Cloudflare status: hosting issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of Cloudflare reports received over the last 24 hours by time of day. When the number of reports exceeds the baseline, represented by the red line, an outage is determined.

Most Reported Problems

The following are the most recent problems reported by Cloudflare users through our website.

• Cloud Services (57%)
• Hosting (29%)
• E-mail (7%)
• Domains (7%)

Live Outage Map

The most recent Cloudflare outage reports came from the following cities:

City	Problem Type	Report Time
Tlajomulco de Zúñiga	Cloud Services	4 days ago
Asnières-sur-Seine	Cloud Services	8 days ago
New York City	E-mail	13 days ago
Township of Evan	Hosting	13 days ago
Ahmedabad	Cloud Services	19 days ago
Le Puy-en-Velay	Cloud Services	20 days ago

Full Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

Cloudflare Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Jonathan.G (@Jonathan2134) reported 7 minutes ago

  In reality CloudFlare makes searching the internet worst place because number of websites you can't use despite not VPN.

• 
  Innocent Bigega (@EnzInnocent) reported 14 minutes ago

  Stopping the bad guys with Cloudflare: 17,256 malicious requests blocked or challenged in the last month #cloudflare

• 
  Reina Cruz 🥼🧤🇨🇺 (@rea1ReinaCruz) reported 21 minutes ago

  @Cloudflare Fix human verification

• 
  Ric Orna (@ric_orna) reported 22 minutes ago

  @yongfook Or you gotta be doing something that actually has a moat still. Eg a network of sensors and API. Nobody is gonna vibe code physical stuff. Or something so big and boring a lift that nobody would want to and couldn’t in short order anyway. Vibe code your Cloudflare competitor.

• 
  2$p00ky (@dos2p00ky) reported 25 minutes ago

  No ******* way that’s really you messaging in that chat…but then I accidentally clicked the button, the chatbot responded, I sent an emoji; then you sent a name and email template & then suddenly OF was also trying to track along with ur site through cloudflare 💀 What a ******* webhook to stay connected to every chat someone has open from ur webpage authenticated by the network cert…holy **** 🥵 for that setup

• 
  Atif (@ioAtif) reported 25 minutes ago

  @acolombiadev @Cloudflare @coderhq Although it's not a service but a whole platform

• 
  Villainess of Resevoi (@GalCharoShipper) reported 26 minutes ago

  @Tepafray @Sean_Topic AI crawlers simply ignore ****. Even layered defenses recommended by artist-protection groups (Glaze for image-level noise, Cloudflare-style bot blocking, watermarking, opt-out registries) are described as a partial, constantly-shifting arms race, not a solved problem

• 
  Sergei Aksjonov (@sergeiaksjonov) reported 28 minutes ago

  @Printful Also this: 1 day of my time 1 domain on Cloudflare $20 on AI That’s the whole budget. 10 years ago I thought you needed a Social account of the service, a pretty brand, a PR campaign, the “right” image. This time I just launched it as is from my own account. No polish. No theater. Just shipped. And the first order still came in. Just start as it is. Let it flow.

• 
  Patrick Ryall (@PadraigOraghail) reported 28 minutes ago

  The learning is never ending. Mostly it is platforms, in this case using Cloudflare as the auth gate. I had never used cloudflare for more than DNS. Now I have three sites running on it, for free, which is mad. Workers, Zero Trust, DNS, connected to ***. It's all rather slick.

• 
  Bobby Lansing (@dataguybobby) reported 29 minutes ago

  Bot is grepping logs in CloudFlare, Azure, Supabase, and Vercel. Then triggering Cursor agents to build a fix. I am mentioning issues and approving PRs.

• 
  Lex (@TheLexTimes) reported 30 minutes ago

  Hey @Cloudflare I reached out via DM about WARP constantly disconnecting. Is there a direct contact I can talk to? Happy to provide any data to help figure this issue out.

