Merge branch 'main' into introduce-boxel-spec

.github/workflows/ci.yaml

@@ -99,6 +99,9 @@ jobs:
       matrix:
         shardIndex: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]
         shardTotal: [12]
+    concurrency:
+      group: matrix-client-test-${{ matrix.shardIndex }}-${{ github.head_ref || github.run_id }}
+      cancel-in-progress: true
     steps:
       - uses: actions/checkout@v4
       - uses: ./.github/actions/init
@@ -249,7 +252,8 @@ jobs:
             "loader-test.ts",
             "module-syntax-test.ts",
             "queue-test.ts",
-            "realm-server-test.ts",
+            "realm-endpoints-test.ts",
+            "server-endpoints-test.ts",
             "virtual-network-test.ts",
           ]
     steps:

.github/workflows/manual-deploy.yml

@@ -74,10 +74,21 @@ jobs:
     with:
       repository: "boxel-realm-server-${{ inputs.environment }}"
       environment: ${{ inputs.environment }}
-      dockerfile: "packages/realm-server/Dockerfile"
+      dockerfile: "packages/realm-server/realm-server.Dockerfile"
       build-args: |
         "realm_server_script=start:${{ inputs.environment }}"
 
+  build-worker:
+    name: Build worker Docker image
+    uses: cardstack/gh-actions/.github/workflows/docker-ecr.yml@main
+    secrets: inherit
+    with:
+      repository: "boxel-worker-${{ inputs.environment }}"
+      environment: ${{ inputs.environment }}
+      dockerfile: "packages/realm-server/worker.Dockerfile"
+      build-args: |
+        "worker_script=start:worker-${{ inputs.environment }}"
+
   build-pg-migration:
     name: Build pg-migration Docker image
     uses: cardstack/gh-actions/.github/workflows/docker-ecr.yml@main
@@ -103,16 +114,43 @@ jobs:
       image: ${{ needs.build-pg-migration.outputs.image }}
       wait-for-service-stability: false
 
+  # the wait-for-service-stability flag doesn't seem to work in
+  # aws-actions/amazon-ecs-deploy-task-definition@v2. we keep getting timeouts
+  # waiting for service stability. So we are manually waiting here.
   post-migrate-db:
     name: Wait for db-migration
     needs: [migrate-db]
     runs-on: ubuntu-latest
     steps:
-      - run: sleep 240
+      - run: sleep 180
+
+  deploy-worker:
+    name: Deploy worker
+    needs: [build-worker, deploy-host, post-migrate-db]
+    uses: cardstack/gh-actions/.github/workflows/ecs-deploy.yml@main
+    secrets: inherit
+    with:
+      container-name: "boxel-worker"
+      environment: ${{ inputs.environment }}
+      cluster: ${{ inputs.environment }}
+      service-name: "boxel-worker-${{ inputs.environment }}"
+      image: ${{ needs.build-worker.outputs.image }}
+      wait-for-service-stability: false
+
+  # the wait-for-service-stability flag doesn't seem to work in
+  # aws-actions/amazon-ecs-deploy-task-definition@v2. we keep getting timeouts
+  # waiting for service stability. So we are manually waiting here.
+  post-deploy-worker:
+    name: Wait for worker
+    needs: [deploy-worker]
+    runs-on: ubuntu-latest
+    steps:
+      - run: sleep 180
 
   deploy-realm-server:
     name: Deploy realm server
-    needs: [build-realm-server, deploy-host, post-migrate-db]
+    needs:
+      [post-deploy-worker, build-realm-server, deploy-host, post-migrate-db]
     uses: cardstack/gh-actions/.github/workflows/ecs-deploy.yml@main
     secrets: inherit
     with:

README.md

@@ -73,7 +73,7 @@ Live reloads are not available in this mode, however, if you use start the serve
 
 #### Using `start:all`
 
-Instead of running `pnpm start:base`, you can alternatively use `pnpm start:all` which also serves a few other realms on other ports--this is convenient if you wish to switch between the app and the tests without having to restart servers. Here's what is spun up with `start:all`:
+Instead of running `pnpm start:base`, you can alternatively use `pnpm start:all` which also serves a few other realms on other ports--this is convenient if you wish to switch between the app and the tests without having to restart servers. Use the environment variable `WORKER_HIGH_PRIORITY_COUNT` to add additional workers that service only user initiated requests and `WORKER_ALL_PRIORITY_COUNT` to add workers that service all jobs (system or user initiated). By default there is 1 all priority worker for each realm server. Here's what is spun up with `start:all`:
 
