that is another good thing to add to the list however.
mobile support should be fairly easy to add.
if it can be of any help: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Classes#Browser_compatibility
really this is where things like babel are perfect, however i am not using it in the project because i wanted to keep things as simple as possible.
@Pasvaz thanks for pointing the issue with the flight tracker. traffic (https://github.com/swimos/traffic), transit (https://github.com/swimos/transit) work on safari and on android.
https://traffic.swim.ai/ and https://transit.swim.ai/ are the hosted versions of those repos respectively
just pushed some quick fixes for safari to github and the live demo site.
Hi @ajay-gov
I am a two day old swimer, so the workings of swim are not yet very clear to me. I'm exploring the possibility of using a webagent as single writer per stream in an eventstore.
How does one co-ordinate planes in such a manner that only one instance of a webagent can exist in a "cluster" of servers, and ensure that requests are routed/re-routed to the node that has that instance running? this would be kind of like what you can achieve with Akka Cluster Sharding. Is this possible with swim?
@oexza great question and this feature is supported In swimos. The way swimos distributes webagents in a cluster is very similar to how data is distributed across a Cassandra cluster(https://docs.datastax.com/en/cassandra-oss/3.0/cassandra/architecture/archDataDistributeDistribute.html). swimos computes a hash token for each agent's uri and the hash token is configured to a certain partition which pins agents to specific servers in the fabric. You can then address that agent using a swim client (or a browser running a swim app) by pointing the client to any plane in the fabric. Swim will transparently demultiplex links opened by external clients, and route them to the appropriate server in the fabric.
For a concrete example on how to do this please refer to the swimos transit repo- https://github.com/swimos/transit. Specifically look at the https://github.com/swimos/transit#run-as-a-fabric section which guides you through the steps of how to setup a two server fabric. The partitions are configured in the server-a.recon and in the server-b.recon. Here we have a partition with two servers with agents being evenly distributed across both servers. Agent Uris that hash to a token that lies between 0x00000000 and 0x7fffffff are pinned to server-a and agent uris that hash to a token that lies between 0x80000000 and 0xffffffff are pinned to server-b
@oexza, thanks so much for checking it out! The k8s question is the next logical step indeed. Since swimos applications are stateful they need to deployed using k8s statefulsets
The pods need to communicate internally but they can use the POD DNS names to look each other up (instead of using static IP addresses).
External access to the k8s cluster can be made using the DNS name provided by the k8s load balancer for this cluster, swimos will then route requests to the appropriate pod internally.
The pods need to communicate internally but they can use the POD DNS names to look each other up (instead of using static IP addresses).
External access to the k8s cluster can be made using the DNS name provided by the k8s load balancer for this cluster, swimos will then route requests to the appropriate pod internally.
Using the transit as a concrete example, say you have the following set up in your k8s yaml file
Service (ns/name): swimos/transit
StatefulSet (ns/name): default/transitthen the DNS and Hostnames will be
StatefulSet Domain: transit.swimos.svc.cluster.local
Pod DNS: transit-{0..N-1}.transit.swimos.svc.cluster.local
Pod Hostname: transit-{0..N-1}For the 2 pod example with server-a as index 0 and server-b as index 1
POD DNS for server-a will be: transit-0.transit.swimos.svc.cluster.local
POD DNS for server-b will be: transit-1.transit.swimos.svc.cluster.local
Also when running in a k8s cluster it will be best to run the app on the same port in all the pods. In order to do that change the port in
run it on port 9002 change the
@web section of the recon file. If you'd like torun it on port 9002 change the
@web section of the recon file to be: @web(port: 9002) {
With the apps running on port 9002, the next thing to fix is the the mesh part for the server recon.
For server-a.recon this will be:
@mesh {
@part {
key: partA
predicate: @hash(0x00000000, 0x7fffffff)
@host {
primary: true
}
}
@part {
key: partB
predicate: @hash(0x80000000, 0xffffffff)
@host {
uri: "warp://transit-1.transit.swimos.svc.cluster.local:9002"
primary: true
}
}
}For server-b.recon this will be:
@mesh {
@part {
key: partA
predicate: @hash(0x00000000, 0x7fffffff)
@host {
uri: "warp://transit-0.transit.swimos.svc.cluster.local:9002"
primary: true
}
}
@part {
key: partB
predicate: @hash(0x80000000, 0xffffffff)
@host {
primary: true
}
}
}
@oexza A lot to take in, so please do let us know if you have more questions. As you can see with the current setup you will need a recon configuration file for each pod in your k8s cluster which is a tedious manual process. We are planning to add a k8s setup feature in our enterprise product swim continuum that will simplify this setup.
thanks so much for the detailed answer @ajay-gov a very lot to take in. this truly is tedious and unsustainable for on demand scaling. One thing that would really ease the pain, is a way to programatically configure the mesh instead of using the server.recon file. this is so that some kind of service discovery mechanism can be used in code instead. this kind of configuration is used alot in Apache Ignite, where discovery is so flexible to the point that even a good ol' database table can be used a registration and discovery point for the cluster. this FLEXIBILITY is HIGLY important. Would there be a way to achieve something similar today in swim?
