Skip to content
Proxy fundamentals

Proxy error: what each message means and how to fix it

Exact proxy error messages from curl, Python, Node, Chrome, Squid and Apache, reproduced on a test rig, with the cause and fix for each.

On this page 14 sections
  1. Find your exact message
  2. How these messages were captured
  3. Find the failing hop with one command
  4. Cause 1: the proxy rejected your credentials (407)
  5. Cause 2: your client cannot reach the proxy
  6. Cause 3: the right address, the wrong protocol
  7. Cause 4: the proxy refused the tunnel by policy (403)
  8. Cause 5: the proxy could not reach the destination (502, 503, 504)
  9. When a website shows “Proxy Error”
  10. Cause 6: TLS interception
  11. Cause 7: the proxy hung up or went silent
  12. Cloudflare 10xx errors behind a proxy
  13. Cause to fix, in one table
  14. Turning exceptions into a cause in Python

A proxy error means the request died at a machine in the middle, not at the website you asked for. Two hops can break: the one from your client to the proxy, and the one from the proxy to the next server. Every message on this page points at one of those two hops, and once you know which, the fix is usually a single line.

If you saw a web page like this one, you met a reverse proxy sitting in front of a website:

Proxy Error

The proxy server received an invalid response from an upstream server.
The proxy server could not handle the request

Reason: Error reading from remote server

That is Apache httpd’s built-in error page. We produced it, with the status line HTTP/1.1 502 Proxy Error, by putting Apache 2.4.67 in front of a backend that hangs up without answering. A visitor cannot fix it from their side. The application behind the site has crashed, stalled or gone away, and the reverse proxy section lists what the owner should look at.

If you write code that sends traffic through a forward proxy, your message reads more like one of these:

curl: (7) CONNECT tunnel failed, response 407
curl: (5) Could not resolve proxy: gw.invalid
requests.exceptions.ProxyError: ... (Caused by ProxyError('Unable to connect to proxy', OSError('Tunnel connection failed: 407 Proxy Authentication Required')))
httpx.ProxyError: 503 Service Unavailable
net::ERR_PROXY_CONNECTION_FAILED
net::ERR_TUNNEL_CONNECTION_FAILED

Find your exact message

Each row is one failure we caused on purpose against servers running on our own machine. The columns show what five clients printed for it. Find your string, then jump to the cause. A dash means we did not run that combination.

Failure curl 8.23.0 Python requests 2.34.2 httpx 0.28.1 Node 24.2.0 fetch Chrome 155
Nothing listening on the proxy port (7) Failed to connect to … over proxy ProxyError('Unable to connect to proxy', NewConnectionError(… [Errno 61] Connection refused ConnectError: [Errno 61] Connection refused fetch failed, cause ECONNREFUSED ERR_PROXY_CONNECTION_FAILED
Proxy hostname does not resolve (5) Could not resolve proxy: gw.invalid NameResolutionError(… Failed to resolve 'gw.invalid' ConnectError: [Errno 8] nodename nor servname provided, or not known fetch failed, cause ENOTFOUND ERR_PROXY_CONNECTION_FAILED
Missing or wrong proxy password (7) CONNECT tunnel failed, response 407 Tunnel connection failed: 407 Proxy Authentication Required ProxyError: 407 Proxy Authentication Required fetch failed, cause Request was cancelled. ERR_INVALID_AUTH_CREDENTIALS
Password contains @ or #, not encoded (5) Could not resolve proxy: ss Failed to resolve 'ss' ConnectError: [Errno 8] … Process crashes: InvalidArgumentError: invalid url —
https:// used for a plain HTTP proxy (35) TLS connect error: … wrong version number Your proxy appears to only use HTTP and not HTTPS ConnectError: [SSL: WRONG_VERSION_NUMBER] cause ERR_SSL_WRONG_VERSION_NUMBER ERR_PROXY_CONNECTION_FAILED
Address is a web server, not a proxy (7) CONNECT tunnel failed, response 501 Tunnel connection failed: 501 Unsupported method ('CONNECT') ProxyError: 501 Unsupported method ('CONNECT') cause Request was cancelled. ERR_TUNNEL_CONNECTION_FAILED
Proxy forbids the destination port (7) CONNECT tunnel failed, response 403 Tunnel connection failed: 403 Forbidden ProxyError: 403 Forbidden cause Request was cancelled. ERR_TUNNEL_CONNECTION_FAILED
Proxy cannot reach the destination (7) CONNECT tunnel failed, response 503 Tunnel connection failed: 503 Service Unavailable ProxyError: 503 Service Unavailable cause Request was cancelled. ERR_TUNNEL_CONNECTION_FAILED
Proxy closes the connection without a reply (56) Proxy CONNECT aborted RemoteDisconnected('Remote end closed connection without response') RemoteProtocolError: Server disconnected without sending a response. No error until our own timeout fired ERR_EMPTY_RESPONSE
Proxy intercepts TLS with its own CA (60) … self-signed certificate in certificate chain (19) SSLError(… CERTIFICATE_VERIFY_FAILED ConnectError: [SSL: CERTIFICATE_VERIFY_FAILED] cause SELF_SIGNED_CERT_IN_CHAIN ERR_CERT_AUTHORITY_INVALID

