Did run a balance operation.
First job failed on map63 after 4 attempts
Second job failed on the same map63
Logs are empty, what is strange it is that jobs became slower.
In both case map63 is executed by the master node which was working before.
Suspecting memory leaks I stop the cluster and started again. Fine
Run a hadoop fsck. fine
3rd time is in progress, but it looks like even slower now.
Any suggestion what to do ?