Would love to hear what worked (or didn’t) for folks here. Did you build it from scratch or stitch together existing tools? Was usage organic or did you have to push it hard internally? Any lessons on what made people actually use it?
Also open to stories where it didn’t work out. Did you end up switching to an off-the-shelf option later?
1. setup a platform using jupyterhub on kubernetes * use case leans towards research and mixed research-dev workflows * user only requires browser installed locally * jupyterhub server selection menu is configured such that * user selects container image to spawn from: these are usually built per project tool chain * user selects node type to run on: for example if user needs gpu resources, then can schedule onto cluster node with gpu devices attached 2. setup a platform that integrates monorepo with vscode devcontainer * use case focuses on developers * requires vscode, docker installed locally * monorepo contains a collection of subprojects for running services and jobs * each subproject define its own devcontainer settings * these follow common patterns per runtime and runtime version needed for project * devcontainer images are pre-built and maintained in onprem container registry
also appreciate the note on the sister team. i’ve seen the same. lots of energy up front but adoption fizzles when teams don’t prioritize the boring stuff like docs and onboarding
mmarian•9mo ago
DeborahEmeni_•9mo ago