Two things stand out. Chrome folds a 403, a 501 and a 503 from the proxy into the same ERR_TUNNEL_CONNECTION_FAILED, and Node’s built-in fetch reported all three, plus the 407, as Request was cancelled. Neither tells you the status the proxy actually sent. curl and both Python libraries do. When a browser or a Node service fails through a proxy, rerun the same request with curl before changing anything else.

How these messages were captured

Every string on this page was produced on 21 October 2026 on macOS (Darwin 25.6.0), each case run once, against processes we started ourselves on 127.0.0.1 and nothing else:

  • Squid 6.13 with three ports: one requiring Basic proxy authentication, one open, and one doing TLS interception with a lab CA (ssl_bump).
  • Apache httpd 2.4.67 as a reverse proxy in front of the misbehaving upstreams below.
  • A Python origin serving HTTP and HTTPS, signed by a separate “Lab Origin CA”.
  • Two hostile upstreams: one that accepts and immediately closes, one that accepts and never answers.

Clients: curl 8.23.0 (Homebrew, OpenSSL 4.0.3) plus the curl 8.7.1 that ships with macOS, Python 3.12.13 with requests 2.34.2, urllib3 2.8.0 and httpx 0.28.1, Node v24.2.0 with NODE_USE_ENV_PROXY=1, and Chrome 155.0.8059.85 headless, driven through the DevTools protocol so we could read the exact net:: error. Hostnames ending in .invalid are reserved for exactly this purpose by RFC 6761 section 6.4, which makes them a safe way to force a DNS failure.

Error numbers inside messages depend on the operating system. macOS prints (61) Connection refused; a Linux build prints (111) for the same refusal, because that is the value of ECONNREFUSED in the Linux errno header. Search for the words, not the number.

Find the failing hop with one command

curl can report the proxy’s answer to CONNECT separately from the target’s answer. That single split settles most arguments about whose fault it is. Save this as proxycheck.sh:

#!/usr/bin/env bash
# proxycheck.sh PROXY_URL [TARGET_URL] [extra curl args]
proxy=$1; target=${2:-https://example.com/}
curl -s -o /dev/null --max-time 15 -x "$proxy" "${@:3}" \
  -w 'connect_to_proxy=%{time_connect}s  proxy_CONNECT=%{http_connect}  target_status=%{http_code}  curl_exit=%{exitcode}  msg="%{errormsg}"\n' \
  "$target"

Output from our rig, one line per failure:

$ ./proxycheck.sh http://127.0.0.1:46999 https://localhost:46443/
connect_to_proxy=0.000000s  proxy_CONNECT=000  target_status=000  curl_exit=7  msg="Failed to connect to localhost:46443 over proxy 127.0.0.1 after 0 ms: Could not connect to server"

$ ./proxycheck.sh http://127.0.0.1:46128 https://localhost:46443/
connect_to_proxy=0.000183s  proxy_CONNECT=407  target_status=000  curl_exit=7  msg="CONNECT tunnel failed, response 407"

$ ./proxycheck.sh http://127.0.0.1:46130 https://localhost:443/
connect_to_proxy=0.000153s  proxy_CONNECT=503  target_status=000  curl_exit=7  msg="CONNECT tunnel failed, response 503"

