Skip to main content
Solved

Cribl Search Performance Impact When Sending Only _raw Field Versus Extracted Fields

  • Dec 24, 2025 8:52 PM UTC
  • 7 replies
  • 168 views

CarlosM
This message originated from Cribl Community Slack.
Click here to view the original link.

Hello team! I’d like to understand the difference between sending only the _raw field to Lake, and sending _raw along with some extracted/parsed fields. I assume it will affect the size of the event, but more importantly, how does it affect the performance in Cribl Search? Additionally, are there any recommendations for using JSON/Parquet data format?

Best answer by jinsea

If you send in _raw and then some parsed fields from that _raw you are going to be duplicating data. Inside the cribl lake you can do things like dataset="staging" AND source_name="syslog_source" | extend raw = parse_json(_raw) in your KQL to more easily see the _raw data :smile:
View original

7 replies

  • Inspiring
  • Answer
  • Dec 24, 2025 8:52 PM UTC
If you send in _raw and then some parsed fields from that _raw you are going to be duplicating data. Inside the cribl lake you can do things like dataset="staging" AND source_name="syslog_source" | extend raw = parse_json(_raw) in your KQL to more easily see the _raw data :smile:

  • Inspiring
  • Dec 24, 2025 8:52 PM UTC
also I think someone mentioned parquet might be better performant but we have only done json so far. :smile:

CarlosM
  • Author
  • New Participant
  • Dec 24, 2025 8:52 PM UTC
awesome, thanks for the info @user! :slightly_smiling_face: It can be more of a Cribl Search question, but do you know what is better for searching performance? For example:
dataset="staging" AND source_name="syslog_source" 
| extend raw = parse_json(_raw)
| where raw.ip=="1.1.1.1"
vs
dataset="staging" AND source_name="syslog_source"  ip=="1.1.1.1"

  • Employee
  • Dec 24, 2025 8:52 PM UTC
Honestly, it depends on the size of the dataset being searched, but they should be fairly close in performance. Ideally, you've got your highly searched fields as top-level events.

  • Employee
  • Dec 24, 2025 8:52 PM UTC
In our current implementation, JSON has better compression, but if your events contain a lot of repeated values in fields (e.g. Kubernetes logs), you'll see smaller file sizes by leveraging Parquet.

  • Employee
  • Dec 24, 2025 8:52 PM UTC
If you end up using Parquet, have those commonly searched fields as top-level fields; Search will be able to only pull out those fields requested rather than all fields. See here for more detail: https://cribl.io/blog/cribl-search-parquet-pushdowns-smooth-like-butter/

CarlosM
  • Author
  • New Participant
  • Dec 24, 2025 8:52 PM UTC
Great info, thanks for your help @user!