Anyone using fly.io from India / Sri Lanka / Pakistan regions - beware of poor routing

Hi everyone, John from the Fly.io infra team here. I’m afraid I can’t give you a very satisfying answer right now, but I just wanted to confirm that, yes, we are aware of the networking situation in India, and, yes, we are investigating how we can improve the situation; however, as sanvit said, these kind of problems are complex, take time to solve, and to a certain degree are beyond our control because they in part result from decisions made by network providers totally independent from us and with whom we lack any relationship at all. Nevertheless, we are working on mitigation strategies, and one major one involves rolling out new hardware to our points of presence across Asia. But commissioning physical servers and getting them in position is, unfortunately, not nearly as convenient as launching a VM. But even when the new servers are in place, this issue will not be “fixed” just, hopefully, “better”, because networking is a constant Red Queen’s race. We will continue to investigate how to improve the situation as networking routes inside and outside India change.

We do, however, very much appreciate everyone taking time to comment here because it helps us gain a better perspective on the problem. Thank you all for letting us know what you’re facing and please be sure that we’re hearing you, even if we can’t snap our fingers and fix it.

3 Likes

This may be far-fetched, but if there’s anything I can do to find someone somewhere in the indian ISPs and hound them to fix some config, please send me an email.

Another issue is that major ISPs in India prefer to route traffic out of the country.

@AkshitGarg that’s crazy! reading that makes my hopes diminish :sob:

Certainly if anyone has any inside connections with an Indian ISP and would be willing to make introductions, we’d be extremely grateful. Leave a message here and we’ll get in touch.

Not sure if something changed in the past 6-8 hours or so, but traffic for me has been consistently going through the ‘sin’ edge (I’m on Bharti Airtel network currently)

curl -I https://<my_apps_hostname> -H "flyio-debug: doit"
HTTP/2 403
access-control-allow-headers: *
date: Wed, 21 Feb 2024 16:32:59 GMT
access-control-allow-origin: *
access-control-allow-methods: GET, POST, PUT, DELETE, OPTIONS
content-length: 16
x-frame-options: SAMEORIGIN
content-type: text/plain; charset=utf-8
x-xss-protection: 1; mode=block
x-content-type-options: nosniff
server: Fly/17d0263d (2024-02-15)
via: 2 fly.io
flyio-debug: {"n":"edge-nac-sin1-e297","nr":"sin","ra":"106.222.202.155","rf":"Verbatim","sr":"sin","sdc":"sin1","sid":"148ed5d6b2d008","st":0,"nrtt":0,"bn":"worker-cf-sin1-d8b0"}
fly-request-id: 01HQ686S1F74XPN5RPRP9DFM12-sin

I just ran a small test using the geonode proxy network again, results are definitely improved. Reliance Jio ISP still does not route correctly, but Airtel seems to be getting it right now.

I’m on Airtel and still connecting to FRA, but it seems there has been some change. It’s a little too soon to tell, but FRA has the least amount of traffic I’ve seen in a while.

Past 6 hours:
image

image

@john-fly / @jssjr - can you tell who the data center provider for bom region is (I read elsewhere you use Equinix), and who is the ISP providing the internet connectivity. I am reaching out to some folks and may be able to make some headway in getting someone relevant for a connect.

Hey! Its me again with a big update. We brought up two new edges in the BOM region with a different network configuration. Seems like the routing is much better now. I am consistently being routed to BOM from DEL.

2 Likes

wait… what part of the network configuration can we control here. can you please elaborate ?

@hi.kanily you don’t have to change any configuration. My above reply meant fly.io edges (my badge doesn’t reflect my staff status yet haha).

Yes, I can confirm that this is true on Airtel also.

flyio-debug: {"n":"edge-nac-bom1-3119","nr":"bom","ra":"27.56.208.29","rf":"Verbatim","sr":"sin","sdc":"sin1","sid":"3d8d979ef5ee89","st":0,"nrtt":64,"bn":"worker-cf-sin1-d8b0"}
fly-request-id: 01HQ7ZW1Y1CWN00V1CWWMN6W5T-bom

Pings are now 60ms!

Last 6 hours:
image

Just brought up a node in BOM, will update tomorrow.

flyio-debug: {"n":"edge-nac-bom1-3119","nr":"bom","ra":"27.56.208.29","rf":"Verbatim","sr":"bom","sdc":"bom1","sid":"3d8d9ee6f00268","st":0,"nrtt":0,"bn":"worker-cf-bom1-9cee"}

Fly team, whatever you’re doing, seems to be working. On a small sample set, now ingress is happening via bom across all ISPs I just hit

This is great! I hope this continues to work.

1 Like

We use two providers in India: NetActuate (ASN 36236) and Cachefly (ASN 30081). We like Cachefly more overall, but the interaction of Cachefly’s peering arrangements in India and our particular configuration is what’s been causing the trouble, and the recent fix we just put out in the past 12 hours involved leaning more on NetActuate. So if you have a connection that could help us improve our peering connections through Cachefly, we would be VERY thankful.

Also, just to be clear, @AkshitGarg is also a member of the Fly.io Infra team. And, unlike me, he has the advantage of actually living in India. :slight_smile:

2 Likes

Also, general response to anyone who makes it down this far: The title of this post says “Anyone from India / Sri Lanka / Pakistan” but all the feedback we’ve gotten so far is people from India. If you are from somewhere else in South Asia, please let us know here how your networking is! It’d be helpful to know if this is a problem across South Asia, or if it somehow stops at the borders of the Republic of India.

1 Like

I had mentioned Sri Lanka and Pakistan because in my proxy testing I saw poor routing from those two countries as well. I just ran it again for SL and PK

Sri Lanka results (anecdotally, these are a little better than the last I remember, but still not fully optimal):

Pakistan results (these are better than before for sure, also better comared to SL, no routing to cdg, fra etc):

1 Like

Just did an Atlas test run in South Korea, and for some reason, SK Telecom is getting 70~80ms ping while it should normally get around 30 ms ping to Tokyo. The ping time looks more like HK to me, but not quite sure. Could you take a look at it (SK Telecom is the biggest cellular provider in South Korea)?

Also, is there any South Korean region in your plans by chance?

image

image

This is much, much better.

It might be that Indian users have less pageviews per visit or something, but I would have expected the traffic thru BOM to be higher.

The regional traffic (month to date visitors is) roughly:
India 8.8k
US+ CA 8.7k
Europe + Russia 5k.

So I would expect BOM traffic to be the same as FRA. Right now, BOM is 11kB and FRA is 15kB (not including the rest of Europe edges).

This is all extremely helpful, thank you so much for taking the time to run these tests.

Funny you should ask…

(This is not a commitment right here that we will be opening in Korea, specifically, but just remarking on the coincidence that you should bring this up while we’re having these conversations internally.)

1 Like