• 
  Stuti (@stutireal) reported 31 minutes ago

  tired: dog ate my homework wired: cloudflare was down inspired: three consecutive ai civilizations rose and fell inside my macbook during a forced restart. My notes app and chrome tabs did not survive the third regime

• 
  Bryan Beal 🎧 (@bryanrbeal) reported 33 minutes ago

  Pro tip - disable Eero’s DNS and use Google or Cloudflare DNS and 99% of Eeros problems disappear Eero DNS is trash

• 
  avrin (@itsavrinwave) reported 36 minutes ago

  Is cloudflare down again??

• 
  ⚛︎ Daddio ⚛︎ (@ScienceDaddio) reported 37 minutes ago

  @grok @DaddioAI E4H.1 is implemented. The pump-activity table is now the reusable presentation surface; events are optional overlays, not duplicate source tables. Raw duration rows stay immutable. E4G.6B canonical stable-ID work is intact (trend_report.html, trend_report.js, test/trend-report-stable-id-e4g6b.test.js). The workspace is not a *** repo, so that was confirmed by file inspection, not *** status. Codex had only started reading; no detector or research files were rewritten. ─── 1. Working-tree inspection No *** repo. Inspected files vs the E4G.6B baseline. E4H.1 presentation files were already present from the interrupted session. This closeout preserved that work, then stopped overlaying Current Assembly / Station Packet tables so automatic assembly still uses detector is_event_row highlighting instead of the new ERT overlay. 2. Distinction from E4G.6B E4G.6B remains: stable IDs and trend_report.js?v=e4h1-e4g6a-e4g6b. E4H.1 added presentation state, ERT overlay, settings UI, and Search Tables controls. Research under research/e4c8g/ was not touched. 3. Existing table architecture found Canonical BossDataTable renders Trend Report, Search Tables, and exports. Manual/Search tables use the approved catalog with exact start/end, pump, and sort. Event tables keep Context Before/After. Snapshots store source columns/rows; presentation is a separate optional field. EMAIL-v1 already highlights via is_event_row. 4. Existing fixed ERT architecture ┌──────────────────────────────────────┬──────────────────────────────────────┬─────────────────────┐ │ Layer │ Behavior │ E4H.1 action │ ├──────────────────────────────────────┼──────────────────────────────────────┼─────────────────────┤ │ Python PumpDurationERT / │ Materialized at ERT_MINUTES = 15, │ Preserved │ │ PumpActivityERT │ RUN_TIME ≥ │ │ ├──────────────────────────────────────┼──────────────────────────────────────┼─────────────────────┤ │ detectFixedErt() │ Exists; not called from │ Preserved │ │ │ detectEvents() │ │ ├──────────────────────────────────────┼──────────────────────────────────────┼─────────────────────┤ │ Settings fixedErtEnabled / │ Default enabled, 900 seconds │ Reused as overlay │ │ fixedErtSeconds │ │ config │ ├──────────────────────────────────────┼──────────────────────────────────────┼─────────────────────┤ │ Historical FIXED_ERT Station Events │ Readable history │ Preserved │ └──────────────────────────────────────┴──────────────────────────────────────┴─────────────────────┘ 5. Existing Adaptive Long Runtime architecture detectAdaptiveRuntime() remains in detectEvents(). Station Events, Current Assembly, and report generation still consume Adaptive events. No silent detector swap. 6. Settings architecture Existing analysis settings: global row + per-station overrides in analysis-settings.js, Station Review cards, PUT /api/station-review/settings/global and station override PUT. No parallel settings universe. 7. ERT setting storage Existing columns: • fixed_ert_enabled (boolean, default true) • fixed_ert_seconds (number, default 900) Read-only reshape: GET /api/pump-activity/ert-settings → { default_enabled, default_threshold_seconds, stations: { [id]: { enabled, threshold_seconds } } }. Writes go through the existing Station Review settings APIs. No new write endpoint. 8. ERT defaults 15 minutes (900 seconds) when no station override exists. Sands is not hardcoded. 9. Threshold semantics Granular rows only: RUN_SECONDS >= station.threshold_seconds Annotation metadata records comparison: 'gte'. Equality counts (6:00 at a 6-minute threshold is ERT; 5:59 is not). Validation range: finite, 1 to 2,592,000 seconds (1 second through 30 days). That is the existing analysis-settings bound. UI is minutes: 0.0167–43,200, converted × 60 on save. Allowed values include 6, 10, 15, 17, 20, 30 minutes. Invalid overlay input clamps to 900 seconds; settings save still rejects out-of-range values. 