Multiverse Feedback: Cardlist | Visual spoiler | Export | Booster | Comments | Search | Recent activity |
Mechanics | Upcoming releases | Skeleton |
CardName: Application Errors Cost: Type: Error Pow/Tgh: / Rules Text: Frequently getting these Application Errors. On some days it's really bad. Loading a cardset or page requires many tries. If working on a new card, it could mislead you to resubmit several times, thus creating multiple copies of the same card. I assume this server runs on ASP.NET because I only ever see these errors on ASP served pages. Quite often, actually, not just this site. Flavour Text: Set/Rarity: Multiverse Feedback Common Frequently getting these Application Errors. On some days it's really bad. Loading a cardset or page requires many tries. If working on a new card, it could mislead you to resubmit several times, thus creating multiple copies of the same card.
I assume this server runs on ASP.NET because I only ever see these errors on ASP served pages. Quite often, actually, not just this site. |
History: [-] Add your comments: |
It's Ruby. But yes, it does seem to be getting worse again.
Indeed, the site runs on Ruby on Rails. The Application Errors are certainly annoying. I have vague metrics of their frequency and I do apologise that it has been somewhat on the rise.
What's particularly annoying is that there's very little I can do about it. My options are:
1) Pay Heroku a lot of money to slightly increase the number of hamsters powering the site - this is an option I apply temporarily if the site gets linked from somewhere high-traffic, but it's too expensive to leave on all the time.
2) Move to a different hosting provider. I looked at quite a few but none of them provide me as good hosting as Heroku provide for free.
3) Move to a self-hosted VM somewhere. This is my preferred option, somewhere like http://edis.at , but it does need quite a lot of time on my part to get it set up, and I've not had the time to put that much time into Multiverse since this came up... pretty much a year ago, in fact. Sorry.
Yeah, that was the impression I had with hosting. I've no better suggestion, I agree the current compromise is probably the only option. (I agree the errors are annoying, but I don't really mind, for me they're usually only very temporary.)
Would it be worth spelling out what regular users would have to contribute to pay for better Heroku hosting? I assume most people wouldn't think it was worth it, but at least you could say "if it's worth this much to anyone, then we can do it, else we're stuck with the status quo" rather than just "no"? (Admittedly accepting money is awkward for lots of other reasons.)
In general this server is actually very slow to load any page. Is it because the database is growing too large?
Yea money is a sensitive issue. But let's at least see what the pricing is for better plans.
Well, several of us can hand over cash. The main problem is that multiverse (possibly the framework multiverse is built on) can only process one thing at once. So when someone is refreshing the recent changes list; submitting a card change grinds to a halt; and vice versa. My understanding is that takes both money and a lot of work to fix. Growing to the size where it becomes a problem is a good problem to have, though :)
I wonder - could each cardset (or chunk of cardsets) be sharded off to a different host+database? The recent changes page would become more like a local feed aggregator with all those problems, though.
Vitenka is right - the issue is parallelism. Because I'm on Heroku's cheapo tier, the whole site is hosted by one process. This does indeed mean that any slow operation (editing cards isn't too bad - the main culprits are viewing visual spoilers or full cardset cardlists) makes everyone else using the site wait until it's finished.
It doesn't take any work at all to add a second Heroku process ("web dyno", as they call it), which would be enough parallelism to fix the issue Vitenka discusses, but it's absurdly expensive - $35 per month.
I've looked at remote-hosted databases a little bit in the past; not found anything that seemed feasible to plug into the Heroku structure I've got.
Oh; I thought adding a dyno didn't actually remove the lock, and so didn't actually help. There was a BIG kerfluffle about it when people realised they'd basically been paying for nothing a few months ago.
Oh, I remember the kerfuffle. I didn't think that was the issue, but I don't recall for sure.