JSON array and stats command

[TSplunk]

Hi,

I am having trouble generating a stats report based on JSON data containing an array.  I want to produce the following report:

ErrorCode	ErrorMessage	Count
212	The image quality is poor	1
680	The image could not be found	1
809	Document not detected	1

When I do the stats command, I do not get any results:

| spath input=jsondata |stats count by "embedded.metadata.data.results{}.notifications.*"

I have to know the error code value in the array in order to get any stats output.  For example:

| spath input=jsondata |stats count by "embedded.metadata.data.results{}.notifications.809"

Result:

embedded.metadata.data.results{}.notifications.809	count
Document not detected	1

Here example of the json data

{
"_embedded": {
  "metadata": {
     "environment": {
         "id": "6b3dc"
      },
"data": {
   "results": [
    {
     "documentId": "f18a20f1",
     "notifications": {
         "212": "The image quality was poor"
       }
   },
{
   "documentId": "f0fdf5e8c",
   "notifications": {
      "680": "The image could not be found"
     }
},
{
   "documentId": "95619532",
   "notifications": {
      "809": "Document not detected"
    }
}
]
}
}
}
}

Thanks in advance for any assistance!!

[yuanliu]

This is an interesting challenge because of the leading underscore (_) in the root node key.  spath can't seem to recognize the path as is. (Could be a subtle bug.)  One potential way to work around this is to use text replacement to get rid of the underscore. But I prefer a more syntactic method.  In Splunk 9, you can do fronjson with an arbitrary prefix so spath can do its job. (You can also use fromjson repeatedly.  But with deep paths like this problem, that's undesirable.)

| fromjson jsondata prefix=my
| spath input=my_embedded path=metadata.data.results{}
| mvexpand metadata.data.results{}
| spath input=metadata.data.results{}
| foreach notifications.*
    [eval ErrorCode = if(isnull(ErrorCode), "<<MATCHSTR>>", ErrorCode), ErrorMessage = mvappend(ErrorMessage, '<<FIELD>>')]
| stats count by ErrorCode ErrorMessage

Your sample data results in

ErrorCode	ErrorMessage	count
212	Document not detected	1
212	The image could not be found	1
212	The image quality was poor	1

This is an emulation you can play with and compare with real data

| makeresults
| eval jsondata = "{
\"_embedded\": {
  \"metadata\": {
     \"environment\": {
         \"id\": \"6b3dc\"
      },
\"data\": {
   \"results\": [
    {
     \"documentId\": \"f18a20f1\",
     \"notifications\": {
         \"212\": \"The image quality was poor\"
       }
   },
{
   \"documentId\": \"f0fdf5e8c\",
   \"notifications\": {
      \"680\": \"The image could not be found\"
     }
},
{
   \"documentId\": \"95619532\",
   \"notifications\": {
      \"809\": \"Document not detected\"
    }
}
]
}
}
}
}"

Note: Once you get past that leading underscore in JSON path, you can then use the text manipulation method proposed by @ITWhisperer (with a small path correction), like this

| fromjson jsondata prefix=my
| spath input=my_embedded path=metadata.data.results{}
| mvexpand metadata.data.results{} ``` not metadata.data.results{}.notifications ```
| rex field=metadata.data.results{} "\"(?<ErrorCode>\d+)\":\s*\"(?<ErrorMessage>[^\"]*)\""
| stats count by ErrorCode ErrorMessage

But text manipulation on structured data is not as robust because a developer/software can always change format in the future without altering semantics.

[ITWhisperer · Accepted solution]

| spath _embedded.metadata.data.results{}.notifications output=results
| mvexpand results
| rex field=results "\"(?<ErrorCode>\d+)\":\s*\"(?<ErrorMessage>[^\"]*)\""
| stats count by ErrorCode ErrorMessage