- Ontario, Canada
- https://www.danmcgrath.ca/
-
eh?
- Joined on
2008-05-08
It would appear that a Cloudflare feature called "workers", which hijacks requests to inject code into a request as part of our recent fund campaign was the cause of the content being cut…
I dug a little deeper and was noticing that the webserver was also missing a merged upstream config option to load mod_http2 and enable the protocol as well.
While the origin server now allows…
I had a chat with a Cloudflare tech, but they only suggested that it could be due to a missing Content-Length header. Unfortunately, we are sending the header from the backend server (origin).…
Last I heard, there was some extra ports enabled in a new rack at the DC awaiting for us to move things, which was planned to be sometime after BCON. Maybe it will help, maybe it won't.
I noticed an interesting case when trying to use github mirror for speedup.
The speed is slow(50-80KB/s) as long as I'm using git bash to clone the mirror. The only case I could get…
Thanks for the heads up. Although I think it's pretty much agreed that anywhere beyond the Netherlands is affected.
LazyDodo (in Canada) and I did another round of extensive tests, and found…
The weekend work at the data center went pretty good. It sounds like things at least helped.
More specifically, the work to deal with the MAC receive errors by swapping out the 10G SFP+…
oops, I realize I closed this earlier. I was wondering what happened when I misclicked that!
I dug a little bit more. I tried different sysctl's on the host, checked the netstat -m nmbcluster values, mtu's, but no matter what I just can't seem to get the window size to go past around 256k…
I did look more into the interface errors, but this time from the host pov. It would seem that, based on the total interface errors and uptime, we are receiving about 1.173 million interface…
No networking expert so feel free to ignore if this is obvious or wrong.
To me it seems a small TCP window size as found by @LazyDodo is exactly the kind of thing that would lead to a…
In terms of network statistics, I took another look at the switch that directly connects to the ISP. There are (still) tons of errors from the server where Phabricator/Git/Svn/Buildbot run, but…
Last weekend when I took a quick peek into this problem, it was found that git checkouts gave results that were fastest in the same rack/network as the VM of gitea.
Performance at home was…