Does anyone know if Pulumi resources have a built-in

retry

mechanism? We sometimes have resource creation fail due to environmental issues. This is usually resolved when re-running

Pulumi Up

, but we would rather just retry a single resource operation, rather than the whole stack.

Curious why you would only want to retry a single resource rather than have Pulumi continue to create the rest of the dependent resources in the stack.

In my case, the

Release

resource of the kubernetes provider sometimes fails because the CRDs are not ready yet on the cluster, so ideally we would just have that

Release

resource try to install the helm chart again.

Also interested in this, for a similar reason where the Release resource is flakey. I just have a retry in place in the CI workflow to rerun the stack

I'm not super deep with the K8s provider, but can you add

depends_on=[my_crd.ready]

? I know that some resource in that provider does have a

ready

output.

In this scenario, are there other resources in your stack would be retried that should not be? I'm asking b/c nothing that succeeded should be retried in a normal Pulumi program. Trying to figure out if there's a potential bug.

No everything else seems to be working as expected. I think this "retry" feature may also be handy with other resources, too. For example,

Command.Local

may be running a curl command that fails because of an intermittent network issue. In that case, it would be nice to tell Pulumi to only retry that resource so I don't have to run the entire stack deployment.

Ok, so use case is "make this resource more fault tolerant so I don't have to re-run pulumi up" (presumably because the preview takes longer than you want). Is that right?

BTW, you can use

command.local

to run

sleep

, which is how I solved for stuff not being ready in another codebase. I wouldn't call it ironclad, but it worked pretty reliably. Then you can make your Helm chart resource depend on the command.

That would normally work if I was installing CRDs independently from my Helm chart, but I would like the

Reelase

resource to install the required CRDs for that chart.

Is this your own chart or someone else? It's generally recommended to install CRDs independently of the chart, as helm doesn't really provide tooling for it (upgrades don't upgrade CRDs, for example).

They are 3rd party charts, like

cert-manager

and

postgres

Without retry support, I'm not sure there's a direct workaround. If it helps, you can pass a URL to a crd manifest to

ConfigYaml

, which is how we deploy cert-manager internally

Thank you for that tip!

I mean

yaml.ConfigFile

We do have an issue to add retries https://github.com/pulumi/pulumi/issues/7932 It's a little complicated and fairly niche (normally it's fine to just run

up

again) but feel free to vote and comment on that.