@oexza programmatically configuring the meshes is definitely the way to go about it. The mesh configurations in the recon file is for convenience (for non-prod or single node deployments) but everything in the recon configuration file can be done programmatically. In fact we did exactly what you are asking for one of our customers where we run a multi-node k8s cluster. We can share this code with you if you'd like. Please do let us know.
@oexza also these programmatic APIs are going to be used by our swim continuum product which will provide a UI console from which you can set up the meshes for your k8s cluster. Hopefully we will have that soon.
@oexza Committed the code in a branch of the transit repo. It can be found here: https://github.com/swimos/transit/tree/feature/cluster-config
The TransitPlane class has the code that programmatically loads the mesh configurations. Please do check out this section of the readme as well
The TransitPlane class has the code that programmatically loads the mesh configurations. Please do check out this section of the readme as well
@oexza you should run it using k8s stateful sets. k8s stateful set will start up node 0 to node 5 in that order. When the
TransitPlane in node 0 starts up it will try to connect to nodes 1-5 and you will see some ConnectionExceptions since those nodes aren't up yet. As the nodes come up these ConnectionExceptions will go away as the swimos kernel will try to reconnect and will eventually connect to those nodes. Usually you will see no exceptions in node 5 since nodes 0-4 would have already started up. Do let us know if you have any questions.
@itsjimos Hi Dimitris, swimOS is a software platform for building real-time applications. It's a self-contained vertically integrated stack, so you don't need an app server or a database or a message broker. swimOS also has UI libraries for real-time visualizations. Please refer to the docs at https://www.swimos.org/.
My only experience with real-time applications was through socket.io for a small side project. If I'm correct swimOS is destined to a wider variety of applications like IoT.
I lean towards web development and I'm curious of the use cases that there might be. I also see it a chance to get into DevOps. Thank you for your replies!
I lean towards web development and I'm curious of the use cases that there might be. I also see it a chance to get into DevOps. Thank you for your replies!
on the UI side of things our Swim Client uses websockets for communication but all that is handled by our client for you. We have a number of demos, many of which have web based UIs. You can find those on our github @ https://github.com/swimos
many of those demos have a live version as well. Some of those are:
http://traffic.swim.ai/
http://transit.swim.ai/
http://swim.flights/
http://swim.farm/
http://traffic.swim.ai/
http://transit.swim.ai/
http://swim.flights/
http://swim.farm/
I am a web dev as well and often build fun things with swim. My latest uses satellite data and three.js to create whats basically a pretty screensaver :D http://satellites.swim.ai/neon.html
Yea, these demos look great. I remembered that a couple of years ago I had some time with InfluxDB a time series database. To my eyes its philosophy is close to real time application technologies and therefore Swim. I will look at the docs and experiment a bit the following weeks. Keep it up!
hi Swim team, any updates on adding auth to server.recon? I am testing out swim now, looks very interesting. I see the Authenticator module but there are no examples , so am not sure how exactly to use it. Also can you comment on RBAC implementation with swim. Can we also control actions and roles with annotation or we write it from scratch ? Thanks!
@abhishaked_twitter We support authentication using the server.recon file now. If you are using google authenticator it will look something like this:
tutorial: @fabric {
@plane(class: "swim.tutorial.TutorialPlane")
myauth: @googleId {
@audience("yourkey.apps.googleusercontent.com")
@email("youremail@gmail.com")
}
}There is a similar way to add OpenId Authenticator parameters as well.
@abhishaked_twitter We also support authorization as well. You may intercept the incoming requests at different levels i.e on connect, on link to the server, on link to an agent, on link to a lane etc. and perform RBAC style authorizations at any of these levels.
You are right though, we haven't documented any of these cleanly yet. Our top priority right now is to revamp our developer site. We will definitely add sample code for authentication and authorization as part of this process.
You are right though, we haven't documented any of these cleanly yet. Our top priority right now is to revamp our developer site. We will definitely add sample code for authentication and authorization as part of this process.
@neeluuuuuu_twitter thanks for your question, please do look at https://www.swimos.org/ for more information on how to use swimOS
Hi, I was going through https://www.swim.ai/architecture/ and it says web agents do aggregate, transform, analyze. Is this for SwimOS or Swim Continuum?
@anoukh_ashley_twitter Those features are available in SwimOS, requiring only that you express them in terms of Web Agents, Lanes, and lifecycle callbacks.
Please see the tutorials for examples: https://www.swimos.org/tutorials/ . Aggregation is outlined under Downlinks, Join Value Lanes, and Join Map Lanes. Transformation and analysis take various forms through all the tutorials, but notably bridging to and from external processes is explained in Ingress Bridges and Egress Bridges.