Elsa Workflows Community, 14 Jun 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Elsa Workflows Community Meeting 13 ( 2022-06-14 )

Description

Meeting 13
Topics

- Elsa 3
○ Toolbar
○ Transposition
○ Prevent cyclic graph detection
- Multitenancy
○ What levels: record level, container/configuration level
- Composite activity stores everything, regardless of storage provider
○ Jay to create an issue
- Auth
○ Jay to write a blog post about authentication & authorization with Elsa
Add a section to homepage pointing to other Elsa related articles?

A

Hello aj did you have any.

B

Topics jake for today's meeting, no specific topic, maybe just a discussion. I think you don't have so much time to to look at the pr.

A

I I have not had any time to look into it. Yeah.

B

I know what you're referring to, unfortunately not yet, but it's on my list- and um I think my next topic for me is um the multi-tenant and I think also I'm working on a authentication and authorization.

B

Maybe I will take a look at the documentation, but maybe I will write some blog posts to help other to configure some some authentication and authorization on the controller.

A

You're talking about writing a blog post, describing the steps wonderful.

B

Yeah, that would be awesome very good. The idea is to explain how you can add the authorization for all the controller and on the ui side, how you can configure the http client of the ui to send a beer automatically for all the requests.

A

Okay, that's cool. Do you have a blog website somewhere that you write more blog posts regularly or once in a while, because I was thinking- maybe we we should add a section on the homepage or the readme of the elsa project, where we can point others to those blog posts. I think I read a blog post from you at some point, but I'm not sure did I lose you no you're. Still there all right, so yeah I'll just get started with elsa 3 updates. There's some noise coming from you. It seems jay. Now it's gone.

A

Thank you! Hey france, hey! Let's go to the elsa 3 updates! Personally. I've been working on the journal, but I wasn't able to make much progress because I ran into some oversight which I will explain later, but before we get to that, let me show something cool. So here we're looking at the new designer and the first thing you may notice, if you, if you're familiar with the new designer.

C

A

There are two new buttons called zoom to fit and outer layout and auto layout is not actually implemented. This was added as part of little user story to add a toolbar, that's extensible, so we can add more buttons to it, but zoom to fit actually works. So let me show you that so now, when I click this it zooms to fit, as you can see, there's still some.

A

It needs some tweaking still because it's not taking into account this toolbar itself and maybe some more freaking room would be, would be nice and also maybe we don't want to exceed the 100 zoom. So zooming out should be fine, but I don't think it makes much sense to zoom in more than 100 or maybe a little bit more, but not too much probably will make it configurable, but we will see- and of course it also works when you have a very large workflow.

D

A

Like this, for example, and now you can zoom to fit, you know everything fits onto the canvas but again take into account that there's this little issue still. So this is going to be useful, especially once we have other layouts. So then, when you have all sorts of connections and then you made a bit of a spaghetti workflow and you can't identify the flow anymore, you should be able to click outer layout and then it will use probably a some dagger algorithm to lay it out in some pre-structure. So.

C

Yeah question uh there you go, I think you're gonna use the maximum minimum of the boxes that support activities. What if the connectors are extending beyond those maximum minimum, for instance, having a connector line going over the most top one? Do you take that into account or not.

A

You mean for the zoom to fit algorithm yeah. Exactly um I'm not entirely sure I I get the situation. Can you can you rephrase.

C

A

C

Yeah, if you, you have a connector now. This is a good example, so the maximum of the width is actually the connector itself, not the box. Oh.

A

I see a good question: let's see, let's, let's remove this one, so to be clear, the algorithm used is implemented by x6, which is the underlying library that we're using. So I don't know what it's using okay, but let's see what happens yeah it looks like it takes into account the boxes, but not so much. The connectors yeah.

C

Then a bit of briefing rooms should be handling that better exactly.

A

Yeah, so some more tweaking uh is warranted, and I see little things like that horizontal scroll over here unnecessary one step.

C

At a time yeah, it looks a bit already yeah.

A

Exactly so, that's all I have for elsa 3 workflows, elsa 3, workflow progress, as I mentioned, there's some well. I ran into a little issue and I will try to explain so, as you can see here. Let me actually delete this and start with an actual workflow. So let me add this one and then add this one for each, because this is important and then two more of these. So now, let's connect these two and we.

C

A

Ignore the done for now, let me zoom to fit so this workflow. The way it's supposed to work is as follows: the first one there's no context menu here, so this is the starting node. So this will execute first and keep in mind. This is a flowchart style activity and the next activity to execute is determined by the connection right.

A

So we have a connection to the forces and then, as you may recall, from previous sessions, the for each activity has a body property, so the property is of type I activity, so it's not like in elsa 2. Where then the for each has an outcome of body that then schedules the next activity, that's not how it works anymore. Nails are three. The four its activity itself will schedule its body activity.

A

So in order to make that work from a model that's generated by this designer, where everything is just connected nodes, I need to do a little transformation and the transformation would be that I would basically follow the next connection from the designer and then whatever node is connected. I will take that node and add it as a property on the for you, so for it would have a body property, and then I assign that property.

A

The value of the rightline activity in this example- and I call it transposition and it works, except for the fact that I didn't take into account the next one. So here whenever the for each activity would execute, it would successfully schedule the writeline activity, which would then execute, and then because this one is now finished before each activity would complete and because it would complete the entire the outer flow chart would execute or schedule its next child. What it would do is it would see.

A

Oh okay, I have no more activities connected because it would not see this one, because this one is connected to that one, and because this one is completed and there's no other activity. Well, I could, of course, this one here to done so if we would execute this workflow after for each executed, it would have executed this one and then finally, this one, but this one would be skipped. So that's a that was an omission on my end, which I'm fixing and the way to fix.

A

It is by looking ahead a little bit because the the body property is only accepts a single activity, so I cannot put two activities into it, so I could not add this one as a list into the body.

A

So, instead, what needs to happen is this body activity itself needs to be a container, so first I create a container and to that container I then add these activities and then whatever that container may be, would then handle the scheduling of these two activities and because the nature of this designer, with this flowchart without containment, the default container, should be a flowchart. We can add a little bit more logic later on, that can be more intelligent and try and add a sequence for example.

A

So in this case this could be wrapped by sequence and then that sequence would be assigned to the body property which would make sense. This is how you would do it in when you use the c-sharp syntax to define workflows, but from the designer point of view, it's.

B

A

B

A

B

That's uh exactly how I see the thing I think, but uh I think we already discussed on this. We we could modify the designer to uh embed all the activities of the sequence, maybe directly.

B

Yes, I say uh I think we already discussed on this, but it's exactly how I see the thing to move all the the activity in a sequence in a container and after could allow us to embed sequence in the ui component of the forage. That's.

A

B

A

B

Collapse and we can collapse it yes,.

C

B

On user experience perspective, it will be the best thing. I.

A

Agree so so that would be one of the next goals to create a different kind of designer. That does exactly what you just described, jay, where we can contain notes. So we would have a sequential diagram sequence designer and that looks very similar to windows. Workflow foundation.

A

We would have a float chart, a different type of flowchart activity, where you can still have this flow chart style, but each node in the flowchart would itself be a container, so you would expand it and then either add a sequence or any individual activity or another float chart or state machine whatever.

A

Then it would be a more natural fit to the programming model or the workflow structure of valsa 3, because here we are basically mapping a different structure onto the elsa 3 structure, and that would prevent us from having to do what I'm doing right now.

C

A

Reason I am doing this is because this kind of designer is easier to implement and it's more similar to lsat 2.. So those two things I think make it easier for me to at some point release an initial version which will work well and then we can evolve it over time with different kinds of activities in their designers.

A

One thing I was wondering about is what would be the name if we want to keep this activity, which is currently called flowchart, and we want to introduce a another kind of flowchart that is very similar to the windows workflow foundation. Flowchart. We can't use the same names, but if anyone has any suggestions for a name or maybe we could configure the flowchart activity itself to use one designer or the other, it could be. In other words, it could be a configuration, but this is still outstanding for me,.

C

I don't have a name for it. I will think about it, but going through the whole tree it's kind of challenging. I guess because there could be, and for each in that one and if and when do you see that it's ending and which elements because you can have multiple splits, basically in the child's element after the four each you have to take that all the neck into account.

A

Exactly that's the challenge, so, for example, we could have another for each here and and of course, the switz, which is a whole different thing. I haven't worked so that's broken right now, but a switch has its own set of rules to provide more connections to child activities. So the way I'm going about that is, of course, there's recursion into it. So what I'm basically doing, I'm taking all the connections starting with the root, so the flow chart has this notion of the the first or the starting node and using that as a root.

A

I take all of its connections so in here there's this one connection and then, at the other end of this connection, there's this target. So this is the source node. This is the target node. I then look at its connections, so here there's two outgoing connections, the first outgoing connection.

A

This one has a target of rightline, but for each connection I also look at the source port, so body support done is a port and I look at the name of this port and if the port name matches an outbound property on the activity, then I transpose the target of this connection into that property, and this is what I do recursively.

