Azure-functions-host: HostID Truncation can cause collisions

[mathewc]

Currently when generating a default host ID we use the host name (slot host name) and we truncate to 32 characters max (code here). This ensures that the generated ID conforms to the core SDK length restrictions (code here).

This truncation can of course open the possibility for naming collisions, particularly in the case of slots. In slot scenarios, if the site name is over 32 characters long and a slot is created that starts with the same 32 characters and is only disambiguated in later characters, both the production and the slot site will be using the same host ID. This can lead to issues. For example, TimerTrigger uses the host ID as a component of the blob lease path. In this case the timer function will only be able to run in one of the sites because they're competing for the same lock. Similarly, customers often have apps deployed to different regions using the same long naming path, varying only in later name components (e.g. region/environment), and can run into this.

[paulbatum]

This is a dangerous change to make for v1 but we should fix it in v2. This would require some updates to the scale controller as I believe it has this logic duplicated.

[paulbatum]

I've confirmed the scale controller will need an update, but it should be straightforward once we know what the new logic for generating the host ID is.

[paulbatum]

We had to punt this due to higher priority issues and the need to coordinate the change across multiple components. Revisit again in V3.

[mathewc]

As a workaround for customers running into this issue, in Functions v2, you can set an explicit HostID in app settings, using a different ID for each environment. The app setting name to use is AzureFunctionsWebHost:hostId. For Functions v1 you can specify the ID in host.json via the hostId property. The host ID value should be unique for all apps/slots you're running. The important thing is that the IDs are under 32 characters. The restrictions for HostIds that the value must satisfy are here. Another way to generate an ID would be to take a GUID, remove the dashes and make it lower case, e.g. 1835D7B5-5C98-4790-815D-072CC94C6F71 => 1835d7b55c984790815d072cc94c6f71

[ankitkumarr]

Had a customer issue last week, where this truncation caused collision between multiple function apps sharing the same storage account.

@fabiocav, is this already planned for V3?

[fabiocav]

Deferring this work as it would have scale controller dependencies and does not align with the timing.

[mike-urnun-msft]

@paulbatum @fabiocav Per #1904 regarding the same detail on architecture center doc, we communicate that the Function App name length to be 1-60 - could you advise what the appropriate length that we should update it to?

[paulbatum]

@mike-urnun-msft good catch. staying in the range of 1-32 will avoid this bug.

[dgard1981]

Though there is no specific mention in this issue, I assume that the solution of explicitly setting the host ID via the 'AzureFunctionsWebHost:hostId' App Setting is still valid for v3 Function Apps?

If so, does the host ID have to remain static, or can it change after every deployment? I ask because our CD pipeline updates App Settings through an ARM template, so if the host ID has to remain static we'd first need to query the App Settings to get the current value of 'AzureFunctionsWebHost:hostId' for the slot we are deploying to so that it can be set back with the same value.

Also, does this only apply to a Function App with a slot, or does the value of 'AzureFunctionsWebHost:hostId' have to be unique across Function Apps? For example, if you have these two Function Apps with no slots...

• my-really-really-long-functionapp-uksouth
• my-really-really-long-functionapp-ukwest

...could both have the same truncated value of 'my-really-really-long-functionap'?

[tyler555g]

Why is this not mentioned in the docs or warned upon creation? This has been known since 2017, but no obvious mention of it anywhere that I can find. ( I very well could be blind, so if so please correct me. )

[davidmrdavid]

+1 I just recently got hit with a case that appears to be, at least partially, caused by this limitation.