Doug Madory (also on Bluesky)
@DougMadory
Followers
10K
Following
7K
Media
1K
Statuses
6K
Director of Internet Analysis at @Kentikinc. Alum of @dynresearch @renesys @uva @dartmouth @usairforce
New Hampshire
Joined August 2013
Throttling of Social Media in Türkiye During Protests: An Analysis of OONI Data https://t.co/DfgtIWPkj8 Our latest report shares OONI data on the throttling of Twitter/X, Facebook, Instagram, WhatsApp, YouTube, TikTok, and Signal in Türkiye during protests earlier this year.
0
2
5
As a follow-up, I published this tweet at 5:12pm ET (22:12 UTC) on Nov-14, these routes were withdrawn at 22:36 UTC on Nov-14. 🤷♂️ (cc @DaryllSwer)
@DaryllSwer Please pass on that when @Starlink recently moved routes from AS14593 to AS45700 (PT Starlink Services Indonesia), they never changed the ROAs to match the new origin. Those routes are being evaluated as RPKI-invalid, are being filtered and are essentially useless.
1
2
7
We’ll be watching for how these mitigations roll out. Outages like this are happening more than they should lately. They shouldn’t happen, but when they do, honest post-mortems help make the Internet stronger. #Cloudflare #InternetOutage #IncidentResponse
0
1
2
Waymo’s fully autonomous driving technology offers greater independence for people with intellectual and developmental disabilities.
0
0
0
Incidents like this are a reminder that complexity is the enemy of reliability — even for the best-resourced networks. Cloudflare’s transparency here is valuable for the whole industry.
1
0
5
In the write-up, Cloudflare outlines several follow-ups: stricter controls around internally-generated config files, a global kill-switch for features, and deeper reviews of failure modes tied to memory and feature-limits.
1
0
0
What’s notable: the outage touched not just their CDN and security services, but also the dashboard, authentication systems, and Workers KV — highlighting how interdependent large platforms are.
1
0
0
By stopping propagation of the bad file and rolling back to a known-good version, traffic started recovering by ~14:30 UTC and fully stabilized by 17:06 UTC.
1
0
0
The culprit was an internal permissions change that caused duplicate feature rows in a ClickHouse query. That oversized file hit a hard limit in their proxy module → panic → cascading failures. Initially, they suspected a large DDoS (reasonable, given the symptoms).
1
1
1
Around 11:20 UTC, Cloudflare’s core traffic began failing worldwide. Many sites behind Cloudflare returned error pages — not because of an attack, but because a feature-file used by their Bot Management system doubled in size unexpectedly.
1
0
0
🧵 Digging into Cloudflare’s detailed post-mortem on the Nov 18, 2025 outage. Tells the story of how a small internal change can ripple across a massive, globally distributed platform. https://t.co/9KvHvcZaXG
blog.cloudflare.com
Cloudflare suffered a service outage on November 18, 2025. The outage was triggered by a bug in generation logic for a Bot Management feature file causing many Cloudflare services to be affected.
Twitter/X coming back after @Cloudflare's platform-wide outage. Traffic volume (bits/sec) by source CDN based on @kentikinc aggregate NetFlow:
1
2
9
Twitter/X coming back after @Cloudflare's platform-wide outage. Traffic volume (bits/sec) by source CDN based on @kentikinc aggregate NetFlow:
Major Internet content provider @Cloudflare is experiencing a platform-wide outage disrupting numerous services and websites around the world. According to @kentikinc aggregate traffic data, disruptions began at 11:32 UTC.
15
5
23
I won’t mince words: earlier today we failed our customers and the broader Internet when a problem in @Cloudflare network impacted large amounts of traffic that rely on us. The sites, businesses, and organizations that rely on Cloudflare depend on us being available and I
1
0
1
I won’t mince words: earlier today we failed our customers and the broader Internet when a problem in @Cloudflare network impacted large amounts of traffic that rely on us. The sites, businesses, and organizations that rely on Cloudflare depend on us being available and I
831
2K
17K
Twitter/X coming back after @Cloudflare's platform-wide outage. Traffic volume (bits/sec) by source CDN based on @kentikinc aggregate NetFlow:
Major Internet content provider @Cloudflare is experiencing a platform-wide outage disrupting numerous services and websites around the world. According to @kentikinc aggregate traffic data, disruptions began at 11:32 UTC.
15
5
23
Major Internet content provider @Cloudflare is experiencing a platform-wide outage disrupting numerous services and websites around the world. According to @kentikinc aggregate traffic data, disruptions began at 11:32 UTC.
0
6
17
@DaryllSwer @Starlink AS45700 started announcing 2605:59c0:502b::/48 and 2605:59c0:5020::/48 back on August 31st. The ROA covering them has always had an origin of AS14593, so they have never been properly routed. https://t.co/9qq2JedbSz
1
1
1
🚀 New on @isoc_pulse — I analyzed 3.5 years of Starlink IP geolocation data. From Gaza to Mars, the results reveal how this network has expanded (and surprised me) around the world.
From Antarctica to Mars, @DougMadory digs further into the @Starlink IP Geolocations he's collected since April 2022
1
6
13
@DaryllSwer Please pass on that when @Starlink recently moved routes from AS14593 to AS45700 (PT Starlink Services Indonesia), they never changed the ROAs to match the new origin. Those routes are being evaluated as RPKI-invalid, are being filtered and are essentially useless.
2
2
3
Be sure to check out the interactive visualization at the end!
1
1
4
🚀 New on @isoc_pulse — I analyzed 3.5 years of Starlink IP geolocation data. From Gaza to Mars, the results reveal how this network has expanded (and surprised me) around the world.
From Antarctica to Mars, @DougMadory digs further into the @Starlink IP Geolocations he's collected since April 2022
1
6
13