Attached disk prevents Instance from being replaced

[synhershko]

The following program launches an instance with an attached disk:

let i = 0;
    let d = new gcp.compute.Disk(runName + "-esdata" + i, {size: dataDiskSize, type: "pd-ssd", zone});
    elasticsearchInstances.push(
        new gcp.compute.Instance(runName + "-elasticsearch-data" + i, {
                machineType: "n1-standard-1",
                zone,
                metadata: {"ssh-keys": sshKey},
                metadataStartupScript: esDataNodeStartupScript,
                bootDisk: {initializeParams: {image: machineImage}},
                attachedDisks: [{source: d}],
                networkInterfaces: [{
                    network: computeNetwork.id,
                    accessConfigs: [{}],
                }],
                scheduling: {automaticRestart: false, preemptible: isPreemptible},
                serviceAccount: {
                    scopes: ["https://www.googleapis.com/auth/cloud-platform", "compute-rw"],
                },
                tags: [clusterName, runName],
            }
        )
    );

Running pulumi up completes fine, but updating some instance parameters (e.g. startup script) requires replacing the machine when running pulumi update. However, the following error is received:

error: Plan apply failed: Error creating instance: googleapi: Error 400: The disk resource 'esdata0-86abfa9' is already being used by 'elasticsearch-data0-5df5543', resourceInUseByAnotherResource

We should be able to replace an instance by detaching the disk from existing instance, and attaching it to the new launched instance.

[cowwoc]

2 years later, any progress on this? Alternatively, is there a workaround?

[leezen]

In this particular case, does deleteBeforeReplace work as a workaround? Right now, what's happening is upon replacement (create new then delete old), the new instance attempts to attach to the disk, which is already in use. By using the deleteBeforeReplace option, the instance should be deleted first, which should allow the disk to be attached to the new replacement.

[wpietri]

It's wild to me that a simple use case like this has not worked for more than 5 years. I'm just trying out Pulumi and it's making me question whether it's the right choice for us.

[zbuchheit]

deleteBeforeReplace was able to resolve this for me, but interestingly enough this behavior didn't seem to happen when using yaml. I only witnessed it while using typescript. I haven't tested in another besides typescript and yaml.