undefined | Better HN

0 pointsgerdesj8mo ago0 comments

ping and mtr only test one thing. I have saws, drills, routers (lol), planes, screwdrivers, hammers and more in my workshop. To be fair a drill driver and a hammer get a lot of jobs done! However, I will get the impact driver out or a fret saw with a very fine blade as the job requires.

The article here is about a loss of DNS service and proves it with ping. That is wrong and you know it. Diagnosing the fault should involve ping but that is not how you conclusively prove DNS is not working.

To be honest you cannot conclusively prove anything in this game but you can at least explore the problem space effectively from your perspective with whatever you have access to. I happen to have a RIPE ATLAS probe at work with a gigantic amount of credit, so I could probably get that system to test Cluodflare DNS from a lot of locations.

If you present to a doctor with some mild but alarming chest pains, I'd hope they wouldn't just look at you and prescribe a course of leeches. A stethoscope is a good start (ICMP) but an ECG (dig) is better. That analogy might need some work 8)

0 comments

indigodaddy8mo ago

If you have a demonstrated network/connectivity problem to an endpoint that provides DNS, then DNS is down (or at the very least degraded) for you. If a functionality of layer 3 is not working, should we expect layer 4 to work, and keep looking into aspects of layer 4 and/or layer 7, or would it make more sense to keep troubleshooting the layer 3 issue?? Any entry level NOC Technician would know at this point that doing digs/queries to the endpoint would not necessarily be meaningful when we have an underlying connectivity/network problem that is the likeliest main contributor to the issue.

gerdesjOP8mo ago

"Any entry level NOC Technician would know at this point"

I'm just a consultant who's been mucking about with networks for 30+ years. I'm sure your highly paid technicians will teach granddad a thing or two.

I note you switch between the OSI seven layer model and the ARPA four layer one with gay abandon. What are you doing at layers five and six?

We are all engineers here (whether chartered or not). The big question is - "Is the service up"? The service is DNS.

We go to the toolbox as any engineer does and use a tool for the job. I can hammer a screw into a wall or use a screwdriver - both will work but one will work effectively. I'll use dig but I imagine that a Windows jockey will use nslookup - both will work.

dig/nslookup fail? OK, now we look at connectivity issues - that's when ping comes in. However we do not own the DNS service and we cannot know that it is now dropping pings for some reason. Then we might play games with packet generators and Wireshark to try and determine what is going on. However, we do not run that failing service and all we can conclusively ... conclude is that for us, it is not working.

That's a far cry from Cloudflare DNS is down for everyone. We can only conclude that Cloudflare DNS is not working for me.

indigodaddy8mo ago

You seem to be not addressing my main point, which is, once we are confident we have a network/connectivity issue, what is the benefit of now focusing on the outcomes of DNS queries? How does that help us at this point, when we know that DNS is not working for us in large part due to not being able to reliably connect to the endpoint itself?

In regard to an endpoint out of our control, once we demonstrate we cannot connect to it or serious connectivity problems in general, "is the service (that the endpoint provides) up?" is not a question that we need or should be trying to answer at that point.

That's cool though, if you want, you can just keep doing digs to an endpoint that is degraded from a network perspective, while I keep trying to troubleshoot why we have packet loss to the endpoint..

vel0city8mo ago

Plenty of hosts may respond to DNS while filtering ICMP. Showing a ping failure as an example of some authoritative layer 3 failure shows a misunderstanding of what ping is doing.

indigodaddy8mo ago

Sure, but here we are talking about an endpoint that we know should/previously responded to ICMP, and then are subsequently having a problem with it. So if we are now having a problem with the service provided by the endpoint, AND we see not insignificant packet loss on MTR/ping (or intermittent TTL exceeded which points to route issues), then we can be pretty certain we have a connectivity/network/route problem. Which is a problem at layer 3. My point in this whole thing is that once we know that, it makes no sense to say, oh let's shift to or we really should be "troubleshooting the service/application that the endpoint is providing" whether that be https or DNS or whatever. No, we keep troubleshooting the network/connectivity issues if/once we are confident that the problem lies therein.

1 more reply

esseph8mo ago

Ping and MTR are actually several different tools spread across them.

Connectivity over ICMP / UDP / TCP, DNS resolution, Autonomous System path, MPLS circuit, IPv4 / IPv6 routing, circuit to endpoint latency, per hop firewall configuration, device packet security configuration, jitter, MTU, and probably some other things I'm forgetting.

A carpenter knows their tools.

gerdesjOP8mo ago

"A carpenter knows their tools."

Quite, and they also know when to use them effectively.

I have no idea what "Autonomous System path" is but it looks like someone searching terms. An Autonomous System is a BGP thing.

You say "I'm forgetting" and I say - you don't have much skin in this game.

I've spent roughly 30 years getting to grips with this stuff.

esseph8mo ago

I have helped designed some of the network hardware and software you may have used, I'm not sure how that's relevant. Pointless D measuring.

My point stands, which is: There are a lot of capabilities in these tools that should not be overlooked or dismissed.

In addition, reachability of the service is one of the things you would note with said tools as you work your way through the stack. You can even use MTR to see if the DNS server is holding port 53 open.

j / k navigate · click thread line to collapse

0 comments

indigodaddy8mo ago

gerdesjOP8mo ago

"Any entry level NOC Technician would know at this point"

I'm just a consultant who's been mucking about with networks for 30+ years. I'm sure your highly paid technicians will teach granddad a thing or two.

I note you switch between the OSI seven layer model and the ARPA four layer one with gay abandon. What are you doing at layers five and six?

We are all engineers here (whether chartered or not). The big question is - "Is the service up"? The service is DNS.

That's a far cry from Cloudflare DNS is down for everyone. We can only conclude that Cloudflare DNS is not working for me.

indigodaddy8mo ago

That's cool though, if you want, you can just keep doing digs to an endpoint that is degraded from a network perspective, while I keep trying to troubleshoot why we have packet loss to the endpoint..

vel0city8mo ago

Plenty of hosts may respond to DNS while filtering ICMP. Showing a ping failure as an example of some authoritative layer 3 failure shows a misunderstanding of what ping is doing.

indigodaddy8mo ago

1 more reply

esseph8mo ago

Ping and MTR are actually several different tools spread across them.

A carpenter knows their tools.

gerdesjOP8mo ago

"A carpenter knows their tools."

Quite, and they also know when to use them effectively.

I have no idea what "Autonomous System path" is but it looks like someone searching terms. An Autonomous System is a BGP thing.

You say "I'm forgetting" and I say - you don't have much skin in this game.

I've spent roughly 30 years getting to grips with this stuff.

esseph8mo ago

I have helped designed some of the network hardware and software you may have used, I'm not sure how that's relevant. Pointless D measuring.

My point stands, which is: There are a lot of capabilities in these tools that should not be overlooked or dismissed.

j / k navigate · click thread line to collapse