Keeping local modified elasticsearch.yml on dpkg upgrade causes ES start to fail

[slovdahl]

Elasticsearch version (bin/elasticsearch --version): 6.0.0

Plugins installed: []

JVM version (java -version): OpenJDK 1.8.0_151

OS version (uname -a if on a Unix-like system): Linux desk 4.10.0-35-generic #39~16.04.1-Ubuntu SMP Wed Sep 13 09:02:42 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux

Description of the problem including expected versus actual behavior:
During the upgrade from ES 5.6.1 to 6.0 with the deb package, dpkg asked if I wanted to keep my local elasticsearch.yml, or replace it with the updated default one. I chose to retain my local one (containing only two extra rows in the end of the file, a custom cluster.name and network.host).

After the upgrade, I tried to manually restart ES. But it refused to start, because it tried to create log files in /usr/share/elasticsearch/, and not /var/log/elasticsearch/ as previously.

Copying the updated path.logs setting from the new default elasticsearch.yml and starting ES again fixes the problem, but this seems lika a very intrusive change to me, and I think it's going to cause a lot of pain for users upgrading their ES nodes to 6.0.0.

Steps to reproduce:

1. Download and install ES 5.x.
2. Modify /etc/elasticsearch/elasticsearch.yml
3. Download and upgrade to ES 6.0.0.
4. Choose to retain the local version of /etc/elasticsearch/elasticsearch.yml
5. Restart ES, it won't start.

Provide logs (if relevant):
es-upgrade.log
es-startup.log

[jasontedor]

I am sorry that you experienced trouble but this is covered in the breaking changes for 6.0. Additionally, it's a good practice to compare differences in package-shipped configuration files when the installation presents to you that there are differences.