Difference between using xmlkv and KV_MODE=xml

[pasokkum]

Hi,
I am getting inputs in the form of xml files.. To extract the fields from xml, do i need to use xmlkv in search or KV_MODE=xml in props.conf?
Which one gives better performance and what is the difference between the two?

[bmunson_splunk]

The underlying code for both is the same so the performance won't be much different.  The difference is when do you want these fields extracted and when don't you. 

KV_MODE=xml will be always done for that sourcetype. 
xmlkv will only be done when you use it in a search string. 
So if you always want all of the fields to be extracted use KV_MODE but if you only want the fields to be occasionally extracted use xmlkv in your search string.
If you only want one or two fields from a big xml file, it might be better to extract them using normal regex extraction

Another use for xmlkv is when not all of your event is clean xml. KV_MODE would fail and not give you the fields. In a search, you can use an eval or rex to extract and clean the xml portion and then run xmlkv on that. 

[ssadanala1]

As per splunk documentation here is the difference

The xmlkv command automatically extracts fields from XML-formatted data. For example, if the XML contains the following in its _raw data . xmlkv command needed to be invoked by the user to get the fields.

KV_MODE = xml is a search time field extraction that happens before the results are fetched to the user .This setting automatically bring the field extractions automatically.

Hence KV_MODE =xml is the best practice and performance wise there wont be much difference (not sure)