Skip to content

Optimize memory footprint for Spark Ingestion Job #1265

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pyalex merged 1 commit into from
Jan 14, 2021
Merged

Optimize memory footprint for Spark Ingestion Job #1265

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions spark/ingestion/src/main/scala/feast/ingestion/BatchPipeline.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,7 @@ object BatchPipeline extends BasePipeline {
}

val validRows = projected
.mapPartitions(metrics.incrementRead)
.map(metrics.incrementRead)
.filter(rowValidator.allChecks)

validRows.write
Expand All @@ -85,7 +85,7 @@ object BatchPipeline extends BasePipeline {
case Some(path) =>
projected
.filter(!rowValidator.allChecks)
.mapPartitions(metrics.incrementDeadLetters)
.map(metrics.incrementDeadLetters)
.write
.format("parquet")
.mode(SaveMode.Append)
Expand Down
4 changes: 2 additions & 2 deletions spark/ingestion/src/main/scala/feast/ingestion/StreamingPipeline.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,7 +107,7 @@ object StreamingPipeline extends BasePipeline with Serializable {
implicit def rowEncoder: Encoder[Row] = RowEncoder(rowsAfterValidation.schema)

rowsAfterValidation
.mapPartitions(metrics.incrementRead)
.map(metrics.incrementRead)
.filter(if (config.doNotIngestInvalidRows) expr("_isValid") else rowValidator.allChecks)
.write
.format("feast.ingestion.stores.redis")
Expand All @@ -122,7 +122,7 @@ object StreamingPipeline extends BasePipeline with Serializable {
case Some(path) =>
rowsAfterValidation
.filter("!_isValid")
.mapPartitions(metrics.incrementDeadLetters)
.map(metrics.incrementDeadLetters)
.write
.format("parquet")
.mode(SaveMode.Append)
Expand Down
24 changes: 16 additions & 8 deletions spark/ingestion/src/main/scala/feast/ingestion/metrics/IngestionPipelineMetrics.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,20 +22,28 @@ import org.apache.spark.sql.Row

class IngestionPipelineMetrics extends Serializable {

def incrementDeadLetters(rowIterator: Iterator[Row]): Iterator[Row] = {
val materialized = rowIterator.toArray
def incrementDeadLetters(row: Row): Row = {
if (metricSource.nonEmpty)
metricSource.get.METRIC_DEADLETTER_ROWS_INSERTED.inc(materialized.length)
metricSource.get.METRIC_DEADLETTER_ROWS_INSERTED.inc()

materialized.toIterator
row
}

def incrementRead(rowIterator: Iterator[Row]): Iterator[Row] = {
val materialized = rowIterator.toArray
def incrementRead(row: Row): Row = {
if (metricSource.nonEmpty)
metricSource.get.METRIC_ROWS_READ_FROM_SOURCE.inc(materialized.length)
metricSource.get.METRIC_ROWS_READ_FROM_SOURCE.inc()

materialized.toIterator
row
}

def incrementRead(inc: Long): Unit = {
if (metricSource.nonEmpty)
metricSource.get.METRIC_ROWS_READ_FROM_SOURCE.inc(inc)
}

def incrementDeadLetters(inc: Long): Unit = {
if (metricSource.nonEmpty)
metricSource.get.METRIC_DEADLETTER_ROWS_INSERTED.inc(inc)
}

private lazy val metricSource: Option[IngestionPipelineMetricSource] = {
Expand Down
2 changes: 1 addition & 1 deletion spark/ingestion/src/main/scala/feast/ingestion/stores/redis/RedisSinkRelation.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ class RedisSinkRelation(override val sqlContext: SQLContext, config: SparkRedisC
// repartition for deduplication
val dataToStore =
if (config.repartitionByEntity)
data.repartition(config.entityColumns.map(col): _*)
data.repartition(config.entityColumns.map(col): _*).localCheckpoint()
else data

dataToStore.foreachPartition { partition: Iterator[Row] =>
Expand Down