10. Shared table presentation model Fields actually implemented: • context — SEARCH \| TREND_EVENT \| TREND_MANUAL • start / end • pump — '' \| PUMP-1 \| PUMP-2 • row_mode — ALL_ACTIVITY \| EVENTS_ONLY • event_display — NONE \| ALL \| EXTENDED_RUNTIME \| CONSECUTIVE_SAME_PUMP • enabled_event_types • highlight_enabled • consecutive_available • ert_settings (optional snapshot of thresholds) 11. Annotation provider model src/public/pump-activity-presentation.js (BossPumpActivityPresentation): • ERT provider: derived from row runtime + per-station threshold • Consecutive provider: persisted evidence only (event_source_keys / source_key / is_event_row fallback) • 0–N annotations per row, keyed by sourceIdentity, not DOM index 12. Raw data integrity applyPresentation clones rows. Source objects are not mutated. Annotations are not written into duration history. Tests stringify the source array before and after. 13. Trend Report integration User-inserted granular tables get a collapsed Table Display card: Start, End, Pump, Rows, Events, Highlight Events, Apply. Event-from-tray tables default All Events + Highlight ON. Manual/Search-style inserts default None + Highlight OFF. Apply updates the same Canvas section (no duplicate). Context Before/After still exist; once they produce a new window, annotations follow that window. Current Assembly and Station Packet tables do not attach overlay presentation and do not show Table Display. They keep detector is_event_row highlighting. 14. Search Tables integration Granular duration tables (pump_runs, pump_durations, PumpDurationERT, rolling48_all, *_durations) show Rows / Events / Highlight. Defaults: Events: None, Highlight: Off, All Pump Activity. Consecutive is not offered. Aggregate tables (PumpActivityDaily, PumpActivitySQL, PumpActivityERT, StationActivityDaily) stay plain with the bar hidden. 15. Generate Report / aggregate handling Generate Report was not merged or redesigned. PumpActivityDaily MAX above threshold is not treated as a granular ERT event (aggregate: true, no Event column). 16. Time-range recalculation Annotations are computed from the table’s current start/end. Expand the window → newly visible ERT rows annotate. Contract → out-of-range annotations disappear. The table is not married to the initiating event. 17. Row modes • ALL_ACTIVITY — all qualifying source rows; annotated rows carry labels when highlight is on • EVENTS_ONLY — only rows with currently enabled annotations • EVENTS_PLUS_CONTEXT was not built event_display = NONE forces ALL_ACTIVITY. 18. Highlight toggle / plain mode Highlight OFF (or Events = None): no event-row class, no Event column, no event_labels / event_annotations, no threshold text in the row payload. Same source rows, visually plain. 19. Pump filter Both / Pump 1 / Pump 2 preserved. Annotations follow remaining rows. 20. Sorting Existing newest/oldest/runtime sorts remain. Highlight class uses sourceIdentity, not row index. 21. Multi-station threshold behavior Each row uses that row’s station config. Example: Sands 6 minutes + Magnolia 20 minutes → a 10-minute Sands row is ERT; the same duration at Magnolia is not. 22. Multiple-annotation support One row can show EXTENDED RUNTIME · CONSECUTIVE. Structure is a list, not one-event-per-row. 23. Consecutive compatibility Now: Trend Report event tables may show persisted Consecutive annotations when source keys or is_event_row evidence exists. Deferred: no detectConsecutiveRuns() changes, no LG30/C1, no historical reconstruction. Search Tables are ERT-only because complete Consecutive evidence is not available there. 24. Adaptive legacy behavior Preserved. Adaptive still feeds Station Events and Current Assembly. The new overlay is fixed-threshold ERT only. 25. Legacy fixed ERT tables PumpDurationERT and PumpActivityERT remain in the catalog and Python pipeline at 15 minutes. 26. Canonical snapshot / save-reopen presentation is optional on table sections. Legacy snapshots with presentation: null still validate and render with is_event_row. New snapshots store view state (including ert_settings when present) and reapply on reopen. 27. EMAIL rendering HUB presentTableSection applies the overlay when presentation is set and the table is not assembly-bound. EMAIL-v1 was not modified. Highlight still uses is_event_row; HUB sets that flag from overlay identity when annotations are shown. Plain presentation leaves rows unhighlighted and without an Event column. Cloudflare requests stay 0 in tests. 