diff --git a/src/gpu/vk/GrVkOpsRenderPass.cpp b/src/gpu/vk/GrVkOpsRenderPass.cpp
index 6b2b51159a..bc01d53c24 100644
--- a/src/gpu/vk/GrVkOpsRenderPass.cpp
+++ b/src/gpu/vk/GrVkOpsRenderPass.cpp
@@ -181,6 +181,11 @@ void GrVkOpsRenderPass::submit() {
 
     // We don't want to actually submit the secondary command buffer if it is wrapped.
     if (this->wrapsSecondaryCommandBuffer()) {
+        // We pass the ownership of the GrVkSecondaryCommandBuffer to the special wrapped
+        // GrVkRenderTarget since it's lifetime matches the lifetime we need to keep the
+        // GrVkResources on the GrVkSecondaryCommandBuffer alive.
+        static_cast<GrVkRenderTarget*>(fRenderTarget)->addWrappedGrSecondaryCommandBuffer(
+                std::move(fCurrentSecondaryCommandBuffer));
         return;
     }
 
diff --git a/src/gpu/vk/GrVkRenderTarget.cpp b/src/gpu/vk/GrVkRenderTarget.cpp
index 2bfa452dfa..1494bca50b 100644
--- a/src/gpu/vk/GrVkRenderTarget.cpp
+++ b/src/gpu/vk/GrVkRenderTarget.cpp
@@ -356,6 +356,11 @@ void GrVkRenderTarget::releaseInternalObjects() {
         fCachedSimpleRenderPass->unref(gpu);
         fCachedSimpleRenderPass = nullptr;
     }
+    for (int i = 0; i < fGrSecondaryCommandBuffers.count(); ++i) {
+        SkASSERT(fGrSecondaryCommandBuffers[i]);
+        fGrSecondaryCommandBuffers[i]->releaseResources(gpu);
+    }
+    fGrSecondaryCommandBuffers.reset();
 }
 
 void GrVkRenderTarget::abandonInternalObjects() {
@@ -380,6 +385,11 @@ void GrVkRenderTarget::abandonInternalObjects() {
         fCachedSimpleRenderPass->unrefAndAbandon();
         fCachedSimpleRenderPass = nullptr;
     }
+    for (int i = 0; i < fGrSecondaryCommandBuffers.count(); ++i) {
+        SkASSERT(fGrSecondaryCommandBuffers[i]);
+        fGrSecondaryCommandBuffers[i]->abandonGPUData();
+    }
+    fGrSecondaryCommandBuffers.reset();
 }
 
 void GrVkRenderTarget::onRelease() {
diff --git a/src/gpu/vk/GrVkRenderTarget.h b/src/gpu/vk/GrVkRenderTarget.h
index 99bac20908..0a4cdc339b 100644
--- a/src/gpu/vk/GrVkRenderTarget.h
+++ b/src/gpu/vk/GrVkRenderTarget.h
@@ -13,14 +13,13 @@
 #include "src/gpu/vk/GrVkImage.h"
 
 #include "include/gpu/vk/GrVkTypes.h"
+#include "src/gpu/vk/GrVkCommandBuffer.h"
 #include "src/gpu/vk/GrVkRenderPass.h"
 #include "src/gpu/vk/GrVkResourceProvider.h"
 
-class GrVkCommandBuffer;
 class GrVkFramebuffer;
 class GrVkGpu;
 class GrVkImageView;
-class GrVkSecondaryCommandBuffer;
 class GrVkStencilAttachment;
 
 struct GrVkImageInfo;
@@ -91,6 +90,10 @@ public:
 
     void addResources(GrVkCommandBuffer& commandBuffer);
 
+    void addWrappedGrSecondaryCommandBuffer(std::unique_ptr<GrVkSecondaryCommandBuffer> cmdBuffer) {
+        fGrSecondaryCommandBuffers.push_back(std::move(cmdBuffer));
+    }
+
 protected:
     GrVkRenderTarget(GrVkGpu* gpu,
                      const GrSurfaceDesc& desc,
@@ -183,6 +186,16 @@ private:
     // VkCommandBuffer and not VK_NULL_HANDLE. In this case the render target will not be backed by
     // an actual VkImage and will thus be limited in terms of what it can be used for.
     VkCommandBuffer fSecondaryCommandBuffer = VK_NULL_HANDLE;
+    // When we wrap a secondary command buffer, we will record GrVkResources onto it which need to
+    // be kept alive till the command buffer gets submitted and the GPU has finished. However, in
+    // the wrapped case, we don't know when the command buffer gets submitted and when it is
+    // finished on the GPU since the client is in charge of that. However, we do require that the
+    // client keeps the GrVkSecondaryCBDrawContext alive and call releaseResources on it once the
+    // GPU is finished all the work. Thus we can use this to manage the lifetime of our
+    // GrVkSecondaryCommandBuffers. By storing them on the GrVkRenderTarget, which is owned by the
+    // SkGpuDevice on the GrVkSecondaryCBDrawContext, we assure that the GrVkResources held by the
+    // GrVkSecondaryCommandBuffer don't get deleted before they are allowed to.
+    SkTArray<std::unique_ptr<GrVkCommandBuffer>> fGrSecondaryCommandBuffers;
 };
 
 #endif