Thanks, Tanishq. Feel free to add related comments to the PR https://github.com/apache/texera/pull/5961 .
On Sat, Jul 4, 2026 at 11:26 AM Tanishq Prashant Gandhi <[email protected]> wrote: > Hi Chen, Yicong, and all, > > Sorry for the slow reply, this email slipped past me for a week. Catching > up now. > > Yicong, thanks for bin/local-dev.sh. I haven't tried it yet, but from the > description it sounds like a lighter approach than mine. > > What I can share is how I got the Kubernetes setup running on my 8 GB > laptop, in case it's useful for anyone testing the Kubernetes-specific > behavior. Everything runs inside a mini Kubernetes cluster (minikube), > which itself uses a good chunk of memory, so the trick was fitting the rest > into what's left. Here's what worked: > > Give Docker enough memory before creating the cluster. In Docker Desktop → > Settings → Resources, set Memory to 7 GB. Then: > minikube start --memory=6000 --cpus=4 --driver=docker > > Turn off the services you don't need. I put this in a small values override > file and passed it with -f: > agentService: > enabled: false # AI assistant > litellm: > enabled: false # AI model router > exampleDataLoader: > enabled: false # sample-data loader > texeraImages: > pullPolicy: IfNotPresent > > Give the install extra time: > helm install texera k8s --namespace texera-dev --create-namespace \ > -f k8s/values-development.yaml -f k8s/values-dev-trimmed.yaml --timeout > 20m > > > With the AI services off, the whole thing fits in about 6 GB, tight, but > stable. > > The reason I went the Kubernetes route is that it mirrors how Texera > actually runs in production, so it lets me test my changes in a real > deployment setup. > > Best, > Tanishq > > On Fri, Jul 3, 2026 at 6:50 PM Chen Li <[email protected]> wrote: > > > Thanks, Yicong. To finish this topic, let's use the tool in > > > > > https://urldefense.com/v3/__https://github.com/apache/texera/pull/5961to__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Mr3uz1cItQMYuQb489hbDyIxoTzsbU_m_HgLAr8-TtJg3XuT_oja4at4zks9Cf9c9dwEzegP3nBrg3A$ > > help developers manage the > > micro services efficiently. > > > > Chen Li > > > > On Mon, Jun 29, 2026 at 10:46 PM Yicong Huang <[email protected]> > > wrote: > > > > > Hi Tanishq, Chen and all, > > > > > > Just wanted to add my two cents here. My local dev env has always been > a > > > headache. I used to use M1 macbook pro + 8GB RAM (from 2022?), and > > > constantly got services OOM, build failure, and slowness. Because of > > this I > > > had to upgrade to M4 mac mini + 24GB RAM this month. The hardware > change > > > solved the issue to a certain extend, where many of the issues remain > on > > > our dev/infra tooling. For instance, there are too many services and > I/my > > > agent always forget to rebuild some of them after change, and it is > > always > > > hard to understand which version (branch/worktree/commit) of service > was > > > started by my agents. > > > > > > To solve this, I have recently introduced a new tool `bin/local-dev.sh` > > > which can help developers to boot up a dev setup with one command. It > can > > > also have options to skip some services if not needed. It can also > manage > > > versions of services and use one command to refresh changed services. > > Here > > > you can read more about it > > > https://urldefense.com/v3/__https://github.com/apache/texera/pull/5961__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Mr3uz1cItQMYuQb489hbDyIxoTzsbU_m_HgLAr8-TtJg3XuT_oja4at4zks9Cf9c9dwEzegPlYuA0A4$ > > , > > > and see demo videos of the usage. > > > > > > The main purpose of this tool was to let agents manage texera services > > > easier during local development, by providing CLI for texera services. > > > Before this, I would ask my agent to launch texera services locally and > > > after 1 hour it still stucks. With this tool + a custome MCP connecting > > to > > > Claude Code, I can switch between branches and get services deployed > in > > > seconds. > > > > > > I believe it can also help human developers as well. For instance, I > also > > > provided `bin/local-dev.sh -i` which is an interactive interface where > > you > > > (human) can manage services, view MEM and CUP usage, and view logs. > From > > > what I see, local-dev mode does not need that much of memory, and we > can > > > further optimize the usage. I do believe running dev services on a 8GB > > RAM > > > machine is possible, as long as we don't do massive data processing > > during > > > dev. > > > > > > I am actively improving this tool, please feel free to try it out and > > > create issues for bug report or feature requests. I hope this can make > > > developers life easier. > > > > > > > > > Best, > > > Yicong Huang > > > [email protected] > > > > > > On Jun 29, 2026 at 9:58 PM -0700, Chen Li <[email protected]>, wrote: > > > > Tanishq, > > > > > > > > Last time you mentioned finding a way to run the system on your > laptop > > > with > > > > 8GB memory, by turning off unneeded microservices. Can you share the > > > > information and add it to our GitHub site to benefit other > developers? > > > > > > > > Thanks, > > > > > > > > Chen > > > > > >
