issue-5293: fixed caching of stale GetNodeAttr responses in case of concurrent WriteData/SetNodeAttr/AllocateData requests

cloud/filestore/libs/vfs_fuse/bench/main.cpp

@@ -0,0 +1,109 @@
+#include <cloud/filestore/libs/vfs_fuse/node_cache.h>
+
+#include <library/cpp/testing/benchmark/bench.h>
+
+#include <util/random/fast.h>
+#include <util/thread/factory.h>
+
+////////////////////////////////////////////////////////////////////////////////
+
+using namespace NCloud::NFileStore;
+using namespace NFuse;
+
+namespace {
+
+////////////////////////////////////////////////////////////////////////////////
+
+void RunBench(ui64 iters, ui32 threads, ui32 shards)
+{
+    TNodeCache cache("fs", shards);
+    std::atomic<ui64> version = 0;
+    const ui64 nodeCount = 1'000'000;
+
+    struct TContext
+    {
+        TManualEvent Ev;
+        std::atomic<ui32> Todo;
+    };
+
+    auto context = std::make_shared<TContext>();
+    context->Todo = threads;
+
+    for (ui32 i = 0; i < threads; ++i) {
+        SystemThreadFactory()->Run(
+            [&cache, &version, iters, threads, context] ()
+            {
+                TReallyFastRng32 rng(777);
+                for (size_t i = 0; i < iters / threads; ++i) {
+                    NProto::TNodeAttr attrs;
+                    attrs.SetId(rng.Uniform(nodeCount));
+                    attrs.SetType(NProto::E_REGULAR_NODE);
+                    cache.UpdateNode(
+                        attrs,
+                        version.fetch_add(1, std::memory_order_release));
+                }
+
+                context->Todo.fetch_sub(1, std::memory_order_release);
+                context->Ev.Signal();
+            });
+    }
+
+    while (context->Todo.load(std::memory_order_acquire)) {
+        context->Ev.WaitI();
+        context->Ev.Reset();
+    }
+}
+
+}   // namespace
+
+////////////////////////////////////////////////////////////////////////////////
+
+Y_CPU_BENCHMARK(UpdateNode1, iface)
+{
+    RunBench(iface.Iterations(), 1, 1);
+}
+
+Y_CPU_BENCHMARK(UpdateNode2, iface)
+{
+    RunBench(iface.Iterations(), 2, 1);
+}
+
+Y_CPU_BENCHMARK(UpdateNode4, iface)
+{
+    RunBench(iface.Iterations(), 4, 1);
+}
+
+Y_CPU_BENCHMARK(UpdateNode8, iface)
+{
+    RunBench(iface.Iterations(), 8, 1);
+}
+
+Y_CPU_BENCHMARK(UpdateNode16, iface)
+{
+    RunBench(iface.Iterations(), 16, 1);
+}
+
+Y_CPU_BENCHMARK(UpdateNode1_16, iface)
+{
+    RunBench(iface.Iterations(), 1, 16);
+}
+
+Y_CPU_BENCHMARK(UpdateNode2_16, iface)
+{
+    RunBench(iface.Iterations(), 2, 16);
+}
+
+Y_CPU_BENCHMARK(UpdateNode4_16, iface)
+{
+    RunBench(iface.Iterations(), 4, 16);
+}
+
+Y_CPU_BENCHMARK(UpdateNode8_16, iface)
+{
+    RunBench(iface.Iterations(), 8, 16);
+}
+
+Y_CPU_BENCHMARK(UpdateNode16_16, iface)
+{
+    RunBench(iface.Iterations(), 16, 16);
+}

cloud/filestore/libs/vfs_fuse/bench/ya.make

@@ -0,0 +1,16 @@
+Y_BENCHMARK()
+
+IF (SANITIZER_TYPE)
+    TAG(ya:manual)
+ENDIF()
+
+SRCS(
+    main.cpp
+)
+
+PEERDIR(
+    cloud/filestore/libs/vfs_fuse
+    cloud/filestore/libs/vfs_fuse/vhost
+)
+
+END()

cloud/filestore/libs/vfs_fuse/fs.cpp

@@ -150,12 +150,12 @@ int ReplyAttr(
     const NCloud::NProto::TError& error,
     fuse_req_t req,
     const struct stat *attr,
-    double attr_timeout)
+    double attrTimeout)
 {
     requestStats.ResponseSent(callContext);
     FILESTORE_TRACK(ResponseSent, (&callContext), "Attr");
 
