stocator icon indicating copy to clipboard operation
stocator copied to clipboard

Spark History Server unable to read rolling logs from COS using stocator

Open shivprakashy opened this issue 3 years ago • 0 comments

When Spark History Server (SHS) is configured to use stocator to read rolling files it fails. Below are the following configurations done (SHS read configurations from spark-default.conf)

Non-Working Configuration

spark.hadoop.fs.cos.hbdevshivcos.endpoint https://s3.us-south.cloud-object-storage.appdomain.cloud
spark.hadoop.fs.cos.hbdevshivcos.access.key <hmac-access-key>
spark.hadoop.fs.cos.hbdevshivcos.secret.key <hmac-secret-key>
spark.hadoop.fs.stocator.cos.impl com.ibm.stocator.fs.cos.COSAPIClient
spark.hadoop.fs.cos.impl com.ibm.stocator.fs.ObjectStoreFileSystem
spark.hadoop.fs.stocator.scheme.list cos
spark.hadoop.fs.stocator.cos.scheme cos
spark.history.fs.logDirectory cos://hbdev-shiv.hbdevshivcos/spark-events
spark.eventLog.dir cos://hbdev-shiv.hbdevshivcos/spark-events

Working Configuration ($ start-history-server --properties-file shs.properties) content of shs.properties below

spark.hadoop.fs.s3a.endpoint=https://s3.us-south.cloud-object-storage.appdomain.cloud
spark.hadoop.fs.s3a.access.key=<testaccesskey>
spark.hadoop.fs.s3a.secret.key=<testsecretkey>
spark.hadoop.fs.s3a.impl=org.apache.hadoop.fs.s3a.S3AFileSystem
spark.hadoop.fs.s3a.scheme.list=s3a
spark.hadoop.fs.s3a.scheme=s3a
spark.history.fs.logDirectory=s3a://hbdev-shiv/spark-events
spark.eventLog.dir=s3a://hbdev-shiv/spark-events

What differs in s3a and COS implementation to read rolling log files needs to be investigated and fixed. More details to follow...

shivprakashy avatar Dec 23 '22 16:12 shivprakashy