Better handling of inline scripts in ingest

[clintongormley]

Anywhere that scripts are used, we allow writing an inline script in a short form, eg:

{ "script": {
    "lang": "painless",
    "inline": "ctx.foo='bar'
}}

can also be written as:

{ "script": "ctx.foo='bar'"}

If I try this in ingest:

POST _ingest/pipeline/_simulate
{
  "pipeline": {
    "processors": [
      {
        "script": "ctx.foo='bar'"
      }
    ]
  },
  "docs": [
    {
      "_index": "t",
      "_type": "t",
      "_id": "1",
      "_source": {}
    }
  ]
}

I get:

     {
        "type": "class_cast_exception",
        "reason": "java.lang.String cannot be cast to java.util.Map"
      }

While I understand that all processors expect an object and so the short form may not be supported (although it'd be nice to make an exception for scripts), we should at least have an exception which is easier to understand.

[Dimitrees]

Hello, I am new to both your project and working with GitHub in general. Can I work on resolving this issue since its labeled as Low Hanging Fruit?

Thanks

[asettouf]

Hello,

@Dimitrees are you still working on it, or can I pick it up?

Thanks,

Adonis

[Dimitrees]

Its yours, sorry for not unassigning myself earlier.

[asettouf]

Here is a PR regarding this issue. As it is one of my first PR, I look forward to your feedback.

[liketic]

@talevy Can I work on this? Thanks. 😄

[talevy]

yeah @liketic!

If you look at my comment here: #24410 (review), that is what I had in mind for a possible solution.

[liketic]

@talevy Thanks for your help. Seems like this issue is much complexer than I thought. We read the processor configs as List<Map<String, Map<String, Object>>>, and in all methods, we assume the value of each key in a Map is a Map, however, script allows the value to be a String. That's the problem. So I think we can resolve this through:

1. Read processor configs as List<Map<String, Object>>. But for method readProcessor, we need to detect if the value object is a Map or a String. If it's a String, need to normalize it to a Map.
2. In ScriptProcessor, parse Script by Script.parse as you mentioned. and clear all script related properties.

I didn't find another place can call the readScript method, because the script parsing is done in ScriptProcessor and ScriptProcessor is created in factory. So seems like this resolution is another work around. I'm not a expert to es, so I'm not sure if this resolution is make sense to you.
Happy to know your thoughts. Thanks very much!

[liketic]

@talevy I opened a PR as I thought. Please add your comments or close it if it does not make sense. Thanks. 😄