Better handling of tabs vs spaces in elasticsearch.yml

[ppf2]

Consider the following example, the settings below look fine to the human eye:

But if we show the actual hidden characters, the hosts line is actually using a tab instead of spaces to indent the entry:

$
discovery:$
  zen:$
    ping:$
      multicast:$
        enabled: false$
      unicast:$
^Ihosts: ["localhost:9300","localhost:9301"]$
$

As a result, ES silently ignores the hosts entry observed in debug logging:

[2014-10-28 10:49:01,271][DEBUG][discovery.zen.ping.unicast] [Franz Kafka] using initial hosts [], with concurrent_connects [10]

While yaml does not consider tabs valid, it would be nice if there is additional validation implemented in ES to catch this.

[johtani]

Now, replace tab with two spaces.
https://github.com/elasticsearch/elasticsearch/blob/master/src/main/java/org/elasticsearch/common/settings/loader/YamlSettingsLoader.java#L41-41

[clintongormley]

@johtani thanks for spotting that. that's a bug. you can't assume two spaces for a tab, because the user might be using 4 spaces etc, which messes up their configuration.

Tabs are illegal in YAML and should throw an error.

[wwken]

Yep I agree with @clintongormley . It should throw an exception instead.