[-] proxymint@programming.dev 2 points 2 days ago

thanks, good catch. it got stuck when js was off, we fixed it today, try a refresh if you want

[-] proxymint@programming.dev 0 points 2 days ago* (last edited 2 days ago)

Hmm, interesting, I thought I was human. Anyway, as far as I can tell this is within the rules here?

-2

Everybody runs the IP check, sees the exit address and calls it done. Fair enough, but a site can ask a lot more than that, and these are the ones that bite:

IPv6. Your device has a public v6 address, your proxy only carries v4, and some traffic just walks around it. Turn IPv6 off in the browser profile or on the adapter.

DNS. Resolver in your own country while the exit is in another one? That's a leak. On SOCKS5, use socks5h:// so names resolve at the proxy.

WebRTC can hand out another public address. Via, Forwarded and X-Forwarded-For can carry your real IP. And your browser clock saying one time zone while the IP says another is a flag all by itself.

Also, easy to forget: a system proxy only covers apps that bother to read it, and an extension only covers its own browser.

We put all seven checks on one page. Your browser asks a few public services the questions directly and we store nothing: https://proxymint.com/tools/proxy-check/

Disclosure: it's our tool, we run a proxy service.

3

Per the Apache advisory, it's improper synchronization in assign_headers() inside mod_http2, and it can end in RCE. Range is httpd 2.4.55 to 2.4.70, plus standalone mod_h2 2.0.8 to 2.0.42. Wiz said no public exploit when they wrote it up on the 11th. Stuck on an old build for now? Protocols http/1.1 takes HTTP/2 off the table. Worth knowing: Debian's packaged http2.conf has Protocols h2 h2c http/1.1, so a2enmod http2 enables both. Also, Cloudflare's HTTP/2 to origin is on by default, so a CDN in front isn't automatically a shield. Commands and distro notes at the link. Disclosure: it's our blog, we run a proxy service.

[-] proxymint@programming.dev 2 points 4 days ago

Fair. We run proxies, so "proxy or site?" is the question we care about, and I didn't trust the error messages to answer it.

The point is what you do after. If the proxy hung you rotate it out. If the site hung you slow down on that site. requests blames the site by name either way, so you can end up throttling a site that's fine and keeping a dead proxy. That's why.

-6

If a scraper starts throwing ReadTimeout and the first instinct is to double every timeout, read the message again. It can point at the wrong place.

We made each phase of a proxied HTTPS request stall on purpose in a local lab (curl 8.5.0, requests 2.34.2, httpx 0.28.1). For a proxy that accepts the connection and never answers CONNECT, curl said "Proxy CONNECT aborted due to timeout". requests raised ReadTimeout with the target's hostname in it, though the target was never contacted. httpx said ReadTimeout too, but only after its read timer, so 8 seconds instead of 5.

The tell in requests: if the number in "(read timeout=N)" equals your connect value, the stall was in the tunnel or TLS, not on the site.

Two more from the same run. An unreachable proxy raises ProxyError, which isn't a subclass of requests.exceptions.Timeout, so code that only catches Timeout lets it crash the worker. And with no timeout set at all, curl and requests both sat there for 134 seconds, which is the Linux kernel's SYN retries and not the client.

curl -w with time_connect, http_connect, time_appconnect and time_starttransfer splits the four phases, and honestly that's the first thing I'd run.

Full table: https://proxymint.com/proxy-timeout/

Disclosure: our blog, we run a proxy service.

[-] proxymint@programming.dev 1 points 1 week ago

it does when the other side allowlists a hostname. plenty of firewalls and saas allowlists only take an ip or a cidr though, and they don't re-resolve, so the entry breaks the next time your isp renumbers you. it also only helps if the traffic actually leaves from that same connection.

3

A partner wants to allowlist your IP and someone says "then we need a static IP". Maybe, but direction matters, and a lot of teams pay for the wrong thing here.

Inbound services need a static address on their own line. Outbound traffic only needs a fixed exit, which can be a VPN gateway or a cloud server, so check what you already run before you buy anything.

And "DHCP Enabled" on your machine only describes the LAN side. It says nothing about your public address. For that you need about a week of logging and a look at the router's WAN address, which is also how you find out you're behind CGNAT.

Dynamic addresses change on three clocks: the lease, ISP renumbering (often every 24 hours where they do it), and outages long enough for the lease to lapse. So a network that held for two days hasn't proved much.

[-] proxymint@programming.dev 3 points 1 week ago

nice. the ones people tend to forget are alpine-based containers that run sshd inside. on alpine ssh -V won't settle it, since the fix is a backport, so check the package with apk info openssh instead.

3

If you ticked a "block AI training" option on Cloudflare at some point, this is worth five minutes. Cloudflare's docs (updated Nov 2) say training actions also apply to mixed-purpose crawlers used for both search and training. Their July blog post names Googlebot, Applebot and BingBot as examples.

There were reports in August of Googlebot getting 403s on sitemaps. Nobody has confirmed the cause as far as I can find, so don't panic over it. It could be another setting entirely, Bot Fight Mode came up in the same thread.

Still, checking is cheap. Look at crawl stats in Search Console for a jump in 403s since mid-September, then read which Cloudflare setting is on. I'd rather spend five minutes on that than learn about it from a traffic graph.

Has anyone here actually seen this on their own site?

