Eric Swanson lachesis

I'm experiencing the weirdest internet issue. I use Comcast Xfinity in El Cerrito, CA. I have a new Arris SB6190 cable modem running firmware version 9.1.93N. A few days ago, I noticed that new tabs start taking a few seconds to load, and curl requests to example.com took ~5.1 seconds. However, if I specified -4 or -6 to curl, it resolved immediately. I immediately suspected DNS, but "host" resolved immediately as well.

So this was too weird to ignore. I set my resolv.conf to point to 8.8.8.8 (instead of my EdgeRouter X) and did:

time strace -f -T curl -s http://www.example.com > /dev/null 2> bad
time strace -f -T curl -4 -s http://www.example.com > /dev/null 2> good

Now if you look at the resulting logs, you'll see that curl starts a new thread to handle DNS resolution. In the good logs, the relevant bit looks like:

[pid 12438] socket(AF_INET, SOCK_DGRAM|SOCK_NONBLOCK, IPPROTO_IP) = 3 <0.000005>

	#!/usr/bin/python
	import argparse
	import re
	import sys
	import requests

	def main():
	SHOW_GROUPS = [
	r'SYNOPSIS',
	r'DISCUSSION',

	#!/usr/bin/python
	# usage: python3 dump_cosmos_staking_rewards.py < address_list.txt > staking_rewards.tsv
	import decimal
	import fileinput
	import collections
	import re
	import sys
	import requests
	import datetime
	import json

	#!/usr/bin/python3
	import json
	import sys
	import requests

	def get_transactions(address):
	resp = requests.get(
	'https://api-osmosis.cosmostation.io/v1/account/new_txs/' + address,
	params={'limit':5000, 'from': 0},
	timeout=31,

	# License: Public Domain, use as you like, no warranty
	import time
	import threading
	import pylru # from pip

	class ExpiringLRUCache(object):
	def __init__(self, size, exp=7200):
	self.exp = exp
	self.size = size
	self.lru = pylru.lrucache(size)

	#!/usr/bin/env python3
	# Get duke energy hourly electricity usage
	# before running: pip3 install requests
	# use env vars to set username, password, and lookback days
	# e.g. [email protected] DUKE_PASSWORD=hunter11 DUKE_DAYS_BACK=14 python dukeusage.py

	import os
	import datetime
	import json
	import re

	#!/usr/bin/python3
	# Prints current PM2.5 concentrations based on fire.airnow.gov map
	# pip3 install requests pytz geobuf
	# usage: LAT=37.9053745 LNG=-122.3048239 MAX_FEATURES=6 MAX_DIST=11 python3 airnow.py
	# optional, for AQI instead of raw concentration (µg/m³): pip3 install python-aqi and pass AQI=1
	import collections
	import datetime
	import math
	import os
	import requests

	#!/usr/bin/env python3
	# Subscribes contacts in a CSV file to a Close campaign
	# Requires python3.7 or newer (f-strings)
	# Prepare: save this script in close_sub.py and `pip3 install closeio`
	# CSV should be two columns with email followed by exact Sequence name on Close
	# Usage: KEY=api_yourcloseiokey LIVE=0 python3 close_sub.py leads.csv
	import csv
	import fileinput
	import os
	import re

	#!/bin/bash
	# Get last value of statsd gauge from grafana server
	# Set env vars:
	# GRAFANA_BASE="https://www.example.com" <-- your grafana base URL
	# GRAFANA_API_KEY="eyJryJzd.....IxMj0F9" <-- your grafana API key

	curl --compressed --silent \
	"$GRAFANA_BASE/api/datasources/proxy/1/render" \
	-H "Authorization: Bearer $GRAFANA_API_KEY" \
	--data-urlencode "target=summarize($1, "'"1h", "last")' \

	from __future__ import print_function, division, unicode_literals
	import sys, time, re, struct, base64, hmac, hashlib
	def get_hotp(secret, counter):
	"""Return the HMAC-Based One-Time Password for the the given secret (base32 encoded) and the counter.
	>>> [get_hotp('GEZDGNBVGY3TQOJQGEZDGNBVGY3TQOJQ', i) for i in xrange(10)]
	[755224, 287082, 359152, 969429, 338314, 254676, 287922, 162583, 399871, 520489]
	"""
	secret = re.sub(r'\s*', '', secret) # remove whitespace
	secret = secret.upper() # uppercase
	if len(secret) % 8 != 0: