Also check out https://www.restate.dev/. We chose this internally after evaluating it against temporal, hatchet, and dbos. The docs are pretty good and development locally and deployment to k8s was simple.
A thing to note is that Restate is not open-source (BSL) while dbos (core, anyway) and temporal are. (Haven't heard of hatchet before.)
On the other hand, durable execution in dbos is implemented in libraries so you have different features for different languages (the Go one doesn't support SQLite as a backend for example while the Python one does), whereas Temporal and Restate are not embedded like this.
I landed up creating https://github.com/atmonostorm/derecho to answer the question of "in-process temporal.io". Works pretty well, with some unique features (error-controlled retry behavior) that temporal can't do.
I've used it pretty heavily in production, ~30m workflows of various types from various projects through it at this point. No capacity to support it as OSS though, and my internal persistence backend isn't easily extricable from monorepo, but it's just an interface with conformance tests.
Been eager for something that wasn’t temporal (egregious overhead and annoying multiple services), but they do write this like Temporal… doesn’t exist. They use a lot of the same pioneered techniques (like “our own context type”) that they do.
go-workflows has always been the good alternative, but I’m sure dbos is a bit better supported. Dbos always had some weird gaps (I don’t remember why exactly, I just remember saying “oh well I can’t use this then” more than once), but maybe they’ll close them with the go sdk
Thanks for the comment (author here). I wanted this post to focus on the Golang specific implementation, not dwell on the durable execution ecosystem at large.
With respect to context, I don't know that anyone invented "having their own context". Go interface are extendable and pretty much every major framework I know of implement their own context.
Would love to learn more about the gaps that offset you. We're constantly improving here ;)
In review, I think it might have been the workflow versioning being strange, and the lack of any heartbeating/crash detection for longer running activities
Thanks, I didn't mean it as criticism, I guess my 5am brain thought the way it was worded almost came off as like "look at our unique idea" which was a pretty common pattern.
We bounced off dbos when we found they charge $$$ for their CRUD web GUI ("DBOS conductor"), which they also "strongly recommend" for production use, for good reason.
Conductor is about enterprise features like automatic workflow recovery, alerting, or RBAC. The GUI is a nice to have -- but all your workflow data are in Postgres. You can access it very easily.
The offering would be enticing if some Web GUI features were behind a paywall. Separate "production" from "enterprise".
Right now the messaging is "you shouldn't use DBOS for production unless you are a paying customer", which is odd considering durable execution itself is a production-level concept. So we rolled our own in a few hundred lines of Python.
After spending way too much time debugging runtime problems with python based workflow tools, I’ve been implementing something very similar: DagGo.
DagGo is a type based workflow tool with observably written in Go. Jobs are compile time safe. I’m planning to bring it to feature parity with tools like Dagster over the next few months.
I notice you didn't provide any specific comparison alongside that comment, which makes me feel frustrated because I think the Temporal workflow SDK is very different. Architecturally, Temporal and DBOS are at two opposites of the durable execution spectrum. I'd love to understand what makes you think this work is a mere copy and paste. Would you be willing to share some more with me?
On the other hand, durable execution in dbos is implemented in libraries so you have different features for different languages (the Go one doesn't support SQLite as a backend for example while the Python one does), whereas Temporal and Restate are not embedded like this.
I've used it pretty heavily in production, ~30m workflows of various types from various projects through it at this point. No capacity to support it as OSS though, and my internal persistence backend isn't easily extricable from monorepo, but it's just an interface with conformance tests.
go-workflows has always been the good alternative, but I’m sure dbos is a bit better supported. Dbos always had some weird gaps (I don’t remember why exactly, I just remember saying “oh well I can’t use this then” more than once), but maybe they’ll close them with the go sdk
With respect to context, I don't know that anyone invented "having their own context". Go interface are extendable and pretty much every major framework I know of implement their own context.
Would love to learn more about the gaps that offset you. We're constantly improving here ;)
What do you find strange with workflow versioning? Would love to consider improving the semantics. In fact, we started doing it: https://github.com/dbos-inc/dbos-transact-py/pull/598
See "DBOS vs. other systems" on the github repository page[1]
1: https://github.com/dbos-inc/dbos-transact-golang?tab=readme-...
Right now the messaging is "you shouldn't use DBOS for production unless you are a paying customer", which is odd considering durable execution itself is a production-level concept. So we rolled our own in a few hundred lines of Python.
DagGo is a type based workflow tool with observably written in Go. Jobs are compile time safe. I’m planning to bring it to feature parity with tools like Dagster over the next few months.
https://github.com/swetjen/daggo