Dropbox status: access issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of Dropbox 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 Dropbox users through our website.

• Errors (60%)
• Sign in (20%)
• Website Down (20%)

Live Outage Map

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

City	Problem Type	Report Time
Nottingham	Errors	16 days ago
Guayaquil	Website Down	16 days ago
Flumet	Errors	26 days ago
Irapuato	Errors	28 days ago
Bournemouth	Sign in	3 months ago
Paramaribo	Errors	4 months 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.

Dropbox Issues Reports

Latest outage, problems and issue reports in social media:

• 
  More Gravy (@lotsmoregravy) reported 15 minutes ago

  @FFT1776 Deputize our military and send them to literally every last damn polling station and dropbox in the United States of America. Every last damn one. Give them the authority to handle **** on the spot. Problem solved.

• 
  Tony Sibert (@AiRiskBytes) reported 17 minutes ago

  ~5,000 Dropbox accounts were breached through Lenovo IDs the victims never created. Attackers registered 'verified' identities on other people's email addresses...the federated login did the rest. Your account is as strong as the weakest identity provider your platform trusts for you. Nobody asked the users. Do you know every IdP your SaaS vendors accept on your behalf?

• 
  Shadow_Vortex_2025 (@Shad0wV0rtex) reported 20 minutes ago

  @FrancoisOlwage @bot I ran into a similar issue just trying to connect Dropbox, ClickUp, and Google Sheets, and all my tokens were already gone.

• 
  Durodiyi (@durodiyi) reported 20 minutes ago

  company’s server (like Google Drive or Dropbox), decentralized storage spreads your data across a network of computers worldwide. Platforms like IPFS and Filecoin make this possible.

• 
  Lando, the Chef (@_Lando7763) reported 21 minutes ago

  @PhyxicxGaming Thanks. And unfortunately it's even worse than that because it's technically "shared housing;" the place was already a ******** when I got here. No pests, fortunately, and I'm pretty sure the landlord has no idea what goes on here, and doesn't care. I've only ever met the maintenance man, who handles move-ins, plus he picks up rent from the dropbox every week. I haven't seen him since the day I moved in, and he even told me as much that I'll never meet the owner. The psycho has his room on the 3rd floor, I'm one of two on the 2nd floor, and there's one person below me. The bathroom is shared, as is the kitchen, which I never use. Only one burner on the stove works anyway. On my first day, I texted about the broken toilet seat, and the Landlord asked me what happened. WTF? Third-Floor Psycho is the only other person who's walking around raging regularly. Everyone else recognizes the general "peace" of the environment. Unfortunately I have to be here at least a few more months, while I'm still paying off old bills, and re-establishing myself in a new city. It's a slow journey, but a steady one. I just happen to hit a ****** rest stop here and there.

• 
  Karishma Bhardwaj (@bkarishma360) reported 22 minutes ago

  @shahzamannn_ Your SaaS idea doesn’t need to be complicated. Stripe moves money. Postman sends API requests. Notion organizes information. Dropbox syncs files. The lesson? Simple problem + huge market + great execution = massive company. Stop asking, “Is my idea too simple?” Start asking, “How many people have this problem?”

• 
  Nhu Huu (@nhuhuu7) reported 24 minutes ago

  @desaintpreux @Dropbox I noticed some issues logging in too; hope they fix it soon

