You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hudi.apache.org by vb...@apache.org on 2020/01/23 23:47:53 UTC
[incubator-hudi] branch asf-site updated: [HUDI-403] Publish
deployment guide for writing to Hudi using HoodieDeltaStreamer and Spark
Data Source
This is an automated email from the ASF dual-hosted git repository.
vbalaji pushed a commit to branch asf-site
in repository https://gitbox.apache.org/repos/asf/incubator-hudi.git
The following commit(s) were added to refs/heads/asf-site by this push:
new 41754bb [HUDI-403] Publish deployment guide for writing to Hudi using HoodieDeltaStreamer and Spark Data Source
41754bb is described below
commit 41754bb31bb8656d0570371ba2283c987f9a8c22
Author: Balaji Varadarajan <va...@uber.com>
AuthorDate: Tue Jan 21 15:44:53 2020 -0800
[HUDI-403] Publish deployment guide for writing to Hudi using HoodieDeltaStreamer and Spark Data Source
---
docs/_docs/2_6_deployment.md | 130 +++++++++++++++++++++++++++++++++++++++++--
1 file changed, 126 insertions(+), 4 deletions(-)
diff --git a/docs/_docs/2_6_deployment.md b/docs/_docs/2_6_deployment.md
index 295f8e8..6fdd680 100644
--- a/docs/_docs/2_6_deployment.md
+++ b/docs/_docs/2_6_deployment.md
@@ -11,9 +11,9 @@ This section provides all the help you need to deploy and operate Hudi tables at
Specifically, we will cover the following aspects.
- [Deployment Model](#deploying) : How various Hudi components are deployed and managed.
- - [Upgrading Versions](#upgrading) : Picking up new releases of Hudi, guidelines and general best-practices
+ - [Upgrading Versions](#upgrading) : Picking up new releases of Hudi, guidelines and general best-practices.
- [Migrating to Hudi](#migrating) : How to migrate your existing tables to Apache Hudi.
- - [Interacting via CLI](#cli) : Using the CLI to perform maintenance or deeper introspection
+ - [Interacting via CLI](#cli) : Using the CLI to perform maintenance or deeper introspection.
- [Monitoring](#monitoring) : Tracking metrics from your hudi tables using popular tools.
- [Troubleshooting](#troubleshooting) : Uncovering, triaging and resolving issues in production.
@@ -23,7 +23,129 @@ All in all, Hudi deploys with no long running servers or additional infrastructu
using existing infrastructure and its heartening to see other systems adopting similar approaches as well. Hudi writing is done via Spark jobs (DeltaStreamer or custom Spark datasource jobs), deployed per standard Apache Spark [recommendations](https://spark.apache.org/docs/latest/cluster-overview.html).
Querying Hudi tables happens via libraries installed into Apache Hive, Apache Spark or Presto and hence no additional infrastructure is necessary.
+A typical Hudi data ingestion can be achieved in 2 modes. In a singe run mode, Hudi ingestion reads next batch of data, ingest them to Hudi table and exits. In continuous mode, Hudi ingestion runs as a long-running service executing ingestion in a loop.
+With Merge_On_Read Table, Hudi ingestion needs to also take care of compacting delta files. Again, compaction can be performed in an asynchronous-mode by letting compaction run concurrently with ingestion or in a serial fashion with one after another.
+
+### DeltaStreamer
+
+[DeltaStreamer](/docs/writing_data.html#deltastreamer) is the standalone utility to incrementally pull upstream changes from varied sources such as DFS, Kafka and DB Changelogs and ingest them to hudi tables. It runs as a spark application in 2 modes.
+
+ - **Run Once Mode** : In this mode, Deltastreamer performs one ingestion round which includes incrementally pulling events from upstream sources and ingesting them to hudi table. Background operations like cleaning old file versions and archiving hoodie timeline are automatically executed as part of the run. For Merge-On-Read tables, Compaction is also run inline as part of ingestion unless disabled by passing the flag "--disable-compaction". By default, Compaction is run inline for eve [...]
+
+Here is an example invocation for reading from kafka topic in a single-run mode and writing to Merge On Read table type in a yarn cluster.