4

Got ERR_TUNNEL_CONNECTION_FAILED in Chrome and no idea why? Try the same proxy from Python, because it'll tell you what the browser won't. Chromium drops every CONNECT reply except 200 and 407, so you can't tell a policy block from a dead exit in there.

Python leaves the status in the exception text. You'll see OSError: Tunnel connection failed: 502 Bad Gateway from http.client, and requests wraps that in a ProxyError with the same words inside.

I'd read that string before retrying anything, since a blanket except ProxyError plus a retry loop just burns time on errors that can't heal. A 403 is usually policy, sometimes just a port they don't allow. 407 means check your login. If you get 405 or 501 you aren't talking to a forward proxy at all. Only 502, 503 and 504 are worth backing off on, those happened behind the proxy.

Disclosure: we run a proxy service, none of this depends on whose.

46

From the Alpine advisory and the researcher's notes: musl's crypt() gives back * when a password is over 256 bytes, and sshd then compares it with an account whose stored field is also *. Key-only users have that field, and so does the alpine user on tiny-cloud images. Fix is OpenSSH 10.6p2 or Alpine's backport (on 3.24 that's 10.3_p1-r2), plus an sshd restart. PasswordAuthentication no also closes it. Package table and check commands are in the link. Disclosure: it's our blog, we run a proxy service.

16

Two traps in java.net.http.HttpClient that bite when you route it through a proxy, because neither one looks like a proxy problem at first.

Basic proxy auth is disabled for HTTPS tunnels by default, so your username and password can be perfectly right and the tunnel still fails on auth.

The nastier one: point it at a SOCKS proxy and it gets ignored, so requests leave from the machine's own address. No exception, no warning. Everything looks fine until you check the exit IP, and by then the job may have run for a week from your server.

So whatever client you use, I'd log the exit IP once at startup and fail the run if it matches the host's address. It's a few lines and it catches this whole family of mistakes.

Disclosure: we run a proxy service, this came out of comparing proxy support across languages.

1

"No logs" on a proxy or VPN page tells you very little, and a couple of old court cases show why. In 2011 a UK web proxy and VPN service handed over a user's records under a court order. In 2017 a US cyberstalking complaint relied on records from a VPN whose privacy policy said it kept no logs.

A service that bills per GB and handles abuse reports has to keep something, so I'd stop asking whether a provider logs. Ask which fields, for what purpose, how long, and who can demand them. A policy that names its fields is worth more than a bold claim on the landing page.

The other half is what they can see at all. Over a CONNECT tunnel with certificate checks on, a proxy sees your IP, the hostname, timing and volume. Paths, headers, passwords and page content stay inside TLS. Turn off certificate verification and that stops being true, which is why I'd never leave verify=False in a script.

Disclosure: we run a proxy service, so we keep some of those records too.

[-] proxymint@programming.dev 7 points 1 week ago

git log -S'some_string' -p is the one that pays for itself. it finds the commits where that string was added or removed, so you can track down when a function disappeared without reading through history. -G is the same idea with a regex, matching any changed line.

and git log --follow -p path/to/file keeps going across renames, which plain git log file won't do.

8

Easy one to miss if you run a lot of pages in one automated Chrome behind a proxy: Chrome allows 32 simultaneous connections through an HTTP/1.1 proxy. Past that, requests wait inside the browser before the proxy ever sees them.

So throughput sits flat no matter how many tabs you add, and it's tempting to blame the proxy and go looking for a bigger plan. That won't help. Splitting the work over several browsers will.

Two more Chrome oddities in the same area. A user:pass@ URL in --proxy-server fails with net::ERR_NO_SUPPORTED_PROXIES, because Chrome doesn't use credentials embedded in proxy settings. And a password-protected SOCKS5 server doesn't work at all, since Chrome supports no SOCKS5 auth methods, so you get net::ERR_SOCKS_CONNECTION_FAILED whatever your code does.

7

Easy one to miss if you're checking a Shopify store for other markets: a VPN or proxy won't show you different shipping rates. Shopify ties rates to the shipping zone of the checkout address, so a local IP changes nothing there. A test checkout with a local address does.

Prices, currency, catalog and translations you can check for free in the admin: Markets, click the market, then More actions > View online store. Products > View as shows the market's currency and which products aren't published for it.

What the preview can't show is the IP-driven stuff: the automatic redirect, geo app popups, the cookie banner, apps that block VPNs. For that you need a connection in the country and a clean browser profile, and you have to load the root domain, because Shopify only redirects on the first page a visitor opens. Land on /de/ directly and you skip the very redirect you wanted to test.

The cookie banner is the sneaky one. With automated privacy settings on, Shopify sets it up for UK and EEA visitors when you have active markets there, so from a US desk you just never see it.

How are you handling multi-market QA for stores like this, manual passes or something scripted?

14

If you've got anything running on Google's Custom Search JSON API, it's on a clock. Google closed it to new customers and their overview page says it's gone on January 1, 2027.

Easy to miss in the docs: one call returns 10 results and a query tops out at 100. So pulling the top 100 for a single keyword is 10 billable calls.

For checking your own site's rankings there's an easier path anyway. The Search Console API hands you query, page, country and device rows, up to 25,000 per request, and it's all within Google's rules.

Anyone already moved off Custom Search? What did you switch to?

view more: next ›

proxymint

0 post score
0 comment score
joined 3 weeks ago