diff --git a/internal-packages/run-engine/src/engine/systems/dequeueSystem.ts b/internal-packages/run-engine/src/engine/systems/dequeueSystem.ts
index 5a951a645c..e8333bcce7 100644
--- a/internal-packages/run-engine/src/engine/systems/dequeueSystem.ts
+++ b/internal-packages/run-engine/src/engine/systems/dequeueSystem.ts
@@ -1,6 +1,6 @@
 import type { BillingCache } from "../billingCache.js";
 import { startSpan } from "@internal/tracing";
-import { assertExhaustive } from "@trigger.dev/core";
+import { assertExhaustive, tryCatch } from "@trigger.dev/core";
 import { DequeuedMessage, RetryOptions } from "@trigger.dev/core/v3";
 import { placementTag } from "@trigger.dev/core/v3/serverOnly";
 import { getMaxDuration } from "@trigger.dev/core/v3/isomorphic";
@@ -611,20 +611,24 @@ export class DequeueSystem {
             }
           );
 
-          const run = await prisma.taskRun.findFirst({
-            where: { id: runId },
-            include: {
-              runtimeEnvironment: true,
-            },
-          });
+          // Wrap the Prisma call with tryCatch - if DB is unavailable, we still want to nack via Redis
+          const [findError, run] = await tryCatch(
+            prisma.taskRun.findFirst({
+              where: { id: runId },
+              include: {
+                runtimeEnvironment: true,
+              },
+            })
+          );
 
