[ home ] [ site / arch ] [ pony / oat / anon ] [ rp / art ]

/site/ - Site Issues

The board for discussing site related issues, questions, concerns, and suggestions.
Name
Email
Subject
Comment
File
Flags
Password (For file deletion.)

Site maintenance in progress! Posts made now may be lost.

Ponychan-MLPchan Merger >>>/site/15219

File: 1376717084318.jpg (393.62 KB, 1303x657, cloud error.jpg)

Sersys!CyQeb5KTDM 10304

Recently I get more and more of these. When I try to post I get a 520 error screen then I get redirected here. When I get this I ask around if mlpchan is down but everyone seems to be fine and it's only effecting me. I looked up error 520 and found this:

One potential cause could be that your web server is sending a response header that exceeds CloudFlare's maximum response header size. This could be the case if you're sending an abnormally high number of cookies for example.

I'm not much of an IT guy so if someone could translate this for me I would appreciate it.

Before recently it only happened when I tried to post 2 posts in different threads at the same time.
I'm using windows 8, mozilla and have pppoe internet connection

Any ideas for trouble shooting?

The Vulture!3bqGraff0U 10307

>>10304
I've had the same problem.
Didn't put any thought into it, because after five or so minutes, it is back.

Macil!/5s/Techmk ## Admin 10309

The site server is having a lot of disk errors. I'll work on it.

Vizio!aotNkSnRIQ 10310

>>10309
The problem is still occurring especially in the >BLOC thread.
If you can also move the >BLOC thread link on the genera directory to this that would be nice.
https://mlpchan.net/anon/res/153744.html#153744

Anonymous 10311

It's happening to me as well, getting really annoying.

BMO 10312

I seem to be getting hit by this too.
Just to add more of an idea of the spread, not to really complain.

Anonymous 10313

It's happening to me as well, getting really annoying.

Anonymous 10314

This keeps happening to me as well.

And It is also kicking my setting changes out after a few seconds. And I am getting annoyed with resetting them.

Macil!/5s/Techmk 10315

>>10314
The settings are done client-side, so that would be a separate issue. Is the settings menu closing after you open it, are your settings reverting back in front of you seconds later while the menu is still open, do they reset when you browse to another page, or do they reset after you close your browser?

Anonymous 10316

>>10315

I got it working again, somehow. Thanks for the help anyway.

Anonymous 10329

>>10309
Are they disk errors, or are they concurrency errors? I have a suspicion that they're the latter, and that I have something to do with it.

For the last 6 weeks, I've been coding an archive for the site, and most of that time has been spent on creating a high-speed replicator for the threads. Today was my first time testing it side-wide, and I found that bandwidth was my limiting factor - which means it was a success. Unfortunately, it seems I made a non-trivial assumption about a load balancer existing between the client and the database. Without being too verbose, SQL databases aren't concurrent - they lock while a process is pending on a cluster of data. I'm guessing some people were trying to post while the process was running in certain threads, so they got a lock. The 520 error could be explained by the large number of requests to the server.

Before jumping to conclusions, I would like to reiterate that 1) this is hypothetical and 2) this is my first/last day running this version of the script, so if you've seen this screen before today, it wasn't me. I was trying to delay the archive announcement until the site was up, but it's more important to allay concerns about the site and help Mac troubleshoot. Again, if this is the cause, I apologize for the inconvenience and it shouldn't happen again.

P.S. Like Mac said, >>10314 is a separate issue.

P.S.S. inb4 DDoS. A standard DDoS works by making a large number of long-polling requests to the server, blocking additional requests. These requests were designed to be as quick as possible to prevent blocking, so CloudFlare's standard anti-DDoS mechanism recognized them as legitimate.

tl;dr There's probably nothing wrong with the server, and this was likely a one-time thing.

10330

Oh, and I'll do my best to work more closely with Mac in the future to avoid these kinds of fumbles (if it was me). Again, pardon the interruption.

Macil!/5s/Techmk 10341

>>10329
It wouldn't be a concurrency error: few locks are used (I guess there are very brief ones used internally to make posts not happen at the same instant), and the site is rendered as static files so viewing threads and boards doesn't trigger any database activity anyway. The bandwidth used by the archiver doesn't seem to compare to the regular server backups I do, so don't worry about that too much.

Now the fact that it requested many different files from the server, including many old ones that probably hadn't been viewed in a while that happened to be on old bad disk sectors (and hadn't been recently cached downstream by CloudFlare) probably caused the server to be affected by more disk errors than usual. Because those files would have gotten cached by CloudFlare again recently (and I think the machine repairs some types of bad sectors when it finds them, and your crawler has its copies of those files now) should mean the server shouldn't have a long bout of tripping over a lot of disk errors again any time too soon. I'll work on getting that underlying issue fixed up.

(I'm still planning a nice json interface, but I want to figure out my future pagination strategy and implement it in the interface first. Part of the purpose of a good interface like that is making it future-proof, and I'm not ready to commit to one yet.)

Sersys!CyQeb5KTDM 10343

>>10309
Thank you for your time, If there is anything I can do on my side please do tell.

10367

>>10341
Sorry about the late reply; I took the day off to generate a trip, review the old code, and do some scaffolding.

So, physical disk errors… that's some serious stuff. If the bill goes up, send me the estimate and I'll foot it.

In other news, I reviewed the logs from early Sunday morning and I see that one point the number of simultaneous connections reached somewhere around 165 - all for the intensive task of transferring images. When I designed the code, I thought that the processing time would be n/x for x concurrent connections (and so released the cap on connections), but I see that it's actually n-x, so I'm going to cut that down to around 1-5. That should take care of the 520 issue on this side.

I'm planning to moving the code to a small-scale production server this weekend, and then to it's final home next weekend after some more testing. I'd like to schedule some calibration tests around your schedule (If that's okay with you), so I'll drop you a line about that soon.

I'm glad you mentioned the JSON. The database I'm using - CouchDB - requires everything to be stored in JSON, so I'm already using a JSON representation of the site. I bring that up because I'm planning on writing an inline jQuery module for many of the features in 4ChanX and AppChanX, so we could probably synchronize our efforts on that and any other forthcoming features around a shared interface. I'll send that schema over in the e-mail; while it works for me, it's definitely open for modification. I also remember somebody else asking for a JSON interface for a RSS feed. We should probably get him in on the conversation too.

Lastly, I haven't seen a solution that I consider elegant on the chans I visit for long threads, so what I'm considering is loading in chunks in tandem with infinite scroll. I don't know many people who could read 400 posts in 10 seconds, so it might work.

Sage because we're now looking down the rabbit hole.

Anonymous 10440

File: 1377649619086.png (145.84 KB, 426x359, sad faic.png)

IT'S HAPPENING AGAIN

Macil!/5s/Techmk 10441

>>10440
A final complete solution to the sporadic short downtime is coming soon. It's just been a pain since the server host support staff has working hours that do not match my timezone's waking hours very well. There may be about an hour of downtime when the repairs happen, but then that should fix the issues for good.

10442

File: 1377785729006.jpg (41.9 KB, 400x378, and I owe it all to the pegasu…)

>>10441
>A final solution is coming soon.


Delete Post [ ]
[ home ] [ site / arch ] [ pony / oat / anon ] [ rp / art ]