I'm trying to deploy Aurora on my Mesos Cluster but when it comes to run aurora-scheduler
, it always fails showing a NullPointerException. Here is my setup:
- Docker container based on Ubuntu 14.04
- OpenJDK8
- Mesos 0.22.1
- Aurora 0.10.0
- 3 ZooKeeper instances
- 1 Mesos Master
- 4 Mesos Slaves
I'm pretty sure the ZooKeeper/Mesos setup is up and running since I can use the Marathon framework without any problems.
Here is my scheduler.sh
that I try to run:
#!/bin/bash
# Cluster-specific environment variables
CLUSTER_NAME=test
ZK1=172.24.32.1:2181
ZK2=172.24.32.5:2181
ZK3=172.24.32.9:2181
ZK_QUORUM=$ZK1,$ZK2,$ZK3
AURORA_HOME=/usr/local/aurora-scheduler
# Flags controlling the JVM.
JAVA_OPTS=(
-Xmx2g
-Xms2g
# GC tuning, etc.
)
# Flags controlling the scheduler.
AURORA_FLAGS=(
-http_port=${PORT_WEBUI}
-backup_dir=/data1/aurora_scheduler_backup
-cluster_name=${CLUSTER_NAME}
-mesos_master_address=zk://${ZK_QUORUM}/mesos
-serverset_path=/aurora/scheduler
-thermos_executor_path=/dev/null
-zk_endpoints=${ZK_QUORUM}
)
# Environment variables controlling libmesos
export GLOG_v=1
export LIBPROCESS_PORT=${PORT_LIBPROCESS}
JAVA_OPTS="${JAVA_OPTS[*]}" exec "$AURORA_HOME/bin/aurora-scheduler" "${AURORA_FLAGS[@]}"
I have followed the installation steps available on the Aurora official documentation.
I uploaded the full output on PasteBin.
Your scheduler.sh appears to lack native log initialization flags. Aurora uses Mesos replicated log to store its state. You'd need to at least configure native_log_zk_group_path and native_log_file_path options to tell Aurora about replicated log. More here: https://github.com/apache/aurora/blob/master/docs/storage-config.md