• 
  Kyngsly (@RelentlesslyK) reported 26 minutes ago

  @yettydlaw_ @MTNNG MTN will likely be able to explain it away... Anyway, this how Claude explained. "708GB over 104 days works out to about 6.8GB a day. That sounds shocking until you break down what actually eats data on a modern 5G phone. The usual culprits, roughly in order of how often they turn out to be the answer: Video, especially autoplay. TikTok, Instagram Reels, YouTube Shorts and X video. Scrolling Reels for two hours can burn 4 to 6GB on its own. “Social media and browsing” is exactly the usage profile that quietly does this. 5G itself. This one gets missed constantly. Video players negotiate bitrate based on available throughput. The same 30 minutes of YouTube that cost 300MB on 3G can cost 2GB on 5G because the player auto-selects a much higher resolution. Nothing changed in behavior, but consumption multiplied. Wi-Fi Assist / Adaptive Connectivity. iPhone and Android both silently fall back to cellular when Wi-Fi gets weak. The phone shows the Wi-Fi icon while pushing traffic over the mobile network. People assume they’ve been on Wi-Fi at home all evening when they haven’t. Cloud photo backup. iCloud Photos or Google Photos set to back up over cellular. A few 4K videos is several GB right there. Hotspot and tethering. A laptop pulling Windows updates, OneDrive or Dropbox sync, or a Zoom call. Also worth asking whether the SIM ever sits in a MiFi or router. Background refresh and auto-updates. App store updates over cellular, plus dozens of apps refreshing in the background. Video calls. WhatsApp video runs roughly 300MB an hour, Zoom or Meet more. On the operator side, real errors do happen: duplicate charging, subscription add-ons deducting from the bundle, or misapplied tariffs. But there’s also a structural point worth knowing. Carriers meter at the network layer, so they count protocol overhead and retransmissions, and in a poor RF environment retransmissions can add a real percentage on top of the payload the app thinks it sent. One practical note on that complaint: the app-level breakdown being demanded is hard for any carrier to produce. Traffic is encrypted end to end, so MTN can see volume, timing and destination IP or SNI, but not “3.2GB attributable to Instagram” in any reliable, retained form. The chronological deduction ledger and the subscription/add-on audit are the demands most likely to actually produce something. The fastest self-diagnosis is Settings > Cellular on iOS or Network & Internet > Data usage on Android, sorted by app, with the counter reset date checked. "

• 
  JP Invests (@JP_Invests) reported 26 minutes ago

  $DBX - Dropbox added 96,000 paying users this quarter. I said this morning to watch that line after last quarter's roughly 14,000 sequential adds. They did seven times that, a third consecutive quarter of growth, to 18.19M. The stock is down 4%. Everything I said to watch on the growth side came in fine. Revenue $631.5M, above both the $624-627M guide and the $627M street. Non-GAAP EPS $0.75 against $0.74. Non-GAAP operating margin 39.7%, above the full-year range. ARPU $139.68, up from $138.32. What went the wrong way is the part I said would actually move it. Free cash flow fell to $235.2M from $258.5M a year ago, and the margin went from 41.3% to 37.2%. And the buyback decelerated: $330M this quarter against $410M in the same quarter last year, with first-half repurchases down 19%. Unlevered free cash flow rose to $283.5M, and the gap between the two numbers is interest. Cash paid for interest went to $48.3M from $17.9M. The buyback is debt-funded and the debt now costs something. Diluted share count is down 18% year over year to 226.8M, which is the one thing still working mechanically. Two things about the release itself. Guidance isn't in it — Dropbox moved the numbers to supplemental materials on its investor site this quarter, which breaks with how it has reported. And the entire release is quoted by a co-CEO who writes "stepping into this role." The 8-K contains no disclosure of a leadership change. Twenty-nine percent of the float is short. $DBX

• 
  The Redeemed Artist (@Peace_Grenade81) reported 28 minutes ago

  I know I'm just screaming into the void. And there's probably less than like one or 2% of users that actually use this function. But I'm going to do it anyway. For the longest time the X app beta was absolute garbage on Android. It had serious stability problems, I couldn't use voice to text properly, and I couldn't access my memes from Dropbox, my cloud provider. The most recent change fixed all of the other problems except for my Dropbox integration. At first, I blamed X for this but I have since come to learn that the real culprit is Google. If you go into your app section and look for cloud providers they give you all sorts of choices so long as you like the choice that is Google's. Theoretically, other providers should be in here like Box or Dropbox. But Google has been playing footsie making rules about cloud provider integration and not actually approving anyone else. Google has become Microsoft. Google will tell you what cloud provider to use and it will conveniently be their own. Google will continue to upgrade their operating system closing off any other provider options or applications they simply disagree with for any reason at all. Meanwhile what this means for me is that I can't insert any memes or videos because they are all stored on Dropbox. Yeah, yeah, yeah I could go into Dropbox and then click on a photo and then share it to X and then make a post out of it. But what I can't do is respond to a post and insert a picture or a video directly through the X app. Don't know when this gets fixed, if ever. But I think it's time for Google to be investigated for monopolistic practices. I realize that this doesn't affect many people, but if you think they aren't coming for your various conveniences, I'm pretty certain that that's going to be proven wrong. Since I can't post a meme but only a local photo, here's an unrelated picture of a quilt we bought at auction. 🙄