+
+```java
+[hoodie]$ spark-submit --packages org.apache.hudi:hudi-utilities-bundle_2.11:0.5.1-incubating,org.apache.spark:spark-avro_2.11:2.4.4 \
+ --master yarn \
+ --deploy-mode cluster \
+ --num-executors 10 \
+ --executor-memory 3g \
+ --driver-memory 6g \
+ --conf spark.driver.extraJavaOptions="-XX:+PrintGCApplicationStoppedTime -XX:+PrintGCApplicationConcurrentTime -XX:+PrintGCTimeStamps -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/tmp/varadarb_ds_driver.hprof" \
+ --conf spark.executor.extraJavaOptions="-XX:+PrintGCApplicationStoppedTime -XX:+PrintGCApplicationConcurrentTime -XX:+PrintGCTimeStamps -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/tmp/varadarb_ds_executor.hprof" \
+ --queue hadoop-platform-queue \
+ --conf spark.scheduler.mode=FAIR \
+ --conf spark.yarn.executor.memoryOverhead=1072 \
+ --conf spark.yarn.driver.memoryOverhead=2048 \
+ --conf spark.task.cpus=1 \
+ --conf spark.executor.cores=1 \
+ --conf spark.task.maxFailures=10 \
+ --conf spark.memory.fraction=0.4 \
+ --conf spark.rdd.compress=true \
+ --conf spark.kryoserializer.buffer.max=200m \
+ --conf spark.serializer=org.apache.spark.serializer.KryoSerializer \
+ --conf spark.memory.storageFraction=0.1 \
+ --conf spark.shuffle.service.enabled=true \
+ --conf spark.sql.hive.convertMetastoreParquet=false \
+ --conf spark.ui.port=5555 \
+ --conf spark.driver.maxResultSize=3g \
+ --conf spark.executor.heartbeatInterval=120s \
+ --conf spark.network.timeout=600s \
+ --conf spark.eventLog.overwrite=true \
+ --conf spark.eventLog.enabled=true \
+ --conf spark.eventLog.dir=hdfs:///user/spark/applicationHistory \
+ --conf spark.yarn.max.executor.failures=10 \
+ --conf spark.sql.catalogImplementation=hive \
+ --conf spark.sql.shuffle.partitions=100 \
+ --driver-class-path $HADOOP_CONF_DIR \
+ --class org.apache.hudi.utilities.deltastreamer.HoodieDeltaStreamer \
+ --table-type MERGE_ON_READ \
+ --source-class org.apache.hudi.utilities.sources.JsonKafkaSource \
+ --source-ordering-field ts \
+ --target-base-path /user/hive/warehouse/stock_ticks_mor \
+ --target-table stock_ticks_mor \
+ --props /var/demo/config/kafka-source.properties \
+ --schemaprovider-class org.apache.hudi.utilities.schema.FilebasedSchemaProvider
+```
+
+ - **Continuous Mode** : Here, deltastreamer runs an infinite loop with each round performing one ingestion round as described in **Run Once Mode**. The frequency of data ingestion can be controlled by the configuration "--min-sync-interval-seconds". For Merge-On-Read tables, Compaction is run in asynchronous fashion concurrently with ingestion unless disabled by passing the flag "--disable-compaction". Every ingestion run triggers a compaction request asynchronously and this frequency [...]
+
+Here is an example invocation for reading from kafka topic in a continuous mode and writing to Merge On Read table type in a yarn cluster.
+
+```java
+[hoodie]$ spark-submit --packages org.apache.hudi:hudi-utilities-bundle_2.11:0.5.1-incubating,org.apache.spark:spark-avro_2.11:2.4.4 \
+ --master yarn \
+ --deploy-mode cluster \
+ --num-executors 10 \
+ --executor-memory 3g \
+ --driver-memory 6g \
+ --conf spark.driver.extraJavaOptions="-XX:+PrintGCApplicationStoppedTime -XX:+PrintGCApplicationConcurrentTime -XX:+PrintGCTimeStamps -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/tmp/varadarb_ds_driver.hprof" \
+ --conf spark.executor.extraJavaOptions="-XX:+PrintGCApplicationStoppedTime -XX:+PrintGCApplicationConcurrentTime -XX:+PrintGCTimeStamps -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/tmp/varadarb_ds_executor.hprof" \
+ --queue hadoop-platform-queue \
+ --conf spark.scheduler.mode=FAIR \
+ --conf spark.yarn.executor.memoryOverhead=1072 \
+ --conf spark.yarn.driver.memoryOverhead=2048 \
+ --conf spark.task.cpus=1 \
+ --conf spark.executor.cores=1 \
+ --conf spark.task.maxFailures=10 \
+ --conf spark.memory.fraction=0.4 \
+ --conf spark.rdd.compress=true \
+ --conf spark.kryoserializer.buffer.max=200m \
+ --conf spark.serializer=org.apache.spark.serializer.KryoSerializer \
+ --conf spark.memory.storageFraction=0.1 \
+ --conf spark.shuffle.service.enabled=true \
+ --conf spark.sql.hive.convertMetastoreParquet=false \
+ --conf spark.ui.port=5555 \
+ --conf spark.driver.maxResultSize=3g \
+ --conf spark.executor.heartbeatInterval=120s \
+ --conf spark.network.timeout=600s \
+ --conf spark.eventLog.overwrite=true \
+ --conf spark.eventLog.enabled=true \
+ --conf spark.eventLog.dir=hdfs:///user/spark/applicationHistory \
+ --conf spark.yarn.max.executor.failures=10 \
+ --conf spark.sql.catalogImplementation=hive \
+ --conf spark.sql.shuffle.partitions=100 \
+ --driver-class-path $HADOOP_CONF_DIR \
+ --class org.apache.hudi.utilities.deltastreamer.HoodieDeltaStreamer \
+ --table-type MERGE_ON_READ \
+ --source-class org.apache.hudi.utilities.sources.JsonKafkaSource \
+ --source-ordering-field ts \
+ --target-base-path /user/hive/warehouse/stock_ticks_mor \
+ --target-table stock_ticks_mor \
+ --props /var/demo/config/kafka-source.properties \
+ --schemaprovider-class org.apache.hudi.utilities.schema.FilebasedSchemaProvider \
+ --continuous
+```
+
+### Spark Datasource Writer Jobs
+
+As described in [Writing Data](/docs/writing_data.html#datasource-writer), you can use spark datasource to ingest to hudi table. This mechanism allows you to ingest any spark dataframe in Hudi format. Hudi Spark DataSource also supports spark streaming to ingest a streaming source to Hudi table. For Merge On Read table types, inline compaction is turned on by default which runs after every ingestion run. The compaction frequency can be changed by setting the property "hoodie.compact.inli [...]
+
+Here is an example invocation using spark datasource
+
+```java
+inputDF.write()
+ .format("org.apache.hudi")
+ .options(clientOpts) // any of the Hudi client opts can be passed in as well
+ .option(DataSourceWriteOptions.RECORDKEY_FIELD_OPT_KEY(), "_row_key")
+ .option(DataSourceWriteOptions.PARTITIONPATH_FIELD_OPT_KEY(), "partition")
+ .option(DataSourceWriteOptions.PRECOMBINE_FIELD_OPT_KEY(), "timestamp")
+ .option(HoodieWriteConfig.TABLE_NAME, tableName)
+ .mode(SaveMode.Append)
+ .save(basePath);
+```
+
## Upgrading
New Hudi releases are listed on the [releases page](/releases), with detailed notes which list all the changes, with highlights in each release.
@@ -31,7 +153,7 @@ At the end of the day, Hudi is a storage system and with that comes a lot of res
As general guidelines,
- - We strive to keep all changes backwards compatible (i.e new code can read old data/timeline files) and we cannot we will provide upgrade/downgrade tools via the CLI
+ - We strive to keep all changes backwards compatible (i.e new code can read old data/timeline files) and when we cannot, we will provide upgrade/downgrade tools via the CLI
- We cannot always guarantee forward compatibility (i.e old code being able to read data/timeline files written by a greater version). This is generally the norm, since no new features can be built otherwise.
However any large such changes, will be turned off by default, for smooth transition to newer release. After a few releases and once enough users deem the feature stable in production, we will flip the defaults in a subsequent release.
- Always upgrade the query bundles (mr-bundle, presto-bundle, spark-bundle) first and then upgrade the writers (deltastreamer, spark jobs using datasource). This often provides the best experience and it's easy to fix
@@ -54,7 +176,7 @@ For more details, refer to the detailed [migration guide](/docs/migration_guide.
## CLI
Once hudi has been built, the shell can be fired by via `cd hudi-cli && ./hudi-cli.sh`. A hudi table resides on DFS, in a location referred to as the `basePath` and
-we would need this location in order to connect to a Hudi table. Hudi library effectively manages this table internally, using `.hoodie` subfolder to track all metadata
+we would need this location in order to connect to a Hudi table. Hudi library effectively manages this table internally, using `.hoodie` subfolder to track all metadata.
To initialize a hudi table, use the following command.