$ ./proxycheck.sh http://127.0.0.1:46132 https://localhost:46443/
connect_to_proxy=0.000154s  proxy_CONNECT=200  target_status=000  curl_exit=60  msg="SSL certificate OpenSSL verify result: self-signed certificate in certificate chain (19)"

$ ./proxycheck.sh http://labuser:correct-horse@127.0.0.1:46128 https://localhost:46443/ --cacert certs/origin-ca.pem
connect_to_proxy=0.000154s  proxy_CONNECT=200  target_status=200  curl_exit=0  msg=""

Read it left to right. proxy_CONNECT=000 means you never got an answer from the proxy. Any other value there is the proxy talking. A 200 in that field with a failure afterwards moves the problem past the proxy, to TLS or to the target. If the target itself answers 403 or 429, that is a different family of problem, covered in 403 vs 407 vs 429.

Do not script against curl’s exit code for tunnel failures. On the same machine, against the same Squid, curl 8.7.1 exited 56 with curl: (56) CONNECT tunnel failed, response 407 while curl 8.23.0 exited 7 with the same text. Match on %{http_connect} instead.

Cause 1: the proxy rejected your credentials (407)

The proxy answered, and it wants a username and password it did not get. The verbose exchange shows exactly where:

$ curl -sv -x http://labuser:wrong@127.0.0.1:46128 https://localhost:46443/
* Proxy auth using Basic with user 'labuser'
* Establishing HTTP proxy tunnel to localhost:46443
> CONNECT localhost:46443 HTTP/1.1
> Host: localhost:46443
> Proxy-Authorization: Basic <redacted>
< HTTP/1.1 407 Proxy Authentication Required
< Server: squid/6.13
< X-Squid-Error: ERR_CACHE_ACCESS_DENIED 0
< Proxy-Authenticate: Basic realm="proxy-lab"
* Basic authentication problem, ignoring.
curl: (7) CONNECT tunnel failed, response 407

407 and the Proxy-Authenticate header are defined in RFC 9110 section 15.5.8. The request never left the proxy, so nothing about the target site is involved. Check, in this order: the password itself, the port (many gateways use the port to select a product), any parameters packed into the username, and whether the account authenticates by IP allowlist instead of a password.

The sneakiest version is a password containing @, #, : or /. Our second lab user has the password p@ss#word. Pasted raw into a proxy URL, every client misread it, and the error message does not mention credentials at all:

$ curl -sS -x 'http://labuser2:p@ss#word@127.0.0.1:46128' https://localhost:46443/
curl: (5) Could not resolve proxy: ss

The first @ ends the user information, as RFC 3986 section 3.2.1 specifies, so curl took ss as the proxy host. requests failed the same way (Failed to resolve 'ss'), and Node 24.2.0 refused to start at all, throwing InvalidArgumentError: invalid url from EnvHttpProxyAgent before the script’s first line ran. Percent-encode the password, or keep it out of the URL:

# encoded in the URL: @ is %40, # is %23
curl -x 'http://labuser2:p%40ss%23word@127.0.0.1:46128' https://localhost:46443/

# or passed separately, so no encoding is needed
curl -x http://127.0.0.1:46128 --proxy-user 'labuser2:p@ss#word' https://localhost:46443/

# Python: encode once, never by hand
python3 -c "from urllib.parse import quote; print(quote('p@ss#word', safe=''))"
p%40ss%23word

Both forms returned hello from origin through the authenticating port. Headless Chrome with no credential handler reported the 407 as net::ERR_INVALID_AUTH_CREDENTIALS. Library-by-library setup is in the curl guide and the Python guide.

Cause 2: your client cannot reach the proxy

The client never got a proxy to talk to. Either nothing listens at that address, the name does not resolve, or a firewall drops the packets. Chrome shows this page for all of them:

There is something wrong with the proxy server or the address is incorrect.
ERR_PROXY_CONNECTION_FAILED

Chromium’s own source describes PROXY_CONNECTION_FAILED (error -130) as a failure “in resolving its name, or in connecting a socket to it”, and says explicitly that it does not cover failures during CONNECT. Those get TUNNEL_CONNECTION_FAILED (-111) instead, per net_error_list.h. So the browser’s wording already tells you which hop failed.

Test the socket before you test anything else:

nc -vz gw.example.net 8000          # does anything accept TCP there?
dig +short gw.example.net           # does the name resolve at all?
env | grep -i _proxy                # is something else choosing the proxy for you?

The third line is the one people skip. A stale HTTPS_PROXY in a shell profile, a CI secret or a container image routes every request at a dead address, and the code you are staring at never mentions a proxy. curl confesses it in verbose mode:

$ HTTPS_PROXY=http://127.0.0.1:46999 curl -sv https://localhost:46443/
* Uses proxy env variable HTTPS_PROXY == 'http://127.0.0.1:46999'
*   Trying 127.0.0.1:46999...
* connect to 127.0.0.1 port 46999 from 127.0.0.1 port 56718 failed: Connection refused
* Failed to connect to localhost:46443 over proxy 127.0.0.1 after 0 ms: Could not connect to server

requests reads the same variables by default. Setting Session.trust_env = False turned that failure into a 200 in our test, and NO_PROXY=localhost or curl --noproxy '*' did the same for curl. The rules for which variable wins are in HTTP_PROXY, HTTPS_PROXY and NO_PROXY explained.

The opposite surprise also happens: a proxy you configured is silently not used. Node’s built-in fetch ignored HTTPS_PROXY pointing at a dead port and connected directly, until we set NODE_USE_ENV_PROXY=1, a variable the Node.js CLI docs list as added in v24.0.0 and v22.21.0. Chrome skipped the proxy for localhost until we passed --proxy-bypass-list="<-loopback>". Both explain the “works in curl, fails in the app” report, and its mirror image. Node specifics are in Using proxies in Node.js.

Cause 3: the right address, the wrong protocol

The proxy is up, but the client speaks a different protocol to it than the one it expects. Three variants turned up.

An https:// proxy URL for a plain HTTP proxy. The client starts a TLS handshake and the proxy replies in plain HTTP:

curl: (35) TLS connect error: error:0A00010B:SSL routines::wrong version number

urllib3 recognises this case and says so inside the requests exception: Unable to connect to proxy. Your proxy appears to only use HTTP and not HTTPS, try changing your proxy URL to be HTTP. The urllib3 advanced usage page it links to has the same advice. The scheme in a proxy URL describes how you talk to the proxy, not the site. An HTTPS site through an HTTP proxy is normal and still encrypted end to end, because the TLS runs inside the CONNECT tunnel.

An address that is a web server, not a proxy. Point a client at an ordinary site’s port and the CONNECT request gets a web server’s answer. Python’s built-in server replied 501 Unsupported method ('CONNECT'); curl printed CONNECT tunnel failed, response 501. Other web servers may answer 400, 404 or 405 at that point. The meaning does not change: wrong port, or a gateway hostname copied from the wrong field.

A SOCKS URL aimed at an HTTP proxy port, or the reverse. This one fails silently. curl sent a SOCKS5 greeting to Squid’s HTTP port, Squid waited for an HTTP request line, and nothing happened until our 8-second cap: curl: (28) Connection timed out after 8000 milliseconds. Python failed earlier, on a missing dependency: requests raised InvalidSchema: Missing dependencies for SOCKS support. and httpx raised ImportError: Using SOCKS proxy, but the 'socksio' package is not installed. Install requests[socks] or httpx[socks], as both projects’ docs say, then use the scheme your provider documents for that port. The difference between the two protocols is covered under SOCKS5.

Cause 4: the proxy refused the tunnel by policy (403)

The proxy understood you, knew who you were, and declined. Our open Squid port answered a CONNECT to port 46444 with:

HTTP/1.1 403 Forbidden
X-Squid-Error: ERR_ACCESS_DENIED 0

Access Denied.
Access control configuration prevents your request from being allowed at this time.

Nothing was wrong with the credentials or the target. Squid’s documented default configuration contains http_access deny CONNECT !SSL_ports (http_access), and its default acl SSL_ports port 443 lists a single port (acl). That follows RFC 9110 section 9.3.6, which says proxies supporting CONNECT should restrict it to known ports, and gives a CONNECT to port 25 relaying spam as the reason. Tunnelling to a database, SSH, SMTP or an unusual HTTPS port through a corporate or commercial proxy hits this rule first.

