Guild Wars 2

We found some issues with /v2/characters that necessitated temporarily disabling the /v2/characters endpoints. I’m not sure when we’ll have it back up, but it likely won’t be before the next release.

Sorry

By “next release” I mean “the next time we do a patch that isn’t a hotfix”.

Sorry about the ambiguity.

AZY.8352:

Am I a “v2” character? I haven’t been able to access my API all day…

“v2” refers to the API version — not the character version. The only way to access character data is via the “/v2/characters” API endpoint, which is disabled.

Effectively, all API access to all character data is turned off until I can resolve the issue.

TPMN.1483:

Did this get fixed with the latest patch?
Are we waiting for normal patch Tue for the fix if above is No?

No, it’s not yet fixed.

We tracked down the issue today and the current plan is to get it hotfixed tomorrow evening, but can’t make any promises. Sorry again about the downtime, software is hard :<

Fix went out with today’s hotfix and works correctly; just turned the /v2/characters endpoint back on.

Nekres.1038:

Uhm, does the data update somewhat slower? My deaths went from 5 minutes intervall updates to almost 20 minutes!

Should still be ~5 minutes. There’s an edge case where the map instance server hasn’t persisted character data to the database (IIRC that’s on a 5 minute timer, or when you switch maps or sell something on the trading post etc) and that stale data gets pulled in and cached by the API backend for another 5 minutes. So to the best of my knowledge, the maximum staleness should be around ~10 minutes (though only under special circumstances).

It’s altogether possible there’s a bug somewhere causing data to be cached too long. If the issue doesn’t clear up, consider opening an issue on github (or just posting again here).

bobsort.4097:

It seems last “/” causes a redirect which removes the key from request.

That’s a bug; here’s a tracking issue. It’s probably all the authenticated endpoints, not just /v2/characters. I can probably have that fixed in a week or two, in the meanwhile I recommend not sending the trailing slash in requests.

stress level zero.4907:

Has the false negative bug been fixed where the api would return 40x responses for valid tokens on the characters endpoint?

I found and fixed a bug awhile back that would cause issues like that for the first request once/day or so (when the session created for the token would expire). I haven’t personally seen it happening anymore, but let me know if you’re still seeing it and I’ll investigate some more.

stress level zero.4907:

Lawton Campbell.8517:

I found and fixed a bug awhile back that would cause issues like that for the first request once/day or so (when the session created for the token would expire). I haven’t personally seen it happening anymore, but let me know if you’re still seeing it and I’ll investigate some more.

Hey Lawton, this happened today.

Received an unexpected 400 response:

http://i.imgur.com/umE1OXw.png

Thoughts?

https://github.com/madou/armory-back/blob/master/gw2-fetch/src/controllers/ping-controller.js#L95

I guess I’ll disable deleting the token for now.

Actually, I think we might have done a server deploy around that time — one of the backend servers the API talks to was timing out over two windows last night. Do you happen to have the error text returned by the server? There’s a lot of reasons a key can be invalid (and they all return unique error strings).

The non-transient errors are supposed to be “invalid key” (key doesn’t exist), “banned key” (we’ve blacklisted the key on our side) or “broken key” (something horrible has happened).

Looking at the code, there may be an issue where “invalid key” is incorrectly returned during a temporary failure — need to root around in some code to check what situations can cause that.

In any case, this is definitely not documented well and we really should be returning 5xx responses for transient failures. I’ve created a tracking issue for both the audit and better docs; I would keep seemingly broken API keys around until this gets resolved. Sorry about the hassle