[UI] Binding to app within service creation fails for some services

[thehejik]

Describe the bug
On Epinio v1.9.0 when trying to deploy a new service and binding it to an existing app (in one step) over UI the binding fails. The service itself is deployed.

Service instance was created but failed to reach a deployed state in a reasonable amount of time, no applications were bound

@mmartin24 found out that it depends on used service catalog as it works ok with postgres and mongo but doesn't with redis, mysql and rabbitmq. It seems that more resource hungry services are unable to bind. Also he was unable to reproduce this issue on Epinio v1.8.1.

I was unable to reproduce that over CLI when trying emulate the UI by:

epinio service create mysql-dev mysql-sample --wait && epinio service bind mysql-sample sample

To Reproduce

1. Install Epinio v1.9.0
2. Push an app over UI
3. Create a mysql-dev service instance and bind it to the app in one step over UI (according to the screenshot)
4. Click Create and see error

Expected behavior
The service is created and bound to the app

[enrichman]

This looks like a timeout issue in the UI, because when the service takes too long then the UI will warn you with this error.

Possible fixes:

• increase the timeout
• since the binding of the app doesn't need the full creation of the service it can be done without waiting, maybe with just a small delay after the service creation (just to be sure that the service was "initialized/created"

[richard-cox]

At the moment we wait for a service state of deployed. Before this it's only not ready

There's a comment in ui code Service instance must be `deployed` before they can be bound to apps.

I tried not waiting at all, having a 2 second gap and also waiting for the resource to be returned via the API, and all give the same 404 response to /bind release 'x443639a38b5fbf3853fe23f3bf7d' does not exist.

@enrichman Can we add a new state to the service resource that indicates it's ready to be bound to?

Edit: In case someone else picks this up it's in dashboard/pkg/epinio/edit/bind-apps-mixin.js waitForServiceInstanceDeployed

[mmartin24]

As @richard-cox in CLI if no wait is placed in between svc creation and binding, indeed an error appears not finding the svc:

➜  epinio git:(main) epinio service create postgresql-dev postgres-sample && epinio service bind postgres-sample sample

🚢  Creating Service...
Catalog: postgresql-dev
Service: postgres-sample
Wait For Completion: false

🚢  Binding Service...

❌  error binding service: service bind failed: release 'x75ed1dbc13b646923fddfc26d983' does not exist

However, please note that in the UI this works for postgresql and mongo but not for the rest

Now, if we add just a couple of seconds in between this works:

➜  epinio git:(main) epinio service create postgresql-dev postgres-sample && sleep 2 && epinio service bind postgres-sample sample

🚢  Creating Service...
Catalog: postgresql-dev
Service: postgres-sample
Wait For Completion: false

🚢  Binding Service...

➜  epinio git:(main) epinio service list                                                                                          

🚢  Listing Services...
Namespace: workspace

✔️  Details:
|      NAME       |            CREATED             | CATALOG SERVICE | VERSION |  STATUS  | APPLICATIONS |
|-----------------|--------------------------------|-----------------|---------|----------|--------------|
| postgres-sample | 2023-08-30 11:52:11 +0200 CEST | postgresql-dev  | 15.1.0  | deployed | sample       |

[enrichman]

@richard-cox that's strange because as @mmartin24 tested it should work with just a few seconds wait from the async creation. The service resource should be there and we should be able to bind it.

[richard-cox]

@enrichman It seems a bit of a workaround to add a wait, which would be susceptible to cluster performance. Would there be a large impact to adding a created state? Would the not ready state be enough to indicate the resource was created?

[enrichman]

It is a workaround. The service should be created as soon the create returns, and the not ready state should be the status you are looking for (the service is there, but not ready yet). Are you doing the binding in parallel or after the creation?

This should work:

service create -> OK -> service bind 

if not it's strange. You can avoid the wait with a get, to check for the service:

service create -> OK -> while(service get) -> OK -> service bind

otherwise the most appropriate way on doing this would be probably have a flag to do this in one API request. This would be an enhancement.

[richard-cox]

The current flow is synchronous

• api request to create service
• loop every 2 seconds
  • check if that service state is deployed
  • fetch service
• when deployed make api request/s to bind app

I tested the above with just waiting for the fetch service to return something. Then also if that state is not-ready. Both still fail.

I've created a PR wait for not-ready....and also to then wait 2 seconds. This isn't great, but resolves the issue.

Edit - Might be worth noting i'm seeing really slow response times for service related network requests. Could be many things, but warrants some investigation

[thehejik]

Now it behaves differently but still not OK.

The service creation with bind usually works fine when the app is freshly deployed but if you delete the service and try to create/bind another service to the same app the action does fail on configuration not found. I guess it might be some race condition as the configuration is in some transient state or not created yet. Also I can reproduce this issue on services that were ok before as postgres.

[richard-cox]

@thehejik from what i can tell this might be an api issue (epinio/epinio#2551). Separately I spotted edit service is broken and will be fixed by #340

[mmartin24]

This looks ok to me.
I tested both issues commented on this ticket and they worked for me:

• the original issue (unable to bind at times when doing so in 1 step on several services)
• the secondary issue commented above not being able to bind after a binded service is deleted.

I checked them first here and rechecked again today with v1.10.0-rc1-10-g09a409e9 and latest epinio ui, by creating service and binding in 1 step, deleting binded service and repeating the service creation and binding again in 1 steps.

So looks ok in my opinion.

@thehejik , ok with closing this issue?