFS location defined in `oozie.wf.application.path` property (generated dynamically by maven build process, based on `${dhp.hadoop.frontend.user.name}` and `workflow.source.dir` properties)
`run` profile introduces:
1) executing oozie application uploaded to HDFS cluster using `deploy` command. Triggers `run_workflow.sh` script providing runtime properties defined in `job.properties` file.
Notice: ssh access to frontend machine has to be configured on system level and it is preferable to set key-based authentication in order to simplify remote operations.

View File

@ -25,7 +25,6 @@
<modules>
<module>dhp-workflow-profiles</module>
<module>dhp-aggregation</module>
<module>dhp-distcp</module>
<module>dhp-actionmanager</module>
<module>dhp-graph-mapper</module>
<module>dhp-dedup-openaire</module>