-          if (!run) {
-            //this isn't ideal because we're not creating a snapshot… but we can't do much else
+          // If DB is unavailable or run not found, just nack directly via Redis
+          if (findError || !run) {
             this.$.logger.error(
-              "RunEngine.dequeueFromWorkerQueue(): Thrown error, then run not found. Nacking.",
+              "RunEngine.dequeueFromWorkerQueue(): Failed to find run, nacking directly via Redis",
               {
                 runId,
                 orgId,
+                findError,
               }
             );
             await this.$.runQueue.nackMessage({ orgId, messageId: runId });
diff --git a/internal-packages/run-engine/src/engine/tests/dequeuing.test.ts b/internal-packages/run-engine/src/engine/tests/dequeuing.test.ts
index 8ea142f630..5871bee832 100644
--- a/internal-packages/run-engine/src/engine/tests/dequeuing.test.ts
+++ b/internal-packages/run-engine/src/engine/tests/dequeuing.test.ts
@@ -1,9 +1,10 @@
-import { containerTest } from "@internal/testcontainers";
+import { assertNonNullable, containerTest } from "@internal/testcontainers";
 import { trace } from "@internal/tracing";
 import { DequeuedMessage } from "@trigger.dev/core/v3";
 import { generateFriendlyId } from "@trigger.dev/core/v3/isomorphic";
 import { PrismaClientOrTransaction } from "@trigger.dev/database";
 import { expect } from "vitest";
+import { setTimeout } from "node:timers/promises";
 import { MinimalAuthenticatedEnvironment } from "../../shared/index.js";
 import { RunEngine } from "../index.js";
 import { setupAuthenticatedEnvironment, setupBackgroundWorker } from "./setup.js";
@@ -79,6 +80,137 @@ describe("RunEngine dequeuing", () => {
       await engine.quit();
     }
   });
+
+  containerTest(
+    "Direct nack after dequeue clears concurrency and allows recovery",
+    async ({ prisma, redisOptions }) => {
+      const authenticatedEnvironment = await setupAuthenticatedEnvironment(prisma, "PRODUCTION");
+
+      // Use a short heartbeat timeout so the stalled system recovers the run quickly
+      const pendingExecutingTimeout = 1000;
+
+      const engine = new RunEngine({
+        prisma,
+        worker: {
+          redis: redisOptions,
+          workers: 1,
+          tasksPerWorker: 10,
+          pollIntervalMs: 100,
+        },
+        queue: {
+          redis: redisOptions,
+          masterQueueConsumersDisabled: true,
+          processWorkerQueueDebounceMs: 50,
+        },
+        runLock: {
+          redis: redisOptions,
+        },
+        machines: {
+          defaultMachine: "small-1x",
+          machines: {
+            "small-1x": {
+              name: "small-1x" as const,
+              cpu: 0.5,
+              memory: 0.5,
+              centsPerMs: 0.0001,
+            },
+          },
+          baseCostInCents: 0.0005,
+        },
+        heartbeatTimeoutsMs: {
+          PENDING_EXECUTING: pendingExecutingTimeout,
+        },
+        tracer: trace.getTracer("test", "0.0.0"),
+      });
+
+      try {
+        const taskIdentifier = "test-task";
+
+        // Setup background worker
+        await setupBackgroundWorker(engine, authenticatedEnvironment, taskIdentifier);
+
+        // Trigger a single run
+        const runs = await triggerRuns({
+          engine,
+          environment: authenticatedEnvironment,
+          taskIdentifier,
+          prisma,
+          count: 1,
+        });
+        expect(runs.length).toBe(1);
+        const run = runs[0];
+
+        // Process master queue to move run to worker queue
+        await engine.runQueue.processMasterQueueForEnvironment(authenticatedEnvironment.id, 1);
+
+        // Wait for processing
+        await setTimeout(500);
+
+        // Dequeue from worker queue - this puts run in concurrency sets and creates PENDING_EXECUTING snapshot
+        const dequeued = await engine.dequeueFromWorkerQueue({
+          consumerId: "test_12345",
+          workerQueue: "main",
+        });
+        expect(dequeued.length).toBe(1);
+        assertNonNullable(dequeued[0]);
+
+        // Verify run is in PENDING_EXECUTING state
+        const executionDataBefore = await engine.getRunExecutionData({ runId: run.id });
+        assertNonNullable(executionDataBefore);
+        expect(executionDataBefore.snapshot.executionStatus).toBe("PENDING_EXECUTING");
+
+        // Verify run is in concurrency
+        const envConcurrencyBefore = await engine.runQueue.currentConcurrencyOfEnvironment(
+          authenticatedEnvironment
+        );
+        expect(envConcurrencyBefore).toBe(1);
+
+        // Simulate DB failure fallback: call nackMessage directly via Redis
+        // This is what happens when the catch block can't read from Postgres
+        await engine.runQueue.nackMessage({
+          orgId: authenticatedEnvironment.organization.id,
+          messageId: run.id,
+        });
+
+        // Verify concurrency is cleared - this is the key fix!
+        // Without this fix, the run would stay in concurrency sets forever
+        const envConcurrencyAfter = await engine.runQueue.currentConcurrencyOfEnvironment(
+          authenticatedEnvironment
+        );
+        expect(envConcurrencyAfter).toBe(0);
+
+        // Verify the message is back in the queue
+        const envQueueLength = await engine.runQueue.lengthOfEnvQueue(authenticatedEnvironment);
+        expect(envQueueLength).toBe(1);
+
+        // Wait for the stalled system to detect and recover the PENDING_EXECUTING run
+        // The stalled system will call tryNackAndRequeue which updates Postgres state to QUEUED
+        await setTimeout(pendingExecutingTimeout * 5);
+
+        // Verify the stalled system recovered the run to QUEUED state
+        const executionDataAfterStall = await engine.getRunExecutionData({ runId: run.id });
+        assertNonNullable(executionDataAfterStall);
+        expect(executionDataAfterStall.snapshot.executionStatus).toBe("QUEUED");
+
+        // Process master queue to move the run from env queue to worker queue
+        await engine.runQueue.processMasterQueueForEnvironment(authenticatedEnvironment.id, 1);
+
+        // Wait for processing
+        await setTimeout(500);
+
+        // Dequeue from worker queue - the run should now be available
+        const dequeuedAgain = await engine.dequeueFromWorkerQueue({
+          consumerId: "test_12345",
+          workerQueue: "main",
+        });
+        expect(dequeuedAgain.length).toBe(1);
+        assertNonNullable(dequeuedAgain[0]);
+        expect(dequeuedAgain[0].run.id).toBe(run.id);
+      } finally {
+        await engine.quit();
+      }
+    }
+  );
 });
 
 async function triggerRuns({