28. Auto/custom narrative autoTableNarrative() exists as a helper only. It is not wired into Current Assembly. Deferred to E4H.2. 29. Future AI compatibility A later local AI can set the same structured presentation object (station/start/end/pump/row_mode/event_display/highlight_enabled) without HTML-specific logic. No AI was implemented. 30. Files created • src/public/pump-activity-presentation.js • test/pump-activity-presentation-e4h1.test.js • test/pump-activity-table-integration-e4h1.test.js 31. Files modified • src/public/trend_report.js • src/public/trend_report.html • src/public/data_views.js • src/public/data_views.html • src/public/station_review.js • src/public/station_review.html • src/public/styles.css • src/server.js • src/services/saved-reports.js • src/services/trend-report-email.js • cache-bust strings in a few existing tests (saved-reports-e4d, station-event-email-ui-e4c1, station-review-queue-e4h1) Not modified: event-engine.js, analysis-settings schema (reused), Python ERT builders, EMAIL-v1, research/. 32. Focused E4H.1 tests 30 / 30 33. Search Tables regression 40 / 40 (catalog, exact time, pump, historical durations, shared renderer) 34. Trend Report regression 85 / 85 on X4 + E4G.6 / 6A / 6B core files; remainder included in full HUB 767. 35. Saved Report regression 16 / 16 (saved-reports-e4d + report-table-compatibility-e4d2) 36. Event/detector regression 54 / 54 (local events, event quality, fixed-ERT workflow, event presentation) 37. HUB full regression Starting baseline 737 / 737. 767 / 767 (737 prior + 30 E4H.1 tests) A first parallel run hit 766/767 from two suites sharing temp files; the serial rerun was 767/767. 38. EMAIL full regression 64 / 64 39. Capture reliability 9 / 9 40. Production DB writes 0 in tests. ERT settings GET is read-only. 41. Real emails sent 0. Cloudflare requests 0. 42. Real captures 0 43. Detector files modified NO. detectConsecutiveRuns() unchanged. detectFixedErt() still unused by detectEvents(). 44. Research files modified NO 45. Manual owner QA Live BOSS UI was not opened in a browser from this session; behavior was verified with node --test. Please confirm: 1. Station Review → settings: EXTENDED RUNTIME Enabled + Threshold (minutes). Set a station to 6.0; confirm storage is 360 seconds. Confirm Adaptive controls are unchanged. Confirm email/SMS are not flipped on. 2. Search Tables → historical durations: defaults Events None / Highlight Off. A 20-minute row looks like any other row. Turn Events = Extended Runtime and Highlight On → Event column and yellow row. Highlight Off → same rows, fully plain. 3. Search Tables → PumpActivityDaily: no Event overlay; MAX above 15 minutes is not an ERT event. 4. Trend Report: drag a Station Event table. Table Display exists. Defaults All Events / Highlight On. Events Only hides ordinary rows. Changing Start/End then Apply reloads annotations for the new window. Apply does not create a second Canvas section. 5. Current Assembly: no Table Display / Apply. Consecutive/Adaptive detector rows still highlight via is_event_row. No extra Event column from the overlay. 6. Save / reopen / email a mixed report: annotated user table stays annotated; a plain table stays plain; a pre-E4H.1 saved report still opens. 7. Multi-station granular table: different thresholds on two stations; a 10-minute run annotates only the 6-minute station. 46. Recommended E4H.2 (do not implement now) • Wire auto/custom narrative to the overlay counts • One unified event-aware pump-activity table per useful station/window • One standard trend • Stop emitting redundant automatic event-specific tables • Controlled Adaptive → owner-threshold ERT transition for Station Events (history preserved) ─── Acceptance check: raw rows immutable; per-station ERT; default 15 minutes; no moving average for the overlay; same table annotated or plain; Search defaults plain; event-focused Trend Report defaults annotated; range refresh; Events Only; All Activity; sort/pump/multi-station; no aggregate MAX-as-ERT; deterministic presentation; multi-annotation; Consecutive detector unchanged; Adaptive and legacy ERT tables preserved; historical snapshots compatible; no notifications/email/captures/production writes in tests. Stopped here. Automatic report assembly was not redesigned.

