allixender
commented
9 years ago Update:
I'll use this issue for another Cassandra connection problem I stumbled over. The spark submit for the CassandraIngestCommand includes the parameter --host:
### INGEST SPATIAL GEOTIFF IN CASSANDRA ###
# geotrellis-spark JAR. Shouldn't have to change this one if running in the root folder (remember to run ./sbt "project spark" assembly)
JAR=/home/akmoch/dev/build/geotrellis-cassandra/spark/target/scala-2.10/geotrellis-spark-assembly-0.10.0-SNAPSHOT.jar
# Directory with the input tiled GeoTIFF's, need to comply to ../ne*.tif rule **weird?*
# INPUT=file:/home/akmoch/dev/build/geotrellis-cassandra/raster-test/data/one-month-tiles-tiff
INPUT=/home/akmoch/dev/build/geotrellis-cassandra/raster-test/data/one-month-tiles-tiff
export SPARK_HOME=/home/akmoch/dev/spark-1.2.2-bin-hadoop2.4
# Table to store tiles
TABLE=nexmonth
# Name of the layer. This will be used in conjunction with the zoom level to reference the layer (see LayerId)
LAYER_NAME=nexmonth
# This defines the destination spatial reference system we want to use
# (in this case, Web Mercator)
CRS=EPSG:3857
# true means we want to pyramid the raster up to larger zoom levels,
# so if our input rasters are at a resolution that maps to zoom level 11, pyramiding will also save
# off levels 10, 9, ..., 1.
PYRAMID=true
# true will delete the HDFS data for the layer if it already exists.
# CLOBBER=true
CLOBBER=false
# true will delete the HDFS data for the layer if it already exists.
# CLOBBER=true
CLOBBER=false
# We need to remove some bad signatures from the assembled JAR. We're working on excluding these
# files as part of the build step, this is a workaround.
zip -d $JAR META-INF/ECLIPSEF.RSA
zip -d $JAR META-INF/ECLIPSEF.SF
# Run the spark submit job
# sbt, sbt assembly to gen JAR
$SPARK_HOME/bin/spark-submit \
--class geotrellis.spark.ingest.CassandraIngestCommand \
$JAR \
--host localhost --keyspace ingest \
--crs $CRS \
--pyramid $PYRAMID --clobber $CLOBBER \
--input $INPUT \
--layerName $LAYER_NAME \
--table $TABLEHint! Cassandra keyspace ingest was hardcoded here, should probably factor out as VARIABLE too.
Although obviously --host localhost is provided and in geotrellis-spark CassandraIngestCommand.scala(line 37) sparkConf.set("spark.cassandra.connection.host", args.host) seems to promote this paramter to the spark context which in turn would be evaluated from the spark-cassandra-connector ...
Exception in thread "main" java.io.IOException: Failed to open native connection to Cassandra at {161.65.43.142}:9042.. it consistently want to connect the external host ip, but not localhost :-( With external eth0 IP, spark-master bound to external and cassandra listen_address to external, the CassandraIngest worked now. What a journey. This IP handling needs to be more reliable, particularly for test runs on localhost.
GSoc2015 GeoTrellis Cassandra tracker for shared scratch documentation
If cassandra server is not running, geotrellis-spark (feature-cassandra as of now) tests will fail.
Update 26.05.2015 This is not directly true actually. In fact, the geotrellis-spark cassandra tests use the
useCassandraConfigembedded service from the datastax spark-cassandra-connector, which actually starts a cassandra instance in the JVM, just like we want. However, it has a timeout of 10 seconds and if the cassandra daemon isn't responsive than the test will fail. Also the tests cannot be run in parallel! Spark-Cassandra-Connector EmbeddedCassandra.scalaWorkarounds/solutions, either define switch to skip Cassandra tests if no cassandra instance is available (keyspace config?) or ideally launch embedded instance, similar to accumulo tests
Some links I'll follow up on, where projects use Embedded Cassandra testing environments: