https://pulumi.com logo
Powered by
b

bitter-dentist-28132

10/02/2019, 3:24 PM
did 1.2.0 break upgrading via helm again? (e.g. upgrading the nginx ingress controller)
c

creamy-potato-29402

10/02/2019, 5:35 PM
@bitter-dentist-28132 not that we know of. do you have more details to share? also, when did it break before?
b

bitter-dentist-28132

10/02/2019, 7:27 PM
see #796, 793. now when trying to upgrade the chart, it says 
deployments.apps "nginxingresscontroller-nginx-ingress-controller" already exists
c

creamy-potato-29402

10/02/2019, 7:29 PM
@bitter-dentist-28132 let me see if I can reproduce the issue, though I would note that those aren’t helm-specific, and this one might not be either.
@bitter-dentist-28132 do you happen to have the code handy?
b

bitter-dentist-28132

10/02/2019, 7:30 PM
just adding it generically
Copy code
const nginxIngressController = new k8s.helm.v2.Chart('nginxingresscontroller', {
        chart: 'nginx-ingress',
        namespace: 'default',
        repo: 'stable',
        values: {
            controller: {
                metrics: {
                    service: {
                        omitClusterIP: true,
                    },
                },
                service: {
                    omitClusterIP: true,
                },
            },
            defaultBackend: {
                service: {
                    omitClusterIP: true,
                },
            },
        },
    }, {
        providers: {
            kubernetes: provider,
        },
    });
what changes is the 
metadata.labels.chart
c

creamy-potato-29402

10/02/2019, 7:31 PM
in the values?
there doesn’t seem to be a value with that path in the values
b

bitter-dentist-28132

10/02/2019, 7:35 PM
sorry, what i meant is that it's usually only 
metadata.labels.chart
that changes for the k8s objects, since most of the chart dev doesn't affect "standard" deployments. though it seems this time it's doing a full destroy/create of the deployments.
c

creamy-potato-29402

10/02/2019, 7:35 PM
do you know what you changed to instigate this?
or were you upgrading the chart version
b

bitter-dentist-28132

10/02/2019, 7:38 PM
upgrading the chart version. looks like the first one that failed was 
"nginx-ingress-1.21.0" => "nginx-ingress-1.23.0"
c

creamy-potato-29402

10/02/2019, 7:38 PM
trying now
b

bitter-dentist-28132

10/02/2019, 7:38 PM
looks like that changed to the new 
apiVersion
for deployments
c

creamy-potato-29402

10/02/2019, 7:39 PM
oh
that’s probably it
b

bitter-dentist-28132

10/02/2019, 7:39 PM
i guess pulumi won't order things properly for that kind of operation?
c

creamy-potato-29402

10/02/2019, 7:39 PM
we have a PR in flight for this, but it will take some time to land:
basically, what’s happening is, if you submit a resource that differs only in 
apiVersion
, kubernetes considers these the same resource, but pulumi does not
we are working to fix ASAP
in the meantime, you can manually set the 
apiVersion
to what it was before using the `transforms`—sorry about this.
b

bitter-dentist-28132

10/02/2019, 7:41 PM
if the delete is done before the create, that shouldn't matter though, yes? i guess all deletes are done after all creates?
c

creamy-potato-29402

10/02/2019, 7:42 PM
if you don’t specify a 
.metadata.name
then yes
we will create and then delete
if you do (as ~all charts do) then we will be foreced to delete first
in this case, we don’t know we need to delete these first, so we try to create then delete and we fail
another option is to simply delete that resource, 
pulumi refresh
, and then 
up
b

bitter-dentist-28132

10/02/2019, 7:45 PM
yes indeed. or pin.
c

creamy-potato-29402

10/02/2019, 7:45 PM
Anyway, apologies—I’ll try to get a fix in over the next day. I need to verify this works for all corner cases though.
So it might take a bit longer.
b

bitter-dentist-28132

10/02/2019, 7:49 PM
that's ok, i do feel like simply saying old === new could cause problems so i can see why you'd be hesitant to just accept it.
Powered by