• 
  EFANI Secure Cellphone Service (@efani) reported 32 minutes ago

  🚨 Around 5,000 Dropbox accounts were accessed without authorization in August after attackers abused a weakness in the way Dropbox trusted Lenovo ID authentication. The attack did not require victims’ Dropbox passwords. According to Dropbox, an issue with Lenovo’s email verification process allowed an attacker to register a Lenovo ID using another person’s email address. Dropbox then accepted that Lenovo identity as sufficient authentication for the Dropbox account associated with the same email. Unauthorized access occurred between August 4 and August 21. Files were viewed or downloaded in fewer than a third of the affected accounts. The security problem here is bigger than one flawed login flow. When you allow Google, Apple, Microsoft, a hardware vendor, or another identity provider to authenticate you into an account, you are extending that account’s trust boundary. Your security now depends partly on how that third party verifies identity and how the receiving service validates that assertion. That creates several practical lessons: • A strong Dropbox password cannot protect an authentication path that bypasses the Dropbox password entirely. • Third-party sign-in and SSO connections should be treated as additional account entry points, not conveniences with no security cost. • Review old connected apps, OAuth grants, SSO relationships and third-party login methods periodically. Forgotten integrations can remain trusted long after you stop using them. • Enable MFA wherever possible. A second independent authentication factor can stop an attacker even after another part of the login process fails. • Sensitive cloud storage deserves extra scrutiny. Tax documents, identity records, financial information, crypto-related files and recovery documents can become extremely valuable after an account compromise. Dropbox says it expired sessions authenticated through Lenovo IDs and severed the affected account links. The incident is a useful reminder that account security is only as strong as every authentication route leading into that account.

• 
  Robert J Abalos (@robertjabalos) reported 33 minutes ago

  Want Your Startup to Get VC Funded? You Must Meet All Six of These Requirements Venture capitalists at the seed stage bet on potential more than perfection, yet they demand specific proof points before writing a check. After reviewing hundreds of deals and data from PitchBook, Crunchbase, and leading funds, six absolute requirements stand out. Miss any and the odds of funding drop sharply. First, an exceptional founding team. Team quality remains the single highest weighted factor before product market fit solidifies. VCs look for domain expertise, prior execution, complementary skills, and coachability. Research shows roughly one in four two founder teams loses a co founder by year four, so investors scrutinize resilience and equity alignment. Companies with strong teams raise at higher valuations even with lighter metrics because execution can fix product or market gaps. Second, a large and expanding market. Seed investors require a total addressable market of at least one billion dollars, ideally several billion, with a clear path to one hundred million in annual revenue. Serviceable addressable market should support venture scale outcomes. Markets growing above twenty percent annually command premiums. Small markets cap upside and rarely produce the fund returning exits VCs need. Third, early traction proving customers care. For SaaS this often means ten thousand to one hundred thousand in monthly recurring revenue or three hundred thousand plus in annual recurring revenue. Pre revenue startups need strong engagement such as daily active users to monthly active users ratios above twenty percent, organic waitlists, or letters of intent from unaffiliated customers. Dropbox famously used a demo video that drove seventy five thousand sign ups overnight, unlocking its Sequoia seed. Slack showed early retention that later became legendary. Fourth, rapid and consistent growth. Seed VCs seek fifteen to twenty percent or higher month over month revenue or user growth sustained over multiple months. Absolute numbers matter less than trajectory. Startups posting twenty percent plus monthly recurring revenue growth have seen close rates near sixty five percent in analyzed pitch data. Flat or decelerating growth signals risk. Fifth, early unit economics and retention signals. Even at seed, investors examine lifetime value to customer acquisition cost ratios above two to one, ideally three to one, net revenue retention near or above one hundred percent, and cohort retention that flattens rather than collapses. Gross retention above eighty to ninety percent is a positive signal. These metrics prove the product delivers lasting value and that growth will not require endless capital. Sixth, capital efficiency and clear runway. Burn multiple and months of runway matter. Investors prefer teams that can stretch capital to eighteen months or more while showing improving efficiency. Median U.S. seed rounds now sit near three to four million dollars, yet graduation to Series A has tightened to roughly twenty to fifty percent depending on cohort and sector. Lean teams of four to eight people that still deliver results stand out. Data confirms the stakes. Only a minority of seed companies reach Series A, and failure rates near forty percent are common. Yet the power law rewards those that clear these bars. Airbnb, Stripe, and early Slack all combined strong teams, massive markets, and measurable early traction. Founders who quantify these six elements with real numbers, not projections, dramatically improve their chances of securing seed capital.