A

So that way, this will work because from from this point on, because I noticed that there's this outbound property, this outbound activity property- I should say I know that I will now create a sub flow or a container for now. I'm using a subflow, as I mentioned it could be a container- could be something else to be determined, but by default I will use a subflow and then I basically get all descendant connections. So right now, there's just one two descendants- and this is the this- is the current connection.

A

So two descendants- I basically cut the rest of the tree and then paste it into the subfloor, and I and then I repeat that process into the subflow and then at some point I will get into this portion here.

C

A

Do the same steps, so it's a little bit tricky it works. It would have been simpler if the designer matches the the structure of the underlying model, of course, do.

C

You take an account for an entertainment. Loop was last for each could re-route back to the first right line or as an input for the other four each. I guess yeah, you could do things like this, then recursively you're gonna be having an infinite.

A

Loop here, yeah exactly these things need to be taken into account right now. It's not even finished so, let alone infinite. Looping is not handled, but we will have to handle those scenarios. I don't even think this should be allowed, and the nice thing about containment is that this would not even be possible, but with this free flow editor, you can basically do anything unless we build in these cards right. We could determine that. Oh wait. This is now part of a lower level part of the hierarchy, so this should not be possible, but yeah.

A

D

A

D

Normally there is a cyclic graph detection, which basically says that a child node should not have a path leaning back to the parent, so there are some algorithms for detecting cyclic graphs. I think it could help to prevent this, because if, if we don't prevent that your algorithm itself to find the child activities, it will go into an infinite loop. What did you call this cycle? Cyclic graph graph.

A

Prevention, detection, great yeah, that's that's a great remark. I will look into it and it makes sense the child should not look back into one of its parents like this. Is that the case? Because.

C

A

C

A schedule that uh I saw some samples in elsa 2, where you could say well, if you're, I think, in the documentation, approval thingy.

B

C

You could wait within a split and join that you have to wait and it sends a remark every one day to remind you to approve or not, and then you go back. I don't know if that is going to be the same structure here yeah you can see that it's going to go up again so.

A

Lc2 is interesting: on the one hand, everything is of a single hierarchy, there's there's actually there's no hierarchy, but then, at the same time it tries to model some hierarchies artificially, so by default, everything is flat. So this is a flowchart these activities. None of them are parents of others.

A

They are just connected in this connectome if you will, but then there's activities like for eats or if and then well, let's take d4 eats as an example before eats it's not shown here, but the four eats has an iterate outcome and then, whenever the iterate branch has completely finished, the four years activity itself is scheduled again and basically trying to mimic a parent-child relationship between the parent for each and the child iterate activities. But it's that's not how it's modeled structurally but functionally it's. It tries to achieve that. uh Behavior.

A

Compare that to elsa three in also three. If there is actually an article structure and there we are using this the same flowchart editor. This is more design. This is intended to be used on a level structure, not not an hierarchical structure. So there's a mismatch there. So in lc2 it's it should work. It's it's uh it's fine, except for the fact that the activities like for each try to mimic an hierarchy which it shouldn't. So I think to fix this mismatch in lsat 2.

A

We should remove the parent child's attempt, but we can't do it because we will break everyone in elsa 3. The structure is already hierarchical in nature, but we should fix the designer by supporting containment and even if we have a flowchart designer within that particular flowchart, the the nodes within it will be on the same level, there will not be parent child nodes. In that regard you can have like in a windows foundation.

A

You can add a note and then you can add activities to that note, but those activities will be contained by said note and then there's different kinds of notes. There's a container note. There is a switch node, I believe, but those are specifically crafted for the flowchart model.

C

As long as it's contained and that in the container itself, it's not like an hierarchy.

B

C

It's gonna work and then you should prevent it indeed to have. It relate back to a parent node that would work, but otherwise I couldn't see how it would work. If you have a reminder that is rescheduled after a day over and over again, because then it feels like you have to give it back to your parents, but it's apparently on the same note exactly.

A

If it's a true flowchart, then the reminder and whatever activity you want to do in response to the timer event, they are at the same level, and you would just connect back to the previous, though, because in a floater that's valid in this flowchart. This is actually actually this one is a child of four eats, but it's displayed as a as the same level and that's where the issue lies, but in reality this is a child of this one.

A

So here this could never be a send email activity, for example, while this is a timer when where this is a child of the timer that shouldn't work, but what should work is just having a timer uh node, which then is you know once it triggers it? It's done, and then the next activity gets scheduled through this connection, and that should work fine, but not when this is a container and that's the confusing part in this designer, because there's no way of telling oh, my body is a child, or this is a childhood for it.

A

You don't see it physically, so yeah moving right along what else. So that was a bit of transposition and that issue- and this is interesting jay did you want to talk about the multi-tenancy topic.

B

uh Yes, justice: uh when topics, uh I think we we need to work, because I think there is some missing titan and configuration for a different controller. For example, if you, uh I think, if you save your workflow today, there is no uh information about the tenant. Maybe I have missing some things. There is a test on a get on a ibm list, but not not for all the controller.

B

I think there is a something to do on this, and maybe, as I wrote on the discord, maybe we we have to write some uh elements and discuss on what we want to do with multi-tenancy. Be sure that we want to, for example, a split, a database with the item element or split to a multiple database, etc, etc. Maybe we need to specify different scenarios for for this. This is for multi-tenancy and what you, what you showed on discord, was more about the persistence issue with composite activity.

A

This issue is unrelated to not.

B

The tenants this is another topic, but this is about the persistence of property when you use a composite activity and it's also an interesting topic, because if you load or use a large data and you use a composite activity, there is no way to override the configuration and to be sure that you use transient to not store the data on the, for example, on the database, and this has a limit very, very limiting the performance of the flow.

A

Yeah exactly and that's the problem with this today with composite activities. This is not honored, so it will store everything as part of the workflow and that can be problematic, for, as you say, when using large data or non-serializable objects. This.

B

Is- and I think there is a two level- there is a content property here and, for example, I map the content property for the message property in the build method and, as you see, I have tried to force the transient or the metal property. I think I succeeded to this. It is uh understood by the engine, but the content property is not stored as transient.

B

It is always stored, yeah, exactly.

A

Yeah, we need to fix that. Did you raise an issue for this by any chance that we can uh look into for 2.9? Maybe sure I will return ah perfect. Thank you all right. That's good! So back to the multi-tenancy topic such a interesting, but also complex topic, because there's, as you say, j there's few levels at which you can make this. You know cut, there's multi-tenancy at the database level or at the record level. Actually so, where you use a discriminator on the record level using a column called tenant id for example.

A

So that's implementation in elsa was started, but it's it's not complete and it's it hasn't been completed. It's been worked on an occasion by people who use it in specific scenarios, but it's not ideal, because if you now, you think it's it's there and you start to use it and then you'll keep running into scenarios where multi-tenancy isn't fully completed, and this is uh this is a bit of an issue, so that's record level multi-tenancy and then there's of course, multi-tenancy at the service container level.

A

If you will, and at that level you can configure different database connections. If you want in this way, you can use same database server but use different databases for tenants depending on the database provider. You could even have the same database but use a prefix for a tenant so that you use the same database, but each set of tables are using their own prefix or maybe even their own schema. If database provider is four cents, we don't have such a provided today, except maybe for yes sql.

A

I think yes, sql might allow you to prefix your set of database tables for tenants, but none of this is implemented today. There is a pull request that adds this functionality and this a different kind of multi-tenancy. So this is not record level multi-tenancy, but it's configuration multi-tenancy, basically, where you can use configuration to specify your tenants. So here's a default one, but you could add more tenants to this configuration. The main reason this hasn't been merged yet is because of the implementation. This implementation requires certain modules of elsa to be aware of the multi-tenancy.

C

A

I am not sure, but I think it should be possible to implement container level multi-tenancy with uh unobtrusively sort of without any elsa services having to be aware of the fact that they are part of a tenant. I know that organ core has implemented it. I know artifact has support for multi-tenant service containers, so I discussed this with uh with this contributor and he's he's doing a bit of a proof of concept to see what's possible there. So.

B

This is for elsa too, and I think just about multi-tenancy. There is a question about the database when, uh where you, you store a lot of information, but there is also question about the storage of the event. For example, if you use the service boost, maybe you want to to use multi-tenancy also for the trigger, etc, etc. It's more large than just saving data in the database. Yeah.

A

Good point so, since this is configuration based, you could use a different connection string for different service bus so that your tenant is isolated in that way by using a different service bus altogether. But if you use like use the same database table the same configuration but have multi-tenancy at the record level, then maybe you also want that for the messaging stories where a message belongs to a given tenant and is then routed to the correct consumer of the tenant.

A

This requires some proper design sessions in some prototyping, which we may have want to do for, of course, the lc2 for elsa 3, there's nothing in terms of multi-tenancy, whatever the same architecture would apply there all right. So the issue you mentioned is.

A

Composited all right, these are the topics I had in mind. If anyone has any topic at this point, let me know- and otherwise we can end the session for today, all right thanks everyone for joining and participating and uh we'll see you next week. Thank you.

C

B

D