The fix belongs to whoever runs the proxy: an allowed port, an allowed destination, or a different product. A commercial gateway can also refuse a destination under its own usage policy. Its error documentation, not the status number, tells you which rule you hit.

Cause 5: the proxy could not reach the destination (502, 503, 504)

Your hop worked. The proxy’s hop to the site did not, and the proxy wrote you an error page saying so. Squid’s pages name the failure in an X-Squid-Error header and in plain text:

HTTP/1.1 503 Service Unavailable
X-Squid-Error: ERR_CONNECT_FAIL 61
Connection to 127.0.0.1 failed.
The system returned: (61) Connection refused

HTTP/1.1 503 Service Unavailable
X-Squid-Error: ERR_DNS_FAIL 0
Unable to determine IP address from host name no-such-host.invalid
The DNS server returned: Name Error: The domain name does not exist.

HTTP/1.1 502 Bad Gateway
X-Squid-Error: ERR_ZERO_SIZE_OBJECT 0
Zero Sized Reply
Squid did not receive any data for this request.

HTTP/1.1 504 Gateway Timeout
X-Squid-Error: ERR_READ_TIMEOUT 0
Read Timeout

The 504 arrived 4.68 seconds after the request, with read_timeout 4 seconds in our config. When an upstream hangs, the delay you see is the proxy’s timeout, not the site’s response time.

Status numbers for these failures are not consistent across software. RFC 9209 defines a Proxy-Status header with named error types and a recommended status for each. Here is what the two proxies we ran sent instead:

Upstream failure RFC 9209 type Recommended status Squid 6.13 sent Apache 2.4.67 sent
Connection refused connection_refused 502 503 ERR_CONNECT_FAIL 503 Service Unavailable
DNS name does not exist dns_error 502 503 ERR_DNS_FAIL 500 Proxy Error
Closed with no response connection_terminated 502 502 ERR_ZERO_SIZE_OBJECT 502 Proxy Error
No data before the read timeout connection_read_timeout 504 504 ERR_READ_TIMEOUT 502 Proxy Error
Denied by proxy policy http_request_denied 403 403 ERR_ACCESS_DENIED —

Neither server emitted Proxy-Status in these tests. The practical rule: never diagnose from the number alone. Read the error page body, the vendor header, or the proxy’s log line.

One trap cost us a rerun. Squid connected to localhost over ::1, our origin listened on IPv4 only, and the page said Connection to ::1 failed. A target that works from your laptop can fail through a proxy simply because the proxy prefers IPv6 and the service does not listen on it.

To fix this class: confirm the target is up from somewhere other than the proxy, then from the proxy’s network if you can. If it is transient, retry with a cap and jitter, and honour Retry-After when the proxy sends one. If it is one exit failing while others succeed, the exit is the problem, not your code.

When a website shows “Proxy Error”

This is the page from the top of this article, seen from the site owner’s side. Apache’s mod_proxy produced three variants on our rig, and its error log explained each one:

Backend behaviour Status line Page reason Error log
Closes the connection without a byte 502 Proxy Error Error reading from remote server AH01102: error reading status line from remote server
Accepts, never answers (timeout=3) 502 Proxy Error Error reading from remote server (70007)The timeout specified has expired
Hostname does not resolve 500 Proxy Error DNS lookup failure for: no-such-host.invalid AH00898: DNS lookup failure for
Port refuses connections 503 Service Unavailable The server is temporarily unable to service your request AH00957: http: attempt to connect to 127.0.0.1:46999 (127.0.0.1:46999) failed

The timeout row answered after 3.00 seconds, matching the timeout=3 we set on that ProxyPass. Note that the visitor sees an identical page for a crashed backend and a slow one. Only the log separates them.

Visitor: retry later, try another page on the same site to see whether the whole backend is down, and report it if it persists. Switching browsers, clearing cookies or changing your own proxy will not help, because the failure is between the site’s proxy and the site’s application.

Owner: grep the error log for the AH code, check the backend process and its port, and compare its normal response time with ProxyTimeout or the per-route timeout=; the mod_proxy documentation says ProxyTimeout defaults to the server’s general Timeout. Cloudflare, which is a reverse proxy too, uses its own codes for them, among them 521 (web server is down), 522 (connection timed out), 524 (a timeout occurred) and 526 (invalid SSL certificate), per its 5xx reference.

Cause 6: TLS interception

Some proxies, usually corporate ones or security appliances, decrypt HTTPS, inspect it, and re-encrypt it with a certificate they mint on the spot. Your client checks that certificate, does not recognise the issuer, and stops. Through Squid’s bumping port:

curl: (60) SSL certificate OpenSSL verify result: self-signed certificate in certificate chain (19)
requests.exceptions.SSLError: ... [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: self-signed certificate in certificate chain (_ssl.c:1010)
TypeError: fetch failed   cause: SELF_SIGNED_CERT_IN_CHAIN
net::ERR_CERT_AUTHORITY_INVALID   (Chrome: "Your connection is not private")

The proxy itself answered CONNECT with 200, which is why proxycheck.sh printed proxy_CONNECT=200 next to curl_exit=60. Confirm by comparing the issuer with and without the proxy:

$ echo | openssl s_client -proxy 127.0.0.1:46132 -connect localhost:46443 -servername localhost 2>/dev/null | openssl x509 -noout -issuer
issuer=CN=Lab Intercepting Proxy CA

$ echo | openssl s_client -connect localhost:46443 -servername localhost 2>/dev/null | openssl x509 -noout -issuer
issuer=CN=Lab Origin CA

If the issuer changes, something between you and the site is re-signing traffic. When that is your employer’s sanctioned inspection proxy, trust its CA explicitly for the tools that need it. Each of these turned the failure into hello from origin on our rig:

curl --cacert corp-proxy-ca.pem -x http://proxy.corp.example:3128 https://site.example/
REQUESTS_CA_BUNDLE=/path/corp-proxy-ca.pem python app.py      # requests
httpx.Client(proxy=PROXY, verify="/path/corp-proxy-ca.pem")   # httpx
NODE_EXTRA_CA_CERTS=/path/corp-proxy-ca.pem node app.mjs      # Node

The requests documentation describes REQUESTS_CA_BUNDLE as overriding the default bundle, so point it at a file that also holds the public roots if the same process talks to other sites. Do not reach for curl -k or verify=False as the fix. They hide interception you did not agree to just as well as the kind you did. Sites that pin certificates will keep failing behind an intercepting proxy whatever you trust; see certificate pinning and TLS interception.

Cause 7: the proxy hung up or went silent

A proxy that accepts the connection and then closes it gives the vaguest errors of all:

curl: (56) Proxy CONNECT aborted                 # https target
curl: (52) Empty reply from server               # http target
requests.exceptions.ConnectionError: ('Connection aborted.', RemoteDisconnected('Remote end closed connection without response'))
httpx.RemoteProtocolError: Server disconnected without sending a response.
net::ERR_EMPTY_RESPONSE

Node’s fetch did not fail at all. It waited until our own AbortSignal.timeout(6000) fired and then threw TimeoutError: The operation was aborted due to timeout. A proxy that accepts and never speaks behaves the same in every client: curl ends with (28) Connection timed out after 5001 milliseconds only because we passed -m 5.

Candidates to rule out: a protocol mismatch from cause 3, a proxy at its connection limit, and a load balancer in front of the proxy dropping idle connections. Set a timeout on every proxied request, log the exit and the time to failure, and compare against a second exit before blaming the target.

Cloudflare 10xx errors behind a proxy

A page that says “Error 1020” or “Error 1015” with a Ray ID is not from your proxy. Cloudflare sits in front of the site as a reverse proxy and generated the page at its edge because of a setting the site owner chose. Summarised from Cloudflare’s own pages for each code:

Code Cloudflare’s title Visitor side Site owner side
1003 Access Denied: Direct IP Access Not Allowed The request went to a Cloudflare IP address instead of the domain name. Use the hostname in the URL. —
1005 Access Denied: Autonomous System Number (ASN) banned The owner blocked the network your IP belongs to. Send the owner a screenshot of the error. Search Security Events for the Ray ID and decide whether that ASN should be allowed.
1006, 1007, 1008, 1106 Access Denied: Your IP address has been banned The owner blocked your client IP. Ask them to review it. Review the security settings that blocked the IP. Cloudflare says its support cannot override them.
1009 Access Denied: Country or region banned The owner blocked the country or region of your IP address. Contact the owner. Check the IP Access rules for the reported address.
1010 The owner of this website has banned your access based on your browser’s signature Blocked on the client’s browser signature. Notify the owner. Browser Integrity Check can be turned off in Security Settings.
1015 You are being rate limited Wait, then try again. Cloudflare warns that retrying repeatedly in a short period may extend the block. Review the rate limiting thresholds.
1020 Access denied A security rule with a Block action matched. Send the owner a screenshot with the Ray ID. Find the rule in Security Events and update it. Cloudflare returns 1020 with HTTP 403.

