Closed schneefux closed 7 years ago
Response times seem stable on my end. http://bit.ly/gamelocker-status
Not sure if related- I noticed some matches I played yesterday took a few minutes to show up (checked on a couple different apps) I checked the status page at that time and nothing seemed immediately out of sorts.
@DominicGunn Can you shortly elaborate on what kind of requests you send to measure response times — do they reflect natural API usage? If you query /matches, that might be cached, and our queries might be not.
Our requests hit all date shards with varying filters: always player ID, always sorting createdAt ascending and most of the time the modes casual,ranked.
Mine is coming back under 200-300ms.
Hey folks - This was two weeks after we moved over to our new container platform, so its possible something is weird there. I dialed up resources a bit this afternoon, could you tell me if you're seeing an improvement.
With that said, my graphs also show stable (and actually some improvement?)
@schneefux yours could be network related on your end too right? Hard to tell... :( Where do you host? Is it AWS? What region?
Can only be related to #248. 40 000 requests (accumulated over the past weeks) 60 seconds 2 retries will have a big impact.
We're hosted in the same region as your CDN I think.
I changed the default download start date to 2017-05-01. It's as fast as never before! Not a single request over the past 24h that failed after the first retry, I'm amazed. I dropped the failed requests that were queued up.
Not a single server error for the past two months. nice
As I have written in https://github.com/madglory/gamelocker-vainglory/issues/252#issuecomment-323600491, we are seeing a 2x increase in response time since about Aug 11th, measured as the time between opening the connection and closing it. Here is a graph that shows the averages…
It is possible that this is a problem on our end, or a problem with how we collect the data. @cklugewicz could not confirm it, his data shows stable response times (although measured as time to first byte). Maybe it is related to #248, as timeouts are included in the data that generate these graphs.
Can you confirm / deny that the API became slower? We are mainly hitting the EU, SG and NA shards, and there is no significant shift in our user base compared to the past months.