🔐 How are #GoogleAIStudio deployed apps on #CloudRun protecting the API key so it's not visible in the browser? (although it appeared as if Gemini was called from the client-side...)
➡️ A transparent proxy w/ Service Workers & WebSocket interception prevent key exposure!
【Google Cloud 】Private Service Connect を使ってVPC間の内部通信をする
https://qiita.com/aaaabbbbb/items/604140a1050a71c35aa6?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items
Cloud Runのアクセス制御を完全理解する（Ingress / IAM / Cloud Armor / IAP）
https://qiita.com/created_on_june_19/items/ed7d16eda2264ba690bb?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items
This week I had a bug where my API container was deploying without incident and crashing a second later when /health was checked. The issue was that the crypto provider was not initialized, which caused a panic as soon as /health was checked - it was an ordinary bug.
Because the container started, but then crashed on health check, and it was deployed to #Google #CloudRun, CloudRun kept restarting the container while I was asleep. And crashing it, hitting /health. Then Google's AI must have seen the word "crypto" in the logs because they turned my project off and I woke up to completely broken infrastructure, and an email that said I was CRYPTOMINING in violation of their FREE use agreement.
Now: (a) they charge me for every millisecond of compute time - this is not free use! and (b) crypto means CRYPTOGRAPHY to normal developers. Not freaking crypto currency mining, which I ain't doing.
So now my service is down and I'm waiting for an appeals process - it's been two days.
This was for a personal project but I can't continue to recommend GCP for organizational folks going forward. it is insane to shut down cloud resources on such a thin fact. This is a single 1-CPU CloudRun service that happens to be cycling due to a bug. Stop the API as failing to start properly, and send an email like a NORMAL PERSON. They could absolutely destroy a business this way.
And I think they also need to count false-positive shut downs as part of their uptime calcs. Because ... it is absolutely down time.
Ugh!
Why Serverless VPC Access Exists
https://dhvogel.com/blog/why-serverless-vpc-access-exists.html
Key question: Do you want your developers spending energy on Docker internals, or should a platform team handle this? Or just go with a FaaS framework?
Full tutorial: https://shinglyu.com/web/2025/10/08/serverless-rust-on-gcp-cloud-run-from-basic-deployment-to-optimized-container-builds.html
React × Hono × Docker × Cloud Runではじめて自前APIサーバーを構築した話
https://qiita.com/kikotkk/items/5d1e14d0ada3075f384f?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items
Guillaume Laforge
Qiita - 人気の記事
Return of Aadmaa
Dan
AI News
shinglyu