 | Port  | Description                                                   | Running `start:all` | Running `start:base` |
 | ----- | ------------------------------------------------------------- | ------------------- | -------------------- |
@@ -82,13 +82,17 @@ Instead of running `pnpm start:base`, you can alternatively use `pnpm start:all`
 | :4201 | `/seed` seed realm                                            | ✅                  | 🚫                   |
 | :4202 | `/test` host test realm, `/node-test` node test realm         | ✅                  | 🚫                   |
 | :4205 | `/test` realm for matrix client tests (playwright controlled) | 🚫                  | 🚫                   |
+| :4210 | Development Worker Manager (spins up 1 worker by default)     | ✅                  | 🚫                   |
+| :4211 | Test Worker Manager (spins up 1 worker by default)            | ✅                  | 🚫                   |
+| :4212 | Worker Manager for matrix client tests (playwright controlled - 1 worker) | ✅      | 🚫                   |
+| :4213 | Worker Manager for matrix client tests - base realm server (playwright controlled - 1 worker) | ✅ | 🚫    |
 | :5001 | Mail user interface for viewing emails sent to local SMTP     | ✅                  | 🚫                   |
 | :5435 | Postgres DB                                                   | ✅                  | 🚫                   |
 | :8008 | Matrix synapse server                                         | ✅                  | 🚫                   |
 
 #### Using `start:development`
 
-You can also use `start:development` if you want the functionality of `start:all`, but without running the test realms. `start:development` will enable you to open http://localhost:4201 and allow to select between the cards in the /base and /experiments realm.
+You can also use `start:development` if you want the functionality of `start:all`, but without running the test realms. `start:development` will enable you to open http://localhost:4201 and allow to select between the cards in the /base and /experiments realm. In order to use `start:development` you must also make sure to run `start:worker-development` in order to start the workers (which are normally started in `start:all`.
 
 ### Card Pre-rendering
 
@@ -119,7 +123,7 @@ When running tests we isolate the database between each test run by actually cre
 If you wish to drop the development databases you can execute:
 
 ```
-pnpm drop-all-dbs
+pnpm full-reset
 ```
 
 You can then run `pnpm migrate up` (with `PGDATABASE` set accordingly if you want to migrate a database other than `boxel`) or just start the realm server (`pnpm start:all`) to create the database again.
@@ -219,7 +223,7 @@ There is a ember-freestyle component explorer available to assist with developme
 
 1. `cd packages/boxel-ui/test-app`
 2. `pnpm start`
-3. Visit http://localhost:4210/ in your browser
+3. Visit http://localhost:4220/ in your browser
 
 ## Boxel Motion Demo App
 
@@ -286,7 +290,7 @@ To run the `packages/realm-server/` workspace tests start:
 ### Boxel UI
 
 1. `cd packages/boxel-ui/test-app`
-2. `pnpm test` (or `pnpm start` and visit http://localhost:4210/tests to run tests in the browser)
+2. `pnpm test` (or `pnpm start` and visit http://localhost:4220/tests to run tests in the browser)
 
 ### Boxel Motion
 

docs/realm.md

@@ -16,7 +16,7 @@ The capabilites of the realm are:
 
 The entry point that serves these requests `handle` function in `realm.ts` file. Depending on the `Accept` header (the recognized ones are `application/vnd.card+json`, `application/vnd.card+source`, `application/vnd.api+json`, `text/event-stream`, `text/html`) and the HTTP verb (`GET`, `PATCH`, `POST`, `DELETE` ), it will perform one of the actions listed in the above list. The routing that depends on the MIME type and HTTP method is defined in `router.ts`. There's a special case of requesting the realm root (`/`) with `GET` and `application/vnd.card+json`. This looks for a card instance at `index.json` to return.
 
-The different types of requests, together with its params, are documented in `realm-server-test.ts`.
+The different types of requests, together with its params, are documented in `realm-endpoints-test.ts`.
 
 ## Transpiling code
 

packages/ai-bot/helpers.ts

@@ -156,7 +156,10 @@ export function constructHistory(
         cardFragments,
       );
     }
-    if (event.type !== 'm.room.message') {
+    if (
+      event.type !== 'm.room.message' &&
+      event.type !== APP_BOXEL_COMMAND_RESULT_EVENT_TYPE
+    ) {
       continue;
     }
     let eventId = event.event_id!;
@@ -595,3 +598,24 @@ export function isCommandResultEvent(
     event.content['m.relates_to']?.rel_type === 'm.annotation'
   );
 }
+
+export function eventRequiresResponse(event: MatrixEvent) {
+  // If it's a message, we should respond unless it's a card fragment
+  if (event.getType() === 'm.room.message') {
+    if (event.getContent().msgtype === APP_BOXEL_CARDFRAGMENT_MSGTYPE) {
+      return false;
+    }
+    return true;
+  }
+
+  // If it's a command result with output, we should respond
+  if (
+    event.getType() === APP_BOXEL_COMMAND_RESULT_EVENT_TYPE &&
+    event.getContent().msgtype === APP_BOXEL_COMMAND_RESULT_WITH_OUTPUT_MSGTYPE
+  ) {
+    return true;
+  }
+
+  // If it's a different type, or a command result without output, we should not respond
+  return false;
+}

