If you happen to be at KubeCon EU in Amsterdam, come stop by booth #882 and catch a demo of Depot CI in action.
See you there!
If you happen to be at KubeCon EU in Amsterdam, come stop by booth #882 and catch a demo of Depot CI in action.
See you there!
when it fails at minute 42 on the final step
iykyk
docs.google.com/forms/d/e/1F...
We're gearing up to launch Depot CI:
A faster, more reliable CI provider built for how software is being built today.
Depot CI waitlist in 🧵
🚀 Sherlock just got a major upgrade. It can now analyze your builds and GitHub Actions jobs — read logs, identify errors, and suggest fixes. It can also break down your CI analytics and open pre-filled support tickets. Available now for all Depot users.
depot.dev/blog/teachin...
depot.dev/blog/lambda-...
durable lambda development
is this san francisco traffic or modern code trying to get through decade old infra 🤷♂️
depot.dev/blog/lambda-...
New post: Lambda durable functions — real implementation notes in 🧵
watching your GitHub Actions queue before switching to Depot
weekly reminder, we saved developers over 239,280 hours last week.
Wild to think this number has gone up almost ~60K hours avg per week since this weekly update started a few months ago.
your CI is now the slowest developer on your team
big brain energy
this is the equivalent of giving a formula 1 driver a cozy coup and expecting him to win... impossible
depot.dev/blog/how-simulations-reduce-guesswork
We had a standby pool algorithm with 12+ tunable parameters and no safe way to test changes without risking customer latency.
So we built a simulator with real job data, ran hyperparameter optimization on it, and found a config that's both faster and cheaper.
New post 👇
Just your weekly reminder that we saved developers 282,650 hours last week.
depot.dev/blog/differe...
"just use QEMU microvm"
depot.dev/changelog/20...
You don't need to see 400 lines of "Setting up Node.js" every time you look at a failed job.
GitHub Actions job logs in Depot are now collapsible per step. Click the caret, fold the noise, find the problem.
Full breakdown on our blog: depot.dev/blog/differe...
Hotplugging:
cloud-hypervisor: add vCPUs, resize memory, attach PCI devices on a running VM.
QEMU microvm: not supported at all.
Snapshotting & live migration:
Both support it. Both have caveats. Neither guarantees cross-version compatibility. cloud-hypervisor does it via API. QEMU needs qcow2 disks or external file dumps.
GPU passthrough:
cloud-hypervisor: yes, via VFIO.
QEMU microvm: no PCI support = no GPU passthrough out of the box. (PCIe is configurable since QEMU 5.2, but it's not the default path.)
cloud-hypervisor gives you a REST API over a UNIX socket. Clean, modern.
QEMU microvm gives you QMP (JSON over socket) or libvirt (XML config, custom RPC format, microvm compatibility varies by distro). Functional but rougher.
QEMU microvm vs cloud-hypervisor — which one do you pick for your platform?
We compared the two across 5 dimensions that matter when you're building on top of microVMs. Quick rundown: