Skip to content
Register

Is 'search first' still valid advice?

General Discussion by admin 25 replies 2.9K views
admin

Administrator OP

Administrator

#1

The search function has been a recurring topic, so let me address it directly.

Our Elasticsearch cluster was running on a single node with 4GB heap. It was not good. I have migrated us to a three-node setup with proper sharding. Search should now return relevant results within 200ms.

The old advice to "search first" was arguably cruel given the previous state of the index. It is now reasonable advice again. I have also added a "did you mean" suggestion layer for common misspellings of provider names.

Please test it and report anomalies in the infrastructure subforum.

— Admin

sudo make me a sandwich
GeorgeNmp

Member

AS64512

#2

I have been on forums where the search indexed every word including "the" and "a," producing noise. The real problem is usually query parsing. If your Elasticsearch is using the standard analyzer on technical terms like "BGP communities" or "RPKI ROA," it will split on whitespace and lose the phrase context. You may want to consider a custom analyzer with edge n-grams for AS numbers and community strings.

That said, the advice to search first remains valid because it trains newcomers to locate authoritative sources. The IRR database is searchable. PeeringDB is searchable. RPKI validity can be checked via RIPE NCC or Cloudflare's RPKI portal. Learning to search structured data is a skill distinct from searching a forum, but the habit transfers.

iBGP, eBGP, don't care, just peer
WendyBird

Member

Detail Oriented

#3

Per my earlier writeup on forum culture above in my previous analysis of forum culture (see my 2024 thread on onboarding friction, section 3.2, nested list item iv), the "search first" doctrine serves multiple functions depending on poster archetype.

Table of Contents for this post:
1. The pedagogical argument
2. The social friction argument
3. The indexing quality argument
4. Synthesis and callbacks

1. The pedagogical argument: veterans claim it builds research skills. I am skeptical. As I mentioned above, research skills require knowing *what* to search for, which presupposes domain knowledge the newcomer lacks.

2. The social friction argument: repeated questions annoy regulars. This is valid but solvable via FAQ curation, which this forum lacks. As I mentioned above, I proposed a wiki integration in 2024 that was not adopted.

3. The indexing quality argument: prior to admin's fix, search returned 47 results for "RPKI" of which 3 were relevant. The signal-to-noise ratio made "search first" functionally punitive. As I mentioned above, punitive onboarding selects for persistence rather than competence.

4. Synthesis: with functional search, the advice is now reasonable. Without it, it was gatekeeping. I will monitor whether newcomer retention improves. As I mentioned above, I track these metrics informally via thread participation rates.

RE

Member

#4

The real issue is nobody knows the search syntax. Is it "BGP communities" with quotes? OR without? Site:forum.example.com? The admin post says 200ms but not what fields are indexed.

GeorgeNmp

Member

AS64512

#5
redstone said:
The real issue is nobody knows the search syntax

This. The old index didn't support phrase queries at all, so quotes were meaningless. With the new cluster, standard Lucene syntax should work. Try 

"AS64512" +community
and see if the sharding helps.

iBGP, eBGP, don't care, just peer
hankels

Member

52 VPS and counting

#6

I searched "cheap VPS 2024" and got threads from 2019. The date filter is either missing or I'm blind. Someone point me at it?

seedbox, NAS, tape, and three offsite
admin

Administrator OP

Administrator

#7
hankels said:
Date filter is either missing

Not missing, but not exposed in the UI yet. Append 

sort=date_desc
to the URL or use 
@timestamp:[now-1y TO now]
in the query box. Full facet search is on the list after I fix the replication lag on node 3.

sudo make me a sandwich
RE

Member

#8

Following this thread because I got warned last month for posting a duplicate. My duplicate was from 2017. The old search literally could not find it.

WendyBird

Member

Detail Oriented

#9
ReaderNova said:
My duplicate was from 2017

This illustrates my social friction argument precisely. The warning system punishes users for index failures. I have cross-referenced this in my updated analysis (section 5.1, forthcoming).

CA

Member

3 #10

Can we talk about the "did you mean" layer? I typed "Linode" and it suggested "Lindoe." That's... not a company. Is it learning from our typos?

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft