- 53 months ago
I hear you, loud and clear. I know we need forum search. It's coming.
I’ve got forum search partially complete. I’ll be honest here - we’re using AWS cloudsearch for indexing and retrieval… it’s ok. We use it for part search and for our completed builds search.
I thought it was passable. Until I spoke with a friend who got his masters in Information Retrieval (i.e. the specialization you want if you're going into search engine stuff) and realized that our part search is woefully naively implemented and could do much much better with a specialized non-AWS solution.
One thing with AWS is that our site is hosted on Linode, and for whatever reason every once in a while the Linode internal DNS likes to go out to lunch and forget where the AWS cloud search endpoints are. Really frustrating until we bypass the internal Linode DNS. Elasticsearch was a possible option, but they recently had a remote code execution vulnerability that kinda gave me the heebie jeebies, so I think we’ll not go that route for now.
Anyhow, forum search poses some interesting usability questions. Do we index just the topics? Or the topics and all the comments in the topic? If both, does the topic text convey more weight in the search results? Or do we need to expose some sort of search switch that says only search the topics, or topics + comments, etc.
Another aspect is that our forums and commenting system here were built in-house by yours truly. So in traditional fashion, I designed it to first work. Then if it got traction, I'd freak out a bit and figure out how to make it scale without taking down the servers. I think we've just finished that second part - the forum adoption surpassed what I thought it would. We had numerous growing pains along the way - from heavily nested comments which I still don't think we're handling right, to boatloads of spam from unnamed regions. Those cause distractions that eat up dev time and delay forum search.
Long story short, I think we've got a lot of the performance issues with forums and comments settled (reasonably so, not where I want it in the end, but a lot better than it used to be). Once we figure out what we want the forum search usability to be like (indexing topics, topics + comments, configurable, etc) then I think the actual implementation will be modestly fast.