Workspace deploy fails because storage account is not unique

[t-young31]

Describe the bug
A clear and concise description of what the bug is.

I've just deployed a workspace and get

mError: \u001b[0m\u001b[0m\u001b[1mcreating Azure Storage Account \"stgws2a73\": storage.AccountsClient#Create: Failure sending request: StatusCode=0 -- Original Error: autorest/azure: Service returned an error. Status=<nil> Code=\"StorageAccountAlreadyTaken\" Message=\"The storage account named stgws2a73 is already taken.

looks like the last 4 chars are maybe not unique enough. Seems odd though if it's ~1/10^6 chance

Steps to reproduce

1. Deploy a workspace
2. Be unlucky!

[marrobi]

Very unlucky - depends how many TRE workspaces have been deployed. Maybe we can add the TRE ID into some of these identifiers.

[RIashfaqahmed]

Very unlucky - depends how many TRE workspaces have been deployed. Maybe we can add the TRE ID into some of these identifiers.

thats a good idea!

[JaimieWi]

Just to mention, we have also come across this issue three times in the last week!

[martinpeck]

FYI this is where we generate that name: https://github.com/microsoft/AzureTRE/blob/main/templates/workspaces/base/terraform/locals.tf#L4

We currently take the last 8 characters from a string that already includes the TRE ID. If we took more than 8 (e.g. like we do for the KV name) then we'd pick up the TRE ID as part of the storage account name. However, we have a limit of 24 characters....but that should still be fine.

[tamirkamara]

The root cause of this is using last 4 chars of a uuid (v4=random) for azure resources that need to be globally unique. The complicated factor is that templates make naming assumptions on resources they reference that were created by a parent template. This is also true for some of our python code mainly around the airlock (but is easier to fix).

We discussed a couple of options to resolve this:

1. Increase the uniqueness level of the chars we use but keep the approach of the template developer making naming assumptions. e.g. use more chars of the uuid string (probably not going to help that much), use a real random string rather than the uuid4, etc.
2. Having a way for the template to get the properties of its parents. The pipeline mechanism has something similar but for a specific use-case which we can consider making more generic.
   A template could generate its own random string as it see fit, which will be saved in cosmos (via rp outputs). A descendant template could then reference the property and the pipeline could replace that before sending the operation to the RP.
   This method will be challenging for situations where we have cross workspace references. Maybe in Airlock review? TBD

[marrobi]

Hit this again today...

Status=<nil> Code=\"StorageAccountAlreadyTaken\" Message=\"The storage account named stgws07e4 is already taken.\"\u001b[0m \u001b[31m│\u001b[0m \u001b[0m \u001b[31m│\u001b[0m \u001b[0m\u001b[0m  with azurerm_storage_account.stg, \u001b[31m│\u001b[0m \u001b[0m  on storage.tf line 1, in resource \"azurerm_storage_account\" \"stg\": \u001b[31m│\u001b[0m \u001b[0m   1: resource \"azurerm_storage_account\" \"stg\" \u001b[4m{\u001b[0m\u001b[0m \u001b[31m│\u001b[0m \u001b[0m \u001b[31m╵\u001b[0m\u001b[0m \u001b[31m╷\u001b[0m\u001b[0m \u001b[31m│\u001b[0m \u001b[0m\u001b[1m\u001b[31mError: \u001b[0m\u001b[0m\u001b[1mcreating Azure Storage Account \"stalimappws07e4\": storage.AccountsClient#Create: Failure sending request: StatusCode=0 -- Original Error: autorest/azure: Service returned an error. Status=<nil> Code=\"StorageAccountAlreadyTaken\" Message=\"The storage account named stalimappws07e4 is already taken.

[marrobi]

@tamirkamara where are we with this? Is #3243 a solution?

People keep hitting it, would be good to get a fix in - if no 2 above is to complex, then can we maybe get a more random ID as per 1?

[tamirkamara]

tamirkamara where are we with this? Is #3243 a solution?

People keep hitting it, would be good to get a fix in - if no 2 above is to complex, then can we maybe get a more random ID as per 1?

We have implemented no2 and it's merged.
However we haven't yet managed to go and update all the templates. Plus you had a concern about having the unique string as a property in the template. So I'm not sure when/if/how we'll proceed.

[marrobi]

This is coming up again as an issue. I think will look at a temporary fix just to check the last 4 digits is unique, then longer term when time allows, we can go through and update all the templates.

What was the conversation about the ID being a property? A property of the workspace service?

[marrobi]

@SvenAelterman I think we need a short term workaround for this, I have tested something like:

    async def is_workspace_with_last_4_id(self, workspace_id: str) -> bool:
        from azure.mgmt.storage import StorageManagementClient

        storage_client = StorageManagementClient(credentials.get_credential(), config.SUBSCRIPTION_ID)
        # check for storage account with last 4 digits of workspace_id
        availability_result = storage_client.storage_accounts.check_name_availability(
            {
                "name": f"stgws{workspace_id[-4:]}"
            }
        )
        return not availability_result.name_available

Which works, but as per the discussion above we need to have a more robust solution.

Which could be as you suggested in #3862, use the tre_id in names:

stgws<tre_id>ws<ws_short_id>

But we also need to adjust templates to use outputs of parent templates, such as the workspace storage account name, rather than a hardcoded format. There is also the airlock to consider as per #3243.

I propose a workaround, then we schedule getting this sorted for good, which will be a breaking change on the base workspace and child templates.

[tim-allen-ck]

PR #3863

[marrobi]

Had a chat with people about this and the view is using unique IDs for the full name - all characters, and then tags for the descriptions might be the best approach. If we use the TRE ID then we are going to start running out of characters.