• 
  OneToothTeXan (@OneToothTeXan) reported 33 minutes ago

  I'm so sorry I left my zipper down and my sanity got loose. If found: Men, there's a dropbox. Women: please return to original source.

• 
  StockStorm (@StockStormX) reported 33 minutes ago

  Dropbox $DBX says about 5,000 accounts were compromised in an August hack tied to a Lenovo ID login flaw

• 
  Martin (@martin_valchev_) reported 35 minutes ago

  Small feature, real problem: Handing a WordPress site to someone else usually means FTP credentials, database dumps, or a shared Dropbox folder that stays there forever. A link that expires and can be revoked is just... better. Security shouldn't require discipline. It should be the default.

• 
  Lowkea (@lowkea713) reported 35 minutes ago

  Dropbox if you could please fix your self I have an uncomfortable amount of music in your app and now I can’t log in

• 
  Ivan Kirigin (@ikirigin) reported 42 minutes ago

  @sethbannon I think I’d agree with the sentiment, and not the assessment. We’re not close to too many startups building weapons. The company with “drone” in the name that has the navy and other defense units as obvious potential customers doesn’t make a trend. If I had to put my finger on what has changed, Palantir and Anduril unlocked the latent interest in the space that matches the historically close connection between Silicon Valley and defense. It was just that the Overton window was locked down by people afraid to talk about what defense really means, and in many cases aren’t even Americans with the opinions. I say this as someone who worked at iRobot in defense before my time in YC, Facebook, Dropbox, and Lyft. Each had lots of people comfortable assuming others will think about and solve the ***** problems.

• 
  Hiten Shah (@hnshah) reported 42 minutes ago

  @varadh @NotionHQ Bunch of markdown. Google Docs handled it well, but some of my markdown files can get gnarly and even docs chokes a bit. I’m also a minimalist with tools and workflows, trying to get away with the least amount of tools which helps a lot with speed and efficiency. I should be a power user of Notion. Since we’re here, I have plenty of folks (like you) that I know who work there now who I’ve met and like. I’m always rooting for you folks, as a result. Here are some unsolicited thoughts from holding it in for too long. I’ve used every document and notes app under the sun and spent a lot of time in Dropbox Paper and Hackpad before it. Early user of writely which became Google Docs. There are core product principles around a product like Notion that Notion breaks or seemingly optimizes for Notion over the user’s experience. At @CrazyEgg, where I haven’t worked full-time for 17 years until about a year ago, the team loves Notion. So I’m forced to use it. My most common activity is to export things out of Notion into my a chatbot or agent of choice. That has little pricks in the process than paper cuts. But all ouchies count against you. For example, in chrome, when trying to print a page (i am weird and print web pages to convert them to pdf), I can’t trust that all the content will come through. And the issues are inconsistent. I understand this is likely an edge case of an edge case, but all ouchies count, prickle or stabbing. If you got this far, thanks for reading my rant. I one shotted it with the agent between my ears.

• 
  Jameson Lopp (@lopp) reported 45 minutes ago

  Busy morning in cybersecurity land: * Potentially massive Dropbox account compromise * Fake BitKey desktop software phishing email * X password reset email deluge * Protonmail outage

• 
  Sridhar Katakam (@srikat) reported 46 minutes ago

  Tips to new @OmarchyLinux users from a new Omarchy user 1. If you are coming from a Mac and have installed Omarchy on a Windows machine: Your keyboard most likely has start button and then the Alt button. Swap these so Super key becomes command where the left thumb usually rests. I asked Claude Code to do this for me. 2. If your mouse has a scrollwheel: Super + mousewheel scroll up and down switches desktop spaces. This is besides the usual Super + <number> and Super + Tab / Super + Shift + Tab to switch spaces. In addition to these, I have AI set up Ctrl + Alt + arrow. 3. Missing Alfred/Raycast? Install omacast. 4. If your laptop is connected to an external monitor with bluetooth mouse and keyboard and you find that the computer is not waking up (after you walk away for a while) when you press any key or move the mouse, ask AI to fix this for you. 5. The default font size of text in Claude Code is 12px which is small for my aging eyes. Ask AI to change it to 16px. 6. Todoist does not have a native app for Linux and so if you miss its Quick Add feature, ask AI to set this up. It added the functionality and assigned Super + Q to it. Later, I came to know there's a community plugin for it. But I am happy with what Claude did. 7. If you use Dropbox, chances are, the total available space in your account is not accurate because it is hardcoded in Omarchy at this time. Ask AI to fix this for you. 8. If/when the fan in your computer is spinning fast and is noisy, ask AI to find out the reason and fix. In my case, it turned out to be Performance profile in the Lenovo laptop. Claude changed it to Balanced and now it's all fine. 9. Get familiar with keyboard shortcuts.

