This is the key error.
How did you install Elasticsearch? Did you use the .rpm installer or the .tar.gz installer?
Has this Elasticsearch node been running for more than 10 days before you attempted to open the ML job?
If you are using the .tar.gz installer and your Elasticsearch node has been running for more than 10 days then this advice is probably relevant.
If you are using the .rpm installer then we need to work out why the systemd private temp directory functionality we added in 6.4 didn't work as expected.