by pieterh
on 05 Jul 2009 09:34
Sunday morning, and we (myself and my team) owe the Wikidot community an apology. On 24 June we migrated to a new search engine, and on 25 June Brunhilda reported that "My search box doesn't give any results, whatever you put inside…"
We found the problem rapidly but were faced with a reindexing process that took weeks to run on the whole of Wikidot. We started reindexing sites that were affected but this was still going to take many days. So now we've rewritten the reindexer. Before, it indexed each site in turn, which meant going to the database over and over, to update each piece of the index. Now it indexes the whole of Wikidot in memory, and only writes out the index at the end. This morning we tested it again and it runs in 2.5 hours. That's without indexing tags, but we'll add those back later when the rest is working and stable.
A week without search functionality is far too long, and I'm really sorry for the stress this has caused to affected site admins. Congrats to gerdami and RobElliott for proposing workarounds and to everyone who helped make it less painful.
Update: searching should work on all sites now.
To hijack this thread, we're going to look at metadata. Some ideas we had:
If you have other suggestions, they'll be welcome.
Portfolio
One meta search I'd like to do is all pages and comments by a particular user in a site, e.g. "user:pieterh".
Portfolio
Not certain if this is related, but all searches on my wiki (metroplexity.wikidot.com) now come up "Sorry, no results found for your query."
Cheers!
Kinak
I just tried a search for "+game +character" on metroplexity.wikidot.com and I get a page of results. Can you post some examples of searches that are not working? Thanks.
Portfolio
It's working properly now. Thanks for your help!
Cheers!
Kinak
Hi,
I'm having the same problem on my wiki (peoplesoft.wikidot.com). All I get when I search is "Sorry, no results found for your query."
I have always supported the great work of the people at Wikidot, but I must admit that this is somewhat frustrating and it is hard to tell what is going due to the lack of information (that I can find anyway) on the issue. As far as I know, this is the only post that talks about the problem?
Praj
Ok, here is an update. Last night the index got corrupted, and searching stopped working on all sites. We've now restored the previous index, so searches work again. Now, a technical explanation why this happened, and what we're doing to stop it happening again…
There are two indexers: one takes the whole Wikidot database and builds an index from it. This now runs in about three hours since we rewrote it over the weekend. The second updates the index with new pages. That second, incremental indexer, has a bug: when it runs out of memory, it crashes, and it corrupts the index. (Mainly because the language it's written in, PHP, has no clean way of handling a memory full situation.) Last night, it ran out of memory, crashed, and corrupted the index. From then, all searches showed "no results". So we're now rewriting the incremental indexer in Python, the language we used for the main indexer. No more crashes, no more lost indexes.
Sorry for these fumbles. I hope the new search makes it worthwhile.
Portfolio
Thanks for the update, looks like searches are working again on my site (peoplesoft.wikidot.com).
Praj
We've switched off incremental search until that indexer has been rewritten. We'll do full daily reindexes. This is a temporary workaround, I think it's better to have slightly outdated indexes than the risk of no search results at all.
Portfolio
Any word on whether the forum search is working?
I have a forum at: http://peoplesoft.wikidot.com/forum:start
However as far as I can tell, forum search isn't working at the moment. Searching on wiki pages however is working fine.
Praj
Praj,
thank you for this report. There was a small bug in the indexing system. I've already fixed that. Forums and comments will be searchable in a few hours.
EDIT: already working: http://peoplesoft.wikidot.com/search:site/a/f/q/labels%20in%20query
Piotr Gabryjeluk
visit my blog