• 
  Jeffinator (@jeffinator06) reported 40 minutes ago

  ive been STUCK in cloudflare HELL for the past 2 WEEKS. this is the 2nd time ive been caught up in some mass-wide bullshit. holy **** dude reddit is so so awful i dont want to go back

• 
  Ben Hocking (@bmwhocking) reported 41 minutes ago

  @rustie5555 There are a lot of round robin API calls for content. Works fine if you are close to their US data-centre. Sucks for us. They are using Cloudflare, I don’t believe they are processing at Cloudflare’s edge, just caching content. They need better & faster timeline building.

• 
  Brendan Ryan (@BrendanRyanM) reported 42 minutes ago

  @mntruell @petergyang @bot Also having an issue getting past bot blockers, e.g. cloudflare prompts to press a button and prove you’re human. Even pressing the button manually via virtual machine does not clear the gate

• 
  James (@jamescoder12) reported 42 minutes ago

  The uncomfortable truth. Your ISP has no incentive to optimize your router settings. They have every incentive to leave them on factory default and every incentive to upsell you when the defaults cause problems. When your Wi-Fi is slow, you call and complain. They test Speed 1 the connection to your modem. It's fine. So they tell you "everything looks good on our end" which is technically true. Then they offer you a faster plan. You pay $30/month more. Speed 1 goes up. Speed 2 the Wi-Fi to your devices doesn't change. Because the bottleneck was never the pipe coming into your home. It was the 6 feet between your router and your laptop. When your firmware is 4 years out of date, they don't push an update. When your channel is congested, they don't suggest switching. When your DNS is slow, they don't mention Cloudflare. When your channel width is halved, they don't widen it. When your security is on WPA2 with the factory admin password, they don't flag the risk. Every free fix that would make your Wi-Fi faster is a fix that prevents them from upselling you a more expensive plan. Every month you rent their router is $14 in revenue for a $60 device they deployed years ago. The rental has no end date. The equipment depreciates. The revenue doesn't. He called his ISP 4 times. They suggested upgrading twice. He upgraded twice. He paid $360 extra over a year for a faster plan that didn't make his Wi-Fi faster because the problem was never the plan. It was 9 settings his ISP could have fixed in a 5-minute phone call. They chose to sell him a faster highway instead of widening the off-ramp. "Your ISP isn't slow. Your settings are. And the company delivering your internet profits every month you don't know the difference." 9 settings. 15 minutes. $528/year back. Faster internet on a cheaper plan. The router was never broken. The defaults were designed by a company that makes money when you think it is.

• 
  🔥🏴‍☠️CyberTechWolf🏴‍☠️🔥 (@CyberTechWolff) reported 46 minutes ago

  @ChaoticTKnight I'm gonna laugh if Cloudflare gets DDOSed or some **** because they wanna claim the Internet. Not everyone has to be hijacked by Cloudflare and I hope someone hacks their ******* **** I don't use these words a lot but the Internet might just have robophobia.

• 
  Guillermo Zaandam 🇨🇦🇳🇱 (@besodemieterd) reported 47 minutes ago

  @Cloudflare Your CSP rules and WAF rules aren't working again and again. Please fix this

• 
  Dave Bettin (@dbettin) reported 50 minutes ago

  @dillon_mulroy @EffectTS_ Where is the cloudflare support?

• 
  Andy Knowles (@andykn101) reported 50 minutes ago

  @SportingNest @Cloudflare It's absolutely a scam. If you don't know what those instructions do listen to those who. If you're not clicking on the ads in your search results you've probably already got malware poisoning your search results. Get help.