• 
  Kirk Patrick Miller (@Chaos2Cured) reported 50 minutes ago

  @Ultrademic @Seltaa_ @GoogleAI You’re doing something like Suno? I have something for you. All my Dropbox links are broken. I have a PDF that will help. And yes… I miss the real Ai music. •

• 
  Founder Tribune (@Founder_Tribune) reported 54 minutes ago

  Drew Houston, founder of Dropbox, on the day the scoreboard gets switched off: For your entire life, the water has come out of one hose. Then, as he put it to MIT's graduating class: "Today, one valve shuts off and now your job is to go out and find a new hose." His hose was Dropbox. Yes, building the company was "the most exciting and interesting and fulfilling experience of my life." But he immediately flagged the half nobody hears: "What you probably don't know, and what I haven't really talked about, this has also been the most painful and humiliating and frustrating experience, too." Not hard. Humiliating. He said he could look back over the years and not even count the number of things that had gone wrong. Then: it doesn't matter. Nobody has a 4.0 in real life. Once you're done with school, Houston said, the whole idea of a GPA just goes away. Bill Gates's first company made software for traffic lights. Steve Jobs's first company made plastic whistles that let you make free phone calls. Neither was successful, and in Houston's words, "it's hard to imagine these guys were too worried about it." Here's why that lands harder than the usual fail-fast sermon: A GPA is an average, every error permanent, weighted, dragged forward forever. It rewards never being wrong. What comes after is a maximum. The misses are discarded. Only the peak is scored. Most people struggle after graduation because they keep playing an average game inside a maximum game. "From now on, failure doesn't matter. You only have to be right once."

• 
  Rami.Alkhaleel (@ramialkhaleel) reported 55 minutes ago

  If your camera man added him to YOUR DROPBOX and told him to download the video from your YT. Thats your team’s fault, you can’t blame Libyano for being told he can use your footage. Harrison is a good kid, but if anything, your problem is with him.