Two details from those pages matter when you are debugging through a proxy. Cloudflare states that requests blocked by WAF custom or managed rules do not get a 1020; they get the WAF block response instead, so the absence of a 1020 does not mean no rule fired. And every one of these codes is the site owner’s decision about traffic like yours. The legitimate path is the owner: contact them with the Ray ID, use an official API if one exists, or slow down for 1015. Rotating exits to keep hitting a site that has said no is how a rate limit turns into a ban.

Cause to fix, in one table

Cause Tell-tale text Diagnostic Fix
Proxy credentials response 407, Proxy-Authenticate curl -sv -x …, then read the Proxy-Authenticate line Correct password, port and username parameters; percent-encode or use --proxy-user
Special characters in the URL Could not resolve proxy: plus a fragment of your password Print the parsed host: python3 -c "from urllib.parse import urlsplit; print(urlsplit(URL).hostname)" quote(password, safe='')
Proxy unreachable ERR_PROXY_CONNECTION_FAILED, Could not connect to server nc -vz host port, dig +short host Correct host and port, open the firewall, start the proxy
Stale proxy settings Uses proxy env variable HTTPS_PROXY printenv HTTPS_PROXY https_proxy HTTP_PROXY ALL_PROXY Unset it, add NO_PROXY, or trust_env = False
Wrong proxy scheme wrong version number, appears to only use HTTP Try the same proxy with http:// Use the scheme the proxy speaks, not the site’s
Not a proxy port response 501, 400, 404, 405 on CONNECT curl -v http://host:port/ without -x Use the documented gateway host and port
SOCKS and HTTP mixed up A hang, or Missing dependencies for SOCKS support Check which protocol the port speaks Match the scheme; install requests[socks] or httpx[socks]
Port or destination not allowed response 403, ERR_ACCESS_DENIED Same request to port 443 Ask the proxy operator to allow it, or use another route
Destination unreachable from the proxy 502, 503, 504 with a proxy error page Read X-Squid-Error or the page reason Fix or wait for the target; retry with backoff
Reverse proxy backend down 502 Proxy Error, Bad Gateway Server error log (AH01102, AH00957) Owner restarts or fixes the backend, tunes the timeout
TLS interception self-signed certificate in certificate chain openssl s_client -proxy and compare issuers Trust the sanctioned CA explicitly; never disable verification
Proxy hung up Proxy CONNECT aborted, Empty reply proxycheck.sh against a second exit Timeouts on every request; check connection limits
Cloudflare 10xx Error 10xx and a Ray ID None on your side; the code names the rule type Contact the site owner; wait out 1015

Turning exceptions into a cause in Python

httpx keeps the proxy’s status in a separate exception class, which makes it easy to log the failing hop instead of a stack trace. Tested with httpx 0.28.1 against the rig above:

import sys, httpx

def classify(proxy: str, url: str, verify=True) -> str:
    try:
        r = httpx.get(url, proxy=proxy, timeout=15, verify=verify)
        return f"target answered {r.status_code}"
    except httpx.ProxyError as e:          # proxy answered CONNECT with a non-2xx status
        return f"proxy refused the tunnel: {e}"
    except httpx.ConnectTimeout:
        return "timed out reaching the proxy or the target"
    except httpx.ConnectError as e:        # proxy unreachable, DNS, or TLS verification
        msg = str(e)
        if "CERTIFICATE_VERIFY_FAILED" in msg:
            return f"TLS verification failed (interception or bad cert): {msg}"
        return f"could not reach the proxy: {msg}"
    except httpx.RemoteProtocolError as e:  # proxy or target hung up mid-exchange
        return f"connection closed without a response: {e}"

