12:57 <fabianski> Heya! Is there any documentation about a deployment or updating process of wallaroo and wallaroo applications? Especially in terms of data persistency when restarting wallaroo?

13:38 <pzel> fabijanski: We have some examples of deploying with pulumi+ansible under https://github.com/WallarooLabs/wallaroo_blog_examples/tree/master/provisioned-classifier but that project sets up a one-shot computation pipeline and then shuts it down.

13:38 <pzel> So likely not what you're after.

13:39 <pzel> Then, there's a resilience demo here: https://github.com/WallarooLabs/wallaroo/tree/master/demos/resilience-demo

13:39 <pzel> Which uses the relatively fresh resilience functionality to recover from induced crashes.

13:39 <jtfmumm> The existing documentation for resilience is here: https://docs.wallaroolabs.com/operators-manual/resilience-crash/.

13:39 <pzel> Maybe if you could describe your use case a bit more closely, we could suggest something more specific?

13:40 <pzel> Usually, if you're changing internal protocols, you're likely better off starting a new cluster with the new version of your data and switching over the data sources.

13:49 <fabianski> thanks for the links :)

13:51 <fabianski> rn we're thinking about replacing an existing own fw with a public one. Our usecase is parsing information out of emails and aggregating different information.

13:58 <pzel> fabianski: sounds interesting! By use-case here I was hoping to find out more about your setup: are you running on hardware, containers, etc? And what are the uptime requirements?

13:58 <pzel> because if you want to upgrade Walaroo workers in-place, you'll have to consider backward- and forward- compatibility in your processing functions

14:03 <fabianski> ah sorry! rn I do not have any setup :D just local try-outs. One approach that crossed my mind are containers (aws ecs/fargate clustering). We can accept down times in between deployments since we have a (almost) infinite queueing mechanism (aws sqs/s3) in front of the data processing.

14:08 <pzel> Oh, in that case do take a look at the `provisioned-classifier` examples with ansible. They're a bit ugly because of `pip-install`ling on the production nodes, but apart from that they can set up a cluster in single-digit minutes.

14:09 <pzel> We also use pulumi (www.pulumi.io) to provision machines, but this is orthogonal to deploying the cluster w/ansible. You can deploy on top of existing infrastructure by changing ansible/inventory.yml

14:14 <fabianski> Cool, yeah I was looking at it just now. Thanks that might help :)

14:17 <pzel> Good luck and keep us in the loop. A lot of this stuff is work-in-progress, and we definitely appreciate open source contributions :)

14:20 <fabianski> And another question :) There is nothing like pipeline configuration management in wallaroo right now, is it? Like where I can predefine workers and replace or append them in/to the a certain pipeline without reconfiguring/restarting everything?

14:21 <SeanTAllen> Correct. Pipelines are currently static and can't be changed at runtime.

14:24 <fabianski> Can you tell if sth like that is on the roadmap? :)

14:25 <SeanTAllen> It's not something there has been great demand for and is a rather large undertaking. So, at the moment it isn't on the roadmap. We do work with folks on a commercial basis for add features that they want. If that was a route you wanted to go.

14:27 <fabianski> That's fair, thanks :) Probably not, but I'll get back in case! thanks for the help!

14:27 <SeanTAllen> you're welcome