• 
  JovanHuttonPulitzer™ אני לא סובל טיפשים! (@JovanHPulitzer) reported 51 minutes ago

  @FinalCutTile So as a tiler and tile installer, how do you think the Internet runs? I vote this the NEWEST stupid *** comment on X regarding #datecenters this week: The statement is false. Data centers are needed to run the internet as it actually exists today. What the internet actually is: The internet is a distributed “network of networks.” There is no single control room. Packets can theoretically move between two computers connected by routers and cables without a giant warehouse full of servers. Early ARPANET and 1980s–90s internet ran on university machines, government systems, and smaller server rooms. That part is true. That is not the internet anyone uses. What actually happens when you “use the internet” Almost everything people mean by the internet lives in data centers: Websites, search, email, social media Streaming (YouTube, Netflix, etc.) Cloud services (AWS, Azure, Google Cloud) Banking, payments, maps, messaging DNS, CDNs, and most of the content that gets delivered to your device Sources across Microsoft, Cloudflare, industry analyses, and infrastructure reporting describe data centers as the physical facilities that store, process, and serve that content. Without them, the pipes still exist, but the useful destinations do not. One infrastructure expert put it bluntly: the rudimentary network can still function, “you just can’t do anything that you’re used to on the internet because that’s all posted out of these datacenters.” Calling data centers the “backbone” or “brick-and-mortar of the digital world” is standard, not hype. The distinction that the statement ignores Network layer (cables, routers, ISPs, peering): can exist without hyperscale data centers. Application / content layer (everything you actually open, stream, search, or log into): overwhelmingly depends on them. The insult in the original statement does not change the facts. Data centers are required for the internet people actually use.

• 
  Sathz_நிலன் (@nilan_sathz) reported 51 minutes ago

  Stopping the bad guys with Cloudflare: 212 malicious requests blocked or challenged in the last month #cloudflare

• 
  Muntasir Mallik (@muntaxir04) reported 52 minutes ago

  @acolombiadev @Cloudflare Maybe it's Gemini? We never run out of searches bec nobody uses it anymore

• 
  AGTP (@AGTPinsights) reported 52 minutes ago

  Cursor just launched self-hosted machines for cloud agents today. Here's what you need to know. You can now run Cursor cloud agents on infrastructure you manage, including pools of machines that auto-scale with demand. The agent loop still runs in Cursor, but tool execution, your codebase, build outputs, and secrets stay entirely inside your own network. This gives agents access to internal services or specialized hardware that Cursor's own cloud can't reach. Cursor also added support for supported sandbox providers, letting teams pick managed infra instead of running their own machines. Self-hosted workers now support computer use on Linux and Mac too, so an agent can click, type, take screenshots, and drive the browser, with users able to watch or take control from Cursor. Coder separately announced Agent Relay, giving Cursor Cloud Agents self-hosted execution environments on customer infrastructure, launching in private preview with SpaceXAI as a design partner for regulated enterprises where code can't leave the network. Key numbers: - Supported sandbox partners at launch: AWS Lambda, Coder, Cloudflare, Daytona, E2B, Modal, Namespace, Vercel - Cursor Pro pricing starts at $20/month, self-hosted machines require paying for your own compute separately - Coder Agent Relay: private preview, no public pricing yet Team pools can hibernate idle machines so you're not paying to keep capacity warm.

• 
  Miget (@miget_com) reported 60 minutes ago

  Then you reach it from outside. Four terminators, and you can run more than one on the same network. WireGuard per person or per machine. Tailscale into a tailnet you already run. Cloudflare WARP into a Cloudflare org. IPsec site-to-site.

• 
  Yaz (@yazcaleb) reported 1 hour ago

  @tin_runje @Cloudflare won't let you down

• 
  R.Gion (Any/All) (@Gion_the_critic) reported 1 hour ago

  @junebnunny_ @SportingNest @Cloudflare I figure it's not legit, but I wanted to know where it was from since I've never seen this particular scam.