if __name__ == "__main__":
    print(classify(sys.argv[1], sys.argv[2]))
$ python classify.py http://127.0.0.1:46999 https://localhost:46443/
could not reach the proxy: [Errno 61] Connection refused
$ python classify.py http://127.0.0.1:46128 https://localhost:46443/
proxy refused the tunnel: 407 Proxy Authentication Required
$ python classify.py http://127.0.0.1:46140 https://localhost:46443/
connection closed without a response: Server disconnected without sending a response.
$ python classify.py http://127.0.0.1:46132 https://localhost:46443/
TLS verification failed (interception or bad cert): [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: self-signed certificate in certificate chain (_ssl.c:1010)

With requests the same information is there, but nested: the proxy’s status line sits inside the message as Tunnel connection failed: 407 Proxy Authentication Required, under requests.exceptions.ProxyError. Matching on that substring works across the versions we ran.

Frequently asked questions

What does "Proxy Error" mean on a website?
It is the error page of a reverse proxy in front of the site; the exact wording "Proxy Error" with "Error reading from remote server" comes from Apache httpd's mod_proxy. The proxy could not get a valid answer from the application behind it. A visitor can only retry later; the owner has to check the backend and the proxy's error log.
Is a proxy error my fault or the website's?
Check whether the proxy answered your CONNECT request. With curl, -w '%{http_connect}' prints 000 when you never reached the proxy, 407 or 403 when the proxy refused you, and 200 when the tunnel opened and the failure came later, from TLS or from the site.
What is the difference between ERR_PROXY_CONNECTION_FAILED and ERR_TUNNEL_CONNECTION_FAILED?
Chromium uses ERR_PROXY_CONNECTION_FAILED (-130) when it cannot resolve or connect to the proxy itself, and ERR_TUNNEL_CONNECTION_FAILED (-111) when the proxy answered but the CONNECT tunnel was not established. In our tests a 403, a 501 and a 503 from the proxy all produced the second one.
Why does curl say "CONNECT tunnel failed, response 407"?
The proxy wants credentials it did not receive or did not accept. Check the password, the port and any parameters packed into the username, and percent-encode special characters or pass the credentials with --proxy-user.
What does "proxy error 429" mean in an app?
When an app shows a proxy error with a status code, that code came from a server it reached through a proxy or gateway. 429 is Too Many Requests, defined in RFC 6585: the service rate-limited you. Slow down and honour Retry-After instead of switching proxies.
Can I fix a Cloudflare 1020 or 1015 error myself?
Nothing on your side changes the rule. Cloudflare's documentation says a 1020 means a security rule set by the site owner blocked the request, and a 1015 means the owner's rate limit was exceeded. Send the owner the Ray ID for a 1020, and wait before retrying a 1015, because Cloudflare warns that repeated attempts can extend the block.

Sources

  1. RFC 9110: HTTP Semantics (CONNECT 9.3.6, 407 15.5.8, 502-504 15.6)
  2. RFC 9209: The Proxy-Status HTTP Response Header Field
  3. RFC 3986 section 3.2.1: User Information
  4. RFC 6585 section 4: 429 Too Many Requests
  5. RFC 6761 section 6.4: the reserved .invalid domain
  6. Chromium net_error_list.h
  7. libcurl error codes
  8. Squid configuration directive: http_access
  9. Squid configuration directive: acl
  10. Apache HTTP Server 2.4: mod_proxy
  11. urllib3 advanced usage: HTTPS proxy error with an HTTP proxy
  12. Requests advanced usage: proxies, SOCKS and CA bundles
  13. HTTPX: proxies and SOCKS support
  14. Node.js CLI: NODE_USE_ENV_PROXY
  15. Linux asm-generic/errno.h
  16. Cloudflare: 1xxx errors
  17. Cloudflare: Error 1003
  18. Cloudflare: Error 1005
  19. Cloudflare: Errors 1006, 1007, 1008 or 1106
  20. Cloudflare: Error 1009
  21. Cloudflare: Error 1010
  22. Cloudflare: Error 1015
  23. Cloudflare: Error 1020
  24. Cloudflare: 5xx errors

Read this page as Markdown · Quote it freely under CC BY 4.0 with a link back.