Peering/routing in Spain. Performance 14x WORSE

I have a website on the free plan of Cloudflare.It’s low traffic website that generates less than 0.5 GB of traffic a month. However the performance has gone down the drain lately.

How come a page in the EU is being proxied to the USA if the costs are equivalent?


Even for the free plan this makes no sense.


This is kind of an issue. We are talking about a 14x decrease in performance.

Is this a bug? A temporary thing? A sketchy movement to upsell?

Is this forum just powered by community? No officials from CF here?
Just to know if my concern is ever going to be addressed here.

Traffic on this forum is oddly low, but you will need more info.

What hosting provider are you using for your site?
What residential ISPs are you using in Spain? Telefonica? Voda?
Do you have some ray-ids of the requests?
Are you connected to a VPN, explicitly (mullvad, purevpn, PIA etc…) or implicity via some crappy network AV that has inbuilt privacy ■■■■ (that essentially acts as a VPN - norton/sophos/avast/nod etc…)

If Madrid is down or re-routed, CDG or AMS would be nearest cloudflare POPs, and I can confirm via cloudflared tunnel (as of writing) this is where I’m routed to (MAD and CDG today) using Vodafone ES. I suspect this is a problem with your setup rather than CF issue.

1 Like

Hi matt,

Thanks for taking the time to reply. Much appreciated.
I have nothing fancy. Direct connection to Vodafone. But still routed to USA. Makes no sense, right?
Even if MAD colo is suffering heavy load the logical thing would be to route to another one in EU. But not to cross the Atlantic Ocean twice per request.

Here a RAY-ID of a request from seconds ago:

CF-RAY: 62065472ea830441-MIA

Just to be clear, I don’t work for CF, so the ray-id I won’t be able to use myself, but maybe someone from CF can track it.

To be clear the output of your above is from:

And your hosting company is in the EU… with EU IPs and geo registered to EU. And you’re in Spain.

So thinking this is odd, I tried a few domains under my control, they are hosted in London. I’m in Spain. And oddly as yourself I get YYZ (Canada on /cdn-cgi/trace), it’s definitely not what I expected. Another I get MIA too… Err, wow, thats odd

Interestingly the domain I have on paid Cloudflare Pro, as expected/optimal shows MAD.

1 Like

So it seems something generalised. What I can say is that this was not happening a few weeks ago. It’s a new thing.

Yikes this is scary. We have an ireland site (hosted in London) seeing it bounce around the world in a traceroute to anycast IPs dished out by Cloudflare. I’m upgrading to paid plan now to see if it consistent with what I observed earlier, Cloudflare status page is green though so this is extremely concerning

FYI this site does a good job of displaying which PoP you are likely to be routed to from each Cloudflare plan

I’m wondering if Cloudflare dynamically optimizes the anycast IPs handed out by geo source of traffic, realizing the imperfections and flaws in bgp/anycast. And thus a daily bot scan could bias the anycast IPs dished out.

This is slightly different issue. My closest colo is MAD, your site yield that as does cloudflared tunnel use, and Warp+ usage. All fine and sorted.

The problem is that some of anycast IPs dished out by DNS for cloudflare protected sites, are clearly hitting cloudflare colos in completely different parts of the world (none showing re-routed at the moment). The site I have behind a paid account, the anycast IPs dished out there are correctly hitting the local colocation as observed by cdn-cgi/trace and reflected in mtr/traceroute timings - Im unsure if this is correlated yet.

Holy yikes. The account I just upgraded to paid 15minutes ago has just changed to MAD on cdn-cgi. Wow these are big changes. 2 paid account routing correctly. 2 free still going MIA/YYZ, cloudflare status green, non of major EU data centers re-routed, or nearest to origins rerouted.

Tbh, I was about to write off this post as another user problem, but it looks like you’ve undercovered either a significant disruption/outage in routing or a commercial decision to deliberately sub-optimally route free plans around the world, 2 paid = optimal/okay, 2 free = around the world (where about 40+ more suited colo are available)

1 Like

I hope this is not a sketchy move to push people on to paid plans. There is absolutely zero excuse to route from EU to USA as the costs are more or less the same.

I hear you on that, there no need for this, unfortunately from a devops perspective I need to resolve this now, so we’ve just upgraded the two remaining sites (basically geo domains for an ecommerce site, all LDN hosted). I’ll let you know if 15 minutes if they start routing correctly.

1 Like

The two remaining cloudflare sites have just started routing correctly after upgrading to Pro plan - looking glass traceroutes from around the world looking as expected in terms of timing, cdn-cgi/trace showing MAD from my residential, other traceroutes from various VPS I have fine. One cdn-cgi/trace from a EC2 in AWS London is hitting AMS (another CDG) but that’s okay, I can handle an extra 15ms rtt and all within region - I guess thats part of HA… but not YYZ/MIA

Would love to know what’s going on?

1 Like

Maybe you could open a support request from your account? I guess support turnaround for paid accounts is better than the forum.

Done. I’ve asked support to give one of on-duty net-eng a headsup to this thread, since ops are probably in handover from Lisbon to states at this moment I wouldn’t expect an immediate reply - at least it’s not downtime…

1 Like

Due to the way routing in the country works some requests are sent to data center locations that are not necessarily the closest geographically. For example, if we do not have a peering relationship with your network provider or if that cache location is under heavier load, we may manually route our self-service plans to other datacenters.

Most of the requests are sent to the physically closest cache location, but some requests are sent to nearby ones that are under a lower capacity. If you would like your zone to always be served by the closest geographic location and the highest priority of traffic, please consider upgrading to the enterprise plan by filling out an Enterprise Request Form

With that said we still would like to rule out any routing issues.

Could you please share the plain text output of the following command in plain text?

mtr <>

Please make sure to replace <> with your actual website.

Please also test your website over at Where's it Up: API Make sure to select Madrid and Valencia and provide your result here.

Do you see the expected location if you test your connection with ?



Since upgrading my domains to Pro plan in the last hour, all the requests are hitting the nearest colo for me, correct/expected/optimal, so I will not be able to assist in replicating this right now., Warp+ (via Teams), etc… also hit MAD. Do note though if we take MAD and Spain out the picture for a moment, the anycast IPs dished out under the free plan I was also seeing traffic from Ireland hit MIA (like the OP) observed (seen via a curl to cdn-cgi/trace from IE based VPS) and looking glass traceroute timing matching that (~150ms vs 15ms).

You mention Enterprise specific priority, but could you also please comment more on the difference between free and Pro plan in this respect?

Btw, added Where’s it up to my toolbox, great tool, thanks


I believe that what you and OP are observing relates to Argo and the Cloudflare Global Private Backbone

Priority goes like Free < Pro < Business < Enterprise.

Glad the tool shares would be useful to you going forward.

1 Like