feat: cancel in progress (#325)

* chore: bad project path

* fix: remove tickerId for replay

* feat: cancel from request

* feat: cancel button in UI

* chore: rm comment

* fix: build error

* chore: reason case

* chore: reason string

* fix: linting

---------

Co-authored-by: gabriel ruttner <gabe@hatchet.run>

api-contracts/openapi/openapi.yaml

@@ -110,6 +110,8 @@ paths:
     $ref: "./paths/step-run/step-run.yaml#/stepRunScoped"
   /api/v1/tenants/{tenant}/step-runs/{step-run}/rerun:
     $ref: "./paths/step-run/step-run.yaml#/rerunStepRun"
+  /api/v1/tenants/{tenant}/step-runs/{step-run}/cancel:
+    $ref: "./paths/step-run/step-run.yaml#/cancelStepRun"
   /api/v1/tenants/{tenant}/step-runs/{step-run}/schema:
     $ref: "./paths/step-run/step-run.yaml#/getSchema"
   /api/v1/tenants/{tenant}/worker:

api-contracts/openapi/paths/step-run/step-run.yaml

@@ -104,6 +104,54 @@ rerunStepRun:
     summary: Rerun step run
     tags:
       - Step Run
+
+cancelStepRun:
+  post:
+    x-resources: ["tenant", "step-run"]
+    description: Attempts to cancel a step run
+    operationId: step-run:update:cancel
+    parameters:
+      - description: The tenant id
+        in: path
+        name: tenant
+        required: true
+        schema:
+          type: string
+          format: uuid
+          minLength: 36
+          maxLength: 36
+      - description: The step run id
+        in: path
+        name: step-run
+        required: true
+        schema:
+          type: string
+          format: uuid
+          minLength: 36
+          maxLength: 36
+    responses:
+      "200":
+        content:
+          application/json:
+            schema:
+              $ref: "../../components/schemas/_index.yaml#/StepRun"
+        description: Successfully dispatched the cancellation
+      "400":
+        content:
+          application/json:
+            schema:
+              $ref: "../../components/schemas/_index.yaml#/APIErrors"
+        description: A malformed or bad request
+      "403":
+        content:
+          application/json:
+            schema:
+              $ref: "../../components/schemas/_index.yaml#/APIErrors"
+        description: Forbidden
+    summary: Attempts to cancel a step run
+    tags:
+      - Step Run
+
 getSchema:
   get:
     x-resources: ["tenant", "step-run"]

api/v1/server/handlers/step-runs/cancel.go

@@ -0,0 +1,74 @@
+package stepruns
+
+import (
+	"fmt"
+	"time"
+
+	"github.com/labstack/echo/v4"
+
+	"github.com/hatchet-dev/hatchet/api/v1/server/oas/apierrors"
+	"github.com/hatchet-dev/hatchet/api/v1/server/oas/gen"
+	"github.com/hatchet-dev/hatchet/api/v1/server/oas/transformers"
+	"github.com/hatchet-dev/hatchet/internal/msgqueue"
+	"github.com/hatchet-dev/hatchet/internal/repository/prisma/db"
+	"github.com/hatchet-dev/hatchet/internal/services/shared/tasktypes"
+)
+
+func (t *StepRunService) StepRunUpdateCancel(ctx echo.Context, request gen.StepRunUpdateCancelRequestObject) (gen.StepRunUpdateCancelResponseObject, error) {
+	tenant := ctx.Get("tenant").(*db.TenantModel)
+	stepRun := ctx.Get("step-run").(*db.StepRunModel)
+
+	// check to see if the step run is in a running or pending state
+	status := stepRun.Status
+
+	if status == db.StepRunStatusFailed || status == db.StepRunStatusCancelled || status == db.StepRunStatusSucceeded {
+		return gen.StepRunUpdateCancel400JSONResponse(
+			apierrors.NewAPIErrors("step run is not in a running or pending state"),
+		), nil
+	}
+
+	engineStepRun, err := t.config.EngineRepository.StepRun().GetStepRunForEngine(tenant.ID, stepRun.ID)
+
+	if err != nil {
+		return nil, fmt.Errorf("could not get step run for engine: %w", err)
+	}
+
+	var reason = "CANCELLED_BY_USER"
+
+	// send a task to the taskqueue
+	err = t.config.MessageQueue.AddMessage(
+		ctx.Request().Context(),
+		msgqueue.JOB_PROCESSING_QUEUE,
+		tasktypes.StepRunCancelToTask(engineStepRun, reason),
+	)
+
+	if err != nil {
+		return nil, fmt.Errorf("could not add step queued task to task queue: %w", err)
+	}
+
+	// wait for a short period of time
+	for i := 0; i < 5; i++ {
+		newStepRun, err := t.config.APIRepository.StepRun().GetStepRunById(tenant.ID, stepRun.ID)
+
+		if err != nil {
+			return nil, fmt.Errorf("could not get step run: %w", err)
+		}
+
+		if newStepRun.Status != stepRun.Status {
+			stepRun = newStepRun
+			break
+		}
+
+		time.Sleep(100 * time.Millisecond)
+	}
+
+	res, err := transformers.ToStepRun(stepRun)
+
+	if err != nil {
+		return nil, fmt.Errorf("could not transform step run: %w", err)
+	}
+
+	return gen.StepRunUpdateCancel200JSONResponse(
+		*res,
+	), nil
+}