I recently changed the binding of a parent/child relationship for an

aws.sqs.QueuePolicy

. When it was updated, it created a

replacement

and deleted the old one. Turns out that they’re not distinct entities, they’re just a property of the

aws.sqs.Queue

. That means creating a replacement then deleting the prior essential just deleted the property all together, thus breaking our workflow and losing a few hours of messages for the Queue. This seems to be a bug in the update strategy. Should I create an issue on

pulumi/pulumi-aws

or am I missing something?

@thousands-hairdresser-72380 when you say you changed the parent child relationship, you mean as a resource option? If so, this is expected behaviour and you can use an alias to prevent a replacement

@billowy-army-68599 Yes as a resource option. Replacing isn’t an issue in itself. The issue is it didn’t create two entities, it just modified the same property on the

Queue

entity. So the order of operations was 1. Create the replacement

QueuePolicy

a. This equates to updating the

policy

attribute for the

Queue

b. This is not a separate entity but a property of the

Queue

2. Delete the previous

QueuePolicy

a. Since the

policy

is not a separate entity but a property of the

Queue

, it just deleted the

policy

attribute altogether. The resolution ended up being running

pulumi refresh

so it detected that the policy had been deleted, then running

pulumi up

again.

Again, using an alias would have entirely prevented this

@billowy-army-68599 I’m not sure that an alias is the correct solution here. To provide more context and evidence of that claim, I’ve created an issue here Thank you for taking time to help out!