packages/ai-bot/main.ts

@@ -14,9 +14,9 @@ import {
   isCommandResultStatusApplied,
   getPromptParts,
   extractCardFragmentsFromEvents,
+  eventRequiresResponse,
 } from './helpers';
 import {
-  APP_BOXEL_CARDFRAGMENT_MSGTYPE,
   APP_BOXEL_ACTIVE_LLM,
   DEFAULT_LLM,
 } from '@cardstack/runtime-common/matrix-constants';
@@ -179,12 +179,9 @@ Common issues are:
         if (toStartOfTimeline) {
           return; // don't print paginated results
         }
-        if (event.getType() !== 'm.room.message') {
+        if (!eventRequiresResponse(event)) {
           return; // only print messages
         }
-        if (event.getContent().msgtype === APP_BOXEL_CARDFRAGMENT_MSGTYPE) {
-          return; // don't respond to card fragments, we just gather these in our history
-        }
 
         if (senderMatrixUserId === aiBotUserId) {
           return;

packages/ai-bot/tests/prompt-construction-test.ts

@@ -1290,7 +1290,7 @@ test('Return host result of tool call back to open ai', () => {
                 id: 'https://cardstack.com/base/SkillCard/card-editing',
                 attributes: {
                   instructions:
-                    '- If the user wants the data they see edited, AND the patchCard function is available, you MUST use the "patchCard" function to make the change.\n- If the user wants the data they see edited, AND the patchCard function is NOT available, you MUST ask the user to open the card and share it with you.\n- If you do not call patchCard, the user will not see the change.\n- You can ONLY modify cards shared with you. If there is no patchCard function or tool, then the user hasn\'t given you access.\n- NEVER tell the user to use patchCard; you should always do it for them.\n- If the user wants to search for a card instance, AND the "searchCard" function is available, you MUST use the "searchCard" function to find the card instance.\nOnly recommend one searchCard function at a time.\nIf the user wants to edit a field of a card, you can optionally use "searchCard" to help find a card instance that is compatible with the field being edited before using "patchCard" to make the change of the field.\n You MUST confirm with the user the correct choice of card instance that he intends to use based upon the results of the search.',
+                    '- If the user wants the data they see edited, AND the patchCard function is available, you MUST use the "patchCard" function to make the change.\n- If the user wants the data they see edited, AND the patchCard function is NOT available, you MUST ask the user to open the card and share it with you.\n- If you do not call patchCard, the user will not see the change.\n- You can ONLY modify cards shared with you. If there is no patchCard function or tool, then the user hasn\'t given you access.\n- NEVER tell the user to use patchCard; you should always do it for them.\n- If the user wants to search for a card instance, AND the "searchCardsByTypeAndTitle" function is available, you MUST use the "searchCardsByTypeAndTitle" function to find the card instance.\nOnly recommend one searchCardsByTypeAndTitle function at a time.\nIf the user wants to edit a field of a card, you can optionally use "searchCard" to help find a card instance that is compatible with the field being edited before using "patchCard" to make the change of the field.\n You MUST confirm with the user the correct choice of card instance that he intends to use based upon the results of the search.',
                   title: 'Card Editing',
                   description: null,
                   thumbnailURL: null,
@@ -1474,15 +1474,13 @@ test('Return host result of tool call back to open ai', () => {
           toolCall: {
             type: 'function',
             id: 'tool-call-id-1',
-            name: 'searchCard',
+            name: 'searchCardsByTypeAndTitle',
             arguments: {
               attributes: {
                 description: "Search for card instances of type 'Author'",
-                filter: {
-                  type: {
-                    module: 'http://localhost:4201/drafts/author',
-                    name: 'Author',
-                  },
+                type: {
+                  module: 'http://localhost:4201/drafts/author',
+                  name: 'Author',
                 },
               },
             },
@@ -1613,6 +1611,27 @@ test('Tools can be required to be called if done so in the last message', () =>
   });
 });
 
+test('Tools calls are connected to their results', () => {
+  const eventList: DiscreteMatrixEvent[] = JSON.parse(
+    readFileSync(
+      path.join(
+        __dirname,
+        'resources/chats/connect-tool-calls-to-results.json',
+      ),
+    ),
+  );
+
+  const { messages } = getPromptParts(eventList, '@aibot:localhost');
+  // find the message with the tool call and its id
+  // it should have the result deserialised
+  const toolCallMessage = messages.find((message) => message.role === 'tool');
+  assert.ok(toolCallMessage, 'Should have a tool call message');
+  assert.ok(
+    toolCallMessage!.content!.includes('Cloudy'),
+    'Tool call result should include "Cloudy"',
+  );
+});
+
 module('set model in prompt', () => {
   test('default active LLM must be equal to `DEFAULT_LLM`', () => {
     const eventList: DiscreteMatrixEvent[] = JSON.parse(