Fog-Bank keeps every frame of the world's oldest webcam feed

What the project does

FogCam, described in a Hacker News post as the oldest webcam feed still running, serves its images in a way that leaves no history behind: one picture is published at a time and replaced the moment the next arrives. A project called Fog-Bank has taken on the job of keeping all of it. According to the write-up on fog-bank.org, the archive catches every frame the camera produces and, just as deliberately, records every failure along the way — when a gap opened, how long it lasted, and whether the blame belongs to the archive's own infrastructure or to FogCam itself.

Four collectors, three networks

The ingestion side is built around independence rather than a single hardened poller. Four collectors run across three separate networks, each asking FogCam for its current image on its own schedule: a poller on the operator's own server every 7 seconds, a component called fogline on Cloudflare every 15 seconds that also maintains a 30-day buffer, a Cloudflare worker every 20 seconds that writes directly into permanent storage, and a machine at the operator's home, again every 15 seconds.

Every frame each collector receives is pushed into the archive. Because pushes can fail, reconciliation also runs in the other direction: a sweep every five minutes and a nightly job both pull back anything a push dropped, so a missed frame can be recovered from one of the other collectors rather than lost for good.

Cache-busting and frame handling

The requests themselves are tuned against a real problem the operator hit in August: FogCam's web server was serving cached copies that were minutes old. Each request now carries a unique query string plus no-cache headers to force the origin to hand over the genuinely current file.

Once caught, a frame gets a defined treatment. It is named using FogCam's own upload time, fingerprinted with SHA-256, and annotated without recompressing the image data. It is then stored as lossless JPEG XL and copied to Cloudflare R2 within the same minute it was captured.

A public ledger of outages

A status board probes every stage of the pipeline, from the camera itself through to the feed's YouTube output, every few minutes, and sends the operator a Telegram message the moment any stage stops.

The project also publishes its outage history in order, with each incident labelled as belonging either to FogCam or to the archive. The entries so far include a FogCam server outage on August 13, a 64-hour freeze on FogCam's side from September 8 to 10, and — inside that same freeze — a 40-hour outage of the operator's own server caused by a full disk.

Why it matters

Much of the web is ephemeral by design, and a webcam that shows only its latest frame is an extreme case: without a third party stepping in, the feed's entire history evaporates frame by frame. Fog-Bank is a small, concrete instance of that intervention, applied to a piece of internet history old enough to be worth keeping.

The design is also a transferable template for anyone archiving an unreliable or uncooperative source. Its core moves — multiple independent collectors on different networks, push plus periodic pull reconciliation, cache-busting request headers, content-addressed files via SHA-256 hashing, lossless transcoding, and immediate off-site copies — are precisely the patterns a robust ingestion pipeline needs, and each one is justified in the write-up by an actual failure the operator ran into.

The outage ledger is the most distinctive part. When you depend on someone else's service, the pressing question after any gap is which side broke. By instrumenting both ends and publishing the attribution, the project turns its own reliability into a verifiable public record — good for the archive's credibility, and a neat piece of practical systems writing in its own right.