Hi there I am self-hosting and I'm having an issue deploying our first test bot. I'm running into an error when calling the

$deploy

function from the UI and it's returning "Invalid Reference". I think the issue is coming from how the

executableCode

value is stored. Looking at my bot in the UI and what is stored in the DB, I see this:

"executableCode": { "contentType": "text/typescript", "url": "https://medplum-storage.mydomain.com/signedcloudfronturl", "title": "index.js" } }

but looking at the deploy.test.js the bot is being created with

executableCode: { url: Binary/${binary.id} },

see https://github.com/medplum/medplum/blob/main/packages/server/src/fhir/operations/deploy.test.ts#L99 Given this difference it's pretty clear that when we call

const binary = await repo.readReference<Binary>({
      reference: (bot.executableCode as Attachment).url as string,
    });

it will fail. I'm running on server version 4.4.1. Any advice on how to fix this?

Hi Doug, just confirming: you're trying to deploy this to lambda rather than the vmcontext correct? And in the example above you've edited in the

medplum-storage.mydomain.com

it's not copy/pasted verbatim from the bot JSON right?

Yes and yes. The url in the

"url"

path opens up the

.js

file being served through cloudfront, and I have the following ENV vars setup:

{
  "name": "MEDPLUM_BOT_LAMBDA_ROLE_ARN",
  "value": "arn:aws:iam::my-role"
},
{ "name": "MEDPLUM_BOT_LAMBDA_LAYER_NAME", "value": "medplum-bot-layer" },
{ "name": "MEDPLUM_VM_CONTEXT_BOTS_ENABLED", "value": "false" },
{ "name": "MEDPLUM_DEFAULT_BOT_RUNTIME_VERSION", "value": "awslambda" }

Although this is making me think to double check the lambda role is correct. Yesterday I dealt with a completely obscure error when uploading that turned out to be a lack of S3 permissions b/c I assigned them to my EC2 role instead of my ECS Task Role

Hi @Doug DeBold ! I think the issue here is that in the medplum UI, the binary reference in the bot is resolved into a URL, but in the DB, it is still stored as a reference to the Binary resource (something like the below). I think the deploy operation is failing because it's trying to read executableCode.url as a Reference, but it's getting the resolved signed URL instead. Is there a chance you manually updated the executable code rather than re-deploying or editing in the bot editor UI?

{
  "contentType": "text/typescript",
  "url": "Binary/{binary-id}"
}

Hi Maddy, I suspected this as well and went into the DB and in the DB the

executableCode

is being stored the exact same as I'm seeing it in the UI

Ah I see, how did you deploy this bot that is failing initially? And was that deployment different than the bots that are working?

I have 2 deployments: - staging: Docker on EC2 -- bot deploys from the UI work - prod: ECS on EC2 -- no bot deploy has ever worked here

Hm interesting, I would definitely encourage users to use our CLI or generally our $deploy operation to update their bots. Something like the below could be used if you would rather use FHIR url than a CLI command

import { MedplumClient } from '@medplum/core';

const medplum = new MedplumClient();

// Deploy bot using the $deploy operation
const deployResult = await medplum.post(
  medplum.fhirUrl('Bot', botId, '$deploy'), 
  {
    code: botCode,
    filename: 'index.js'
  }
);

Yeah, I was hoping for a quick and dirty solution to get something out the door. I am now using the CLI directly from my machine and it worked and will worry about merging into our CI/CD pipeline later. On a similar note, I haven't found any good documentation on building rules for subscription criteria. Can you point me towards any?

@Doug DeBold not too much beyond the subscription extensions page https://www.medplum.com/docs/subscriptions/subscription-extensions

Thank you! I somehow missed this page