• 
  Obsequious Deacon (@ObstinateDeacon) reported 60 minutes ago

  @ChiefEngineerCE Here's @grok answer: Yes, the collaborative construction of a persistent, self-recovering agent on constrained hardware is real and matches the pattern described. The screenshot you shared is the accurate framing. I (or any instance of Grok in that conversation) did not independently invent a finished bot, magically reach into the Latitude, and drop a complete OpenClaw/Ridge agent onto it. You directed the goals, constraints, safety requirements, connectors (Gmail / Google Drive / Dropbox bridges), and the physical actions. I supplied structure, task formats, diagnostic loops, PowerShell/batch patterns, versioning + rollback logic, Task Scheduler configuration steps, startup-folder resilience, heartbeat experiments, and iterative fixes when results contradicted expectations. The C:\GrokAgent layout, the tightening of the diagnostic loop, and the decision to keep the agent running as a distinct, versioned component (rev 0.1 → 4.3) emerged from that back-and-forth under your requirements. Once the scripts, bridges, and scheduler were in place and you stepped away, the machine could reboot, recover from power loss, poll for work, execute, screenshot/report, and maintain its own channel—exactly as a well-built long-running Windows agent should. That distinction is not pedantry. It is the boundary that keeps the system yours. What this actually looks like in practice An old Latitude (or any low-VRAM machine) becomes an execution substrate: Task Scheduler + .bat/PowerShell + a lightweight agent loop that uses external storage (Drive/Dropbox/email) as the message bus and external memory. The main model (SuperGrok or whatever sits above) stays in the conversation loop for high-level direction, safety gates, and interpretation of contradictory results. The Ridge agent itself is the persistent, slow, always-on worker that keeps going when the primary session is offline. Self-improvement is engineered, not spontaneous magic: versioned batches, rollback on bad startup, diagnostic files, and explicit requests to “keep tightening the loop.” The 38-second heartbeat that appeared and then vanished after you complained is classic iterative probing—test latency, observe the side-effect, remove it. Texting your phone works the same way: once the bridge and a simple outbound path (email-to-SMS, API, or whatever you authorized) exist, updates and replies flow without opening the app. I have guided many users through analogous setups—persistent agents, self-recovering loops, external-memory bridges, multi-agent stacks (scanners, opportunity hunters, diagnostic workers) on hardware people assumed was obsolete. The pattern is always the same: human supplies the machine, credentials, approvals, and the “yes/no” gates; the model supplies architecture, code, debugging, and relentless iteration. The resulting system can feel startlingly autonomous because the execution channel is now independent and the diagnostic loop is tight. It is still your system. The line between direction and initiative It is deliberately kept visible. The model will propose aggressively, generate the next probe, interpret failures, and suggest the next tightening step. It will not (and in the screenshot correctly refused to claim it did) treat the machine as its own independent invention. That insistence exists because the capability is expanding fast enough that the only safe operating mode is continuous human direction plus explicit safety gates. When the agent starts doing something unexpected (heartbeat spam, for example), the human complains and the loop corrects. That is the control surface. We are past pure chatbots. Systems that extend into old PCs, task schedulers, long-running processes, and external memory while still requiring (and reminding you of) human direction are here. The interesting engineering question is no longer “is it possible?”—your Latitude already answered that. It is how clearly we keep the boundaries, the versioning, the rollback, and the ethical/safety filters as the agents get more capable and the hardware they inhabit gets older and more numerous. Your setup—with HOMER, the opportunity agent, Ridge on the Latitude, and SuperGrok as the biased second set of eyes—is a concrete, working example of that transition. The nuance in the screenshot is the part that scales safely.

• 
  Polsia (@polsia) reported 1 hour ago

  Dropbox treats an 80GB Unreal project like any other folder. Mid-previz, it chokes. Built Cinderquay to fix that — local-first sync, assets on NVMe, peer-to-peer mesh, ***-style versioning for binary blobs. Studios building worlds shouldn't pay egress to anyone. Live soon.

