topic Re: Use a field as the search in searchmatch in Splunk Search

Use a field as the search in searchmatch

triest — Wed, 15 Nov 2017 17:33:49 GMT

I would like to use a field as the string for searchmatch, but that results in an error stating:

Error in 'eval' command: The arguments to the 'searchmatch' function are invalid.

I've attempted to recreate the issue using fake data to demonstrate the issue.

Assuming I have data like the results from the following:

 | makeresults
 | eval data="index=foo|foo|1;index=foo|bar|0;index=bar|bar|1;index=bar|foo|0;index=foo OR index=bar|foo|1;index=foo OR index=bar|bar|1"
 | fields - _time
 | eval data=split(data, ";")
 | mvexpand data
 | rex field=data "^(?<filter>[^\|]+)\|(?<index>[^\|]+)\|(?<expected>[^\|]+)$"
 | fields filter index expected

I'd really like to be able to append | eval actual=if(searchmatch(filter),1,0) and have actual match the expected column

I can copy the string from filter into search match and run them individually, but I need something that scales.

EDIT:
Just to clarify, the example I provided is fairly simple in that it is only filtering by index, but in my actual data I have all sorts of filters so something that parses out the indexes and then checks won't be sufficient for my needs.

Also in my actual data, I don't have the expected column. I added that here just to indicate if I assumed the data should match.

Re: Use a field as the search in searchmatch

MuS — Wed, 15 Nov 2017 18:13:23 GMT

Hi triest,

maybe I misunderstand your question, but how about a case() instead of searchmatch()?
Since searchmatch()takes a regex as argument you will compare against a literal filter in your example.
Your example works btw if you do it like this:

 | eval actual=if(searchmatch("filter"),1,0)

but this will match all events since you have filter in all example events.

Would a case() like this do the thing?

 | eval actual=case(expected="1", "Yes", expected="0", "No", 1=1, "unknown")

Again I may understand the requirement completely wrong ¯\_(ツ)_/¯ 😉

cheers, MuS

Re: Use a field as the search in searchmatch

triest — Wed, 15 Nov 2017 18:33:25 GMT

@MuS,

You are completely correct that in this simple case that would work. I added the expected to show if I thought the filter should match the event or not; in the real data set I wouldn't have that. I'll try to clarify the question; I really appreciate you trying.

Re: Use a field as the search in searchmatch

rojyates — Thu, 31 May 2018 01:49:11 GMT

Hi @triest,

If I understand correctly, then try appending the following as I think this may do what you want:

| eval actual=if(searchmatch("expected=1"), filter, "")

Note the quotes inside the searchmatch() - without which you'd get "Error in 'eval' command."

I've appended it to your query, which gave the following output when I ran it, with the 'actual' column showing the filters that matched, or empty if they didn't:

filter                  index   expected    actual
index=foo               foo     1           index=foo
index=foo               bar     0           
index=bar               bar     1           index=bar
index=bar               foo     0           
index=foo OR index=bar  foo     1           index=foo OR index=bar
index=foo OR index=bar  bar     1           index=foo OR index=bar

Is this what you were trying to do?

Re: Use a field as the search in searchmatch

wrangler2x — Fri, 22 Jun 2018 17:25:43 GMT

searchmatch will not allow a field to be used in place of string. However, the match function of eval will, and match can be made to behave like searchmatch very easily!

| eval searchHits=if(match(_raw,"Type=Error"),1,0)

is the same as:

| eval searchHits=if(searchmatch("Type=Error"),1,0)

Further, match will support the regex pipe, so you can OR as well. Here's an example of both:

... | eval typeField = "Type=Error|Type=Warning"
| eval searchHits=if(match(_raw,typeField),1,0)
| stats count as EventCount sum(searchHits) AS searchHits by Type