lanyuanxiaoyao/hudi
1
0
Fork 0
Code Activity

Fixing README for hudi test suite long running job (#2578)

Browse Source
This commit is contained in:
Sivabalan Narayanan
2021-02-25 19:50:18 -05:00
committed by GitHub
parent 8c2197ae5e
commit 9f5e8cc7c3
1 changed files with 59 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

94
hudi-integ-test/README.md
View File

@@ -270,20 +270,31 @@ spark-submit \
--compact-scheduling-minshare 1
```
For long running test suite, validation has to be done differently. Idea is to run same dag in a repeated manner.
Hence "ValidateDatasetNode" is introduced which will read entire input data and compare it with hudi contents both via
spark datasource and hive table via spark sql engine.
## Running long running test suite in Local Docker environment
For long running test suite, validation has to be done differently. Idea is to run same dag in a repeated manner for
N iterations. Hence "ValidateDatasetNode" is introduced which will read entire input data and compare it with hudi
contents both via spark datasource and hive table via spark sql engine. Hive validation is configurable.
If you have "ValidateDatasetNode" in your dag, do not replace hive jars as instructed above. Spark sql engine does not
go well w/ hive2* jars. So, after running docker setup, just copy test.properties and your dag of interest and you are
good to go ahead.
go well w/ hive2* jars. So, after running docker setup, follow the below steps.
```
docker cp packaging/hudi-integ-test-bundle/target/hudi-integ-test-bundle-0.8.0-SNAPSHOT.jar adhoc-2:/opt/
docker cp demo/config/test-suite/test.properties adhoc-2:/opt/
```
Also copy your dag of interest to adhoc-2:/opt/
```
docker cp demo/config/test-suite/complex-dag-cow.yaml adhoc-2:/opt/
```
For repeated runs, two additional configs need to be set. "dag_rounds" and "dag_intermittent_delay_mins".
This means that your dag will be repeated for N times w/ a delay of Y mins between each round.
This means that your dag will be repeated for N times w/ a delay of Y mins between each round. Note: complex-dag-cow.yaml
already has all these configs set. So no changes required just to try it out.
Also, ValidateDatasetNode can be configured in two ways. Either with "delete_input_data" set to true or without
setting the config. When "delete_input_data" is set for ValidateDatasetNode, once validation is complete, entire input
data will be deleted. So, suggestion is to use this ValidateDatasetNode as the last node in the dag with "delete_input_data".
Also, ValidateDatasetNode can be configured in two ways. Either with "delete_input_data: true" set or not set.
When "delete_input_data" is set for ValidateDatasetNode, once validation is complete, entire input data will be deleted.
So, suggestion is to use this ValidateDatasetNode as the last node in the dag with "delete_input_data".
Example dag:
```
Insert
@@ -294,7 +305,7 @@ Example dag:
If above dag is run with "dag_rounds" = 10 and "dag_intermittent_delay_mins" = 10, then this dag will run for 10 times
with 10 mins delay between every run. At the end of every run, records written as part of this round will be validated.
At the end of each validation, all contents of input are deleted.
For eg: incase of above dag,
To illustrate each round
```
Round1:
insert => inputPath/batch1
@@ -322,6 +333,12 @@ may not scale beyond certain point since input data as well as hudi content's ke
every cycle.
Lets see an example where you don't set "delete_input_data" as part of Validation.
```
Insert
Upsert
ValidateDatasetNode
```
Here is the illustration of each round
```
Round1:
insert => inputPath/batch1
@@ -382,27 +399,14 @@ Above dag was just an example for illustration purposes. But you can make it com
Validate w/o deleting
Upsert
Validate w/ deletion
```
With this dag, you can set the two additional configs "dag_rounds" and "dag_intermittent_delay_mins" and have a long
running test suite.
```
dag_rounds: 1
dag_intermittent_delay_mins: 10
dag_content:
Insert
Upsert
Delete
Validate w/o deleting
Insert
Rollback
Validate w/o deleting
Upsert
Validate w/ deletion
```
Sample COW command with repeated runs.
Once you have copied the jar, test.properties and your dag to adhoc-2:/opt/, you can run the following command to execute
the test suite job.
```
docker exec -it adhoc-2 /bin/bash
```
Sample COW command
```
spark-submit \
--packages org.apache.spark:spark-avro_2.11:2.4.0 \
@@ -424,25 +428,45 @@ spark-submit \
--conf spark.driver.extraClassPath=/var/demo/jars/* \
--conf spark.executor.extraClassPath=/var/demo/jars/* \
--class org.apache.hudi.integ.testsuite.HoodieTestSuiteJob \
/opt/hudi-integ-test-bundle-0.6.1-SNAPSHOT.jar \
/opt/hudi-integ-test-bundle-0.8.0-SNAPSHOT.jar \
--source-ordering-field test_suite_source_ordering_field \
--use-deltastreamer \
--target-base-path /user/hive/warehouse/hudi-integ-test-suite/output \
--input-base-path /user/hive/warehouse/hudi-integ-test-suite/input \
--target-table table1 \
--props test.properties \
--schemaprovider-class org.apache.hudi.utilities.schema.FilebasedSchemaProvider \
--schemaprovider-class org.apache.hudi.integ.testsuite.schema.TestSuiteFileBasedSchemaProvider \
--source-class org.apache.hudi.utilities.sources.AvroDFSSource \
--input-file-size 125829120 \
--workload-yaml-path file:/var/hoodie/ws/docker/demo/config/test-suite/complex-dag-cow.yaml \
--workload-yaml-path file:/opt/complex-dag-cow.yaml \
--workload-generator-classname org.apache.hudi.integ.testsuite.dag.WorkflowDagGenerator \
--table-type COPY_ON_WRITE \
--compact-scheduling-minshare 1
--compact-scheduling-minshare 1 \
--clean-input
--clean-output
```
A ready to use dag is available under docker/demo/config/test-suite/ that could give you an idea for long running
Few ready to use dags are available under docker/demo/config/test-suite/ that could give you an idea for long running
dags.
cow-per-round-mixed-validate.yaml
```
complex-dag-cow.yaml: simple 1 round dag for COW table.
complex-dag-mor.yaml: simple 1 round dag for MOR table.
cow-clustering-example.yaml : dag with 3 rounds, in which inline clustering will trigger during 2nd iteration.
cow-long-running-example.yaml : long running dag with 50 iterations. only 1 partition is used.
cow-long-running-multi-partitions.yaml: long running dag wit 50 iterations with multiple partitions.
```
To run test suite jobs for MOR table, pretty much any of these dags can be used as is. Only change is with the
spark-shell commnad, you need to fix the table type.
```
--table-type MERGE_ON_READ
```
But if you had to switch from one table type to other, ensure you clean up all test paths explicitly before switching to
a different table type.
```
hdfs dfs -rm -r /user/hive/warehouse/hudi-integ-test-suite/output/
hdfs dfs -rm -r /user/hive/warehouse/hudi-integ-test-suite/input/
```
As of now, "ValidateDatasetNode" uses spark data source and hive tables for comparison. Hence COW and real time view in
MOR can be tested.