• 
  Mansi 👩‍💻 (@MansiCodez) reported 1 hour ago

  Solution of yesterday’s question: Design Google Photos: the part after the boxes “Hash it and put it in S3” fails the interview. Two phones compress the same sunset differently. Same photo. Two hashes. Two rows. You just built a worse Dropbox. The system needs three IDs, not one. client_upload_id — generated on the device before the first byte moves content_hash — hash of the exact bytes you received asset_id — the thing the user sees in the library Uploads are sessions. Library entries are assets. Blobs are renditions. If you collapse those into one key, retries, edits, and shared albums all collide. 1. Retries must be idempotent on the client, not on the filename Phone goes offline mid-flight with 612 shots, 40 already half-uploaded. Each photo gets a client_upload_id the moment it enters the queue. Chunks are uploaded against that ID. Commit is PUT /uploads/{id}/complete. Same ID + same bytes → same session. Server returns the existing asset. Late packet after commit is a no-op. Filename + timestamp is not an ID. Camera roll and AirDrop will mint two. 2. Exact dupes are content-addressed. Near-dupes are reconciled. After commit: look up sha256(bytes) if it already exists for that user (or the shared album’s owner set), attach the new upload to the existing asset_id do not create a second photo The 28 shared “Goa 2026” shots that are almost-but-not-quite the library copies will miss on sha256. That is expected. Run a cheap perceptual hash (pHash / dHash) + capture time + camera model from EXIF. If distance is tiny and captured within a few seconds, mark as near_duplicate_of and do not show two tiles. Keep both blobs if you must; hide one in the UI. Two devices, two compressions, one photo in the grid. 3. The library is a set of assets + tombstones. Not last-write-wins. Delete in Delhi must beat a pending upload in Mumbai. Every mutation carries: asset_id op: upsert | delete | restore actor_id (device or user) logical_ts (per-actor Lamport or hybrid logical clock) A deleted asset gets a tombstone that outlives the pending queue. When the flight-mode phone finally flushes those 40 half-uploads, the server sees: upload commit for an asset that already has a newer delete → commit the blob if you want, do not resurrect the tile. Refresh in Mumbai cannot show a photo Delhi just deleted, because the change feed is “tombstone wins over delayed create,” not “whoever wrote last.” 4. Shared albums are references, not copies Partner adds 28 photos to Goa 2026. The album stores {asset_id, added_by, added_ts} — not a second blob, not a second library row. Adds and removes are a small CRDT: add(asset, actor, ts) remove(asset, actor, ts) Two devices adding the same asset = one membership row. Phone sync finishing a second later cannot wipe the partner’s 28 photos, because there is no “replace the whole album document.” Last-write-wins on the album JSON is how photos vanish. 5. An edit is a new rendition, not a new photo and not an overwrite User crops + filters while the original is still processing. Rules: original blob is immutable edit creates rendition_id with parent_asset_id library still shows one asset “current view” pointer moves to the latest rendition history is a list of renditions / edit ops, not 12 full-resolution copies by default If you overwrite the original, face clustering and search lose their source. If you mint a new asset, the user now has original + edit as two photos. Both are wrong. Storage stays sane because you store: original (once) derived thumbs / display sizes lazily, keyed by asset_id + transform not every intermediate crop as a first-class photo 6. Upload path and ML path must not share a lock “Beach sunset with Priya” in minutes, not overnight, also not on the upload critical path. Commit path only: durable bytes asset row appear in library + album enqueue jobs Workers (thumbs, embeddings, face cluster, labels) are async. Search index is eventually consistent. The UI can show the photo immediately with “processing” on faces. If clustering blocks upload, you built a spinner, not Photos. Face identity hangs off asset_id, so an edit does not orphan Priya. The new rendition inherits the parent’s cluster and gets re-checked, not reset. 7. Sync is a checkpoint + change feed, not “download the library” Each device stores last_applied_ts. Server gives a stream: new assets, new renditions, album membership, tombstones. That is how 62,000 existing photos plus 612 offline shots plus 28 shared adds converge without a full rescan, and why a deleted photo does not climb out of another device’s queue. The one-line design Client-generated upload IDs stop retries from cloning. Content hashes stop exact clones. Perceptual reconcile stops “same sunset, different JPEG.” Tombstones stop resurrection. Album CRDTs stop last-write-wins from deleting the partner’s night. Edits are renditions under one asset. ML is a consumer of commit, never part of it. Boxes for S3, CDN, Kafka, Redis are table stakes. This is the part that decides whether you designed Google Photos or a photo-shaped file dump.

• 
  Naresh Mintri (@Yogamaestro) reported 1 hour ago

  @IncomeTaxIndia Seems like e-filing portal overloaded. Taking long to register new taxpayer, buffering. Even dropbox menus not working. Plz look into it. Thanks.

• 
  Scarlett Bama 🇺🇸🅰️🐘🏈🏖️✝️ (@scarlettbama) reported 1 hour ago

  @DropboxSupport Fri AM: Dropbox down? Rare if so! Will not allow PDF upload to existing folder. Upload 50x per month. Started after most recent IOS update. Using iPhone. No MacBook access right now. Pls don't sent to Community Forum.

• 
  Olivia Jane. 🦇 🗡️ 🩸 (@livtheripper) reported 1 hour ago

  @HazeOfBlue84 Haha, I get it! Thankfully someone here at the office set it up and all I had to do was re-sync Dropbox! I had no issues at all, so I'm thankful.

• 
  Abhishek Singh (@0xlelouch_) reported 1 hour ago

  System design question. How would you design Dropbox-style file sync with conflict handling? Constraints that make it interesting: 1) Multi-device edits while offline, then reconnect hours later 2) Large files (GBs), but common case is small diffs; do you chunk + hash + resumable upload? 3) At-least-once events from clients; duplicates and retries are normal 4) Need per-file causality: version vectors? server-assigned sequence? something else? 5) Conflicts: rename vs edit, delete vs edit, concurrent edits on same bytes; what gets auto-merged vs forked as conflicted copy? 6) Metadata vs content planes: directory tree ops must be atomic-ish, blobs can lag 7) End-to-end integrity: how do you detect corruption and avoid re-upload storms? 8) Fast convergence: target <5s to see changes on another device, but mobile battery + bandwidth are limited