-    int res = fuse_reply_attr(req, attr, attr_timeout);
+    int res = fuse_reply_attr(req, attr, attrTimeout);
     if (res != 0) {
         STORAGE_WARN(callContext.LogString()
             << " fuse_reply_attr failed with code " << res);

cloud/filestore/libs/vfs_fuse/fs_directory_content_format.cpp

@@ -0,0 +1,112 @@
+#include "fs_directory_content_format.h"
+
+#if defined(FUSE_VIRTIO)
+#   include<cloud/contrib/virtiofsd/fuse.h>
+#endif
+
+#include <util/system/align.h>
+
+namespace NCloud::NFileStore::NFuse {
+
+////////////////////////////////////////////////////////////////////////////////
+
+TDirectoryBuilder::TDirectoryBuilder(size_t size) noexcept
+    : Buffer(std::make_shared<TBuffer>(size))
+{}
+
+void TDirectoryBuilder::Add(
+    fuse_req_t req,
+    const TString& name,
+    const fuse_entry_param& entry,
+    size_t offset)
+{
+#if defined(FUSE_VIRTIO)
+    size_t entrySize = fuse_add_direntry_plus(
+        req,
+        nullptr,
+        0,
+        name.c_str(),
+        &entry,
+        0);
+
+    Buffer->Advance(entrySize);
+
+    fuse_add_direntry_plus(
+        req,
+        Buffer->Pos() - entrySize,
+        entrySize,
+        name.c_str(),
+        &entry,
+        offset + Buffer->Size());
+#else
+    size_t entrySize = fuse_add_direntry(
+        req,
+        nullptr,
+        0,
+        name.c_str(),
+        &entry.attr,
+        0);
+
+    Buffer->Advance(entrySize);
+
+    fuse_add_direntry(
+        req,
+        Buffer->Pos() - entrySize,
+        entrySize,
+        name.c_str(),
+        &entry.attr,
+        offset + Buffer->Size());
+#endif
+}
+
+TBufferPtr TDirectoryBuilder::Finish()
+{
+    return std::move(Buffer);
+}
+
+////////////////////////////////////////////////////////////////////////////////
+
+NProto::TError ResetAttrTimeout(
+    char* data,
+    ui64 len,
+    const TNodeIdVisitor& visitor)
+{
+#if defined(FUSE_VIRTIO)
+    while (len > sizeof(fuse_direntplus)) {
+        auto* de = reinterpret_cast<fuse_direntplus*>(data);
+
+        if (de->dirent.ino != de->entry_out.attr.ino) {
+            return MakeError(E_INVALID_STATE, TStringBuilder() << "ino mismatch"
+                << " in dirent and attr: " << de->dirent.ino << " != "
+                << de->entry_out.attr.ino);
+        }
+
+        if (visitor(de->dirent.ino)) {
+            de->entry_out.attr_valid = 0;
+            de->entry_out.attr_valid_nsec = 0;
+        }
+
+        const ui64 fullSize = sizeof(fuse_direntplus)
+            + AlignUp<ui64>(de->dirent.namelen, sizeof(ui64));
+
+        if (len < fullSize) {
+            return MakeError(E_INVALID_STATE, TStringBuilder() << "expected >= "
+                << fullSize << " bytes of dir content, have " << len
+                << " bytes");
+        }
+
+        len -= fullSize;
+        data += fullSize;
+    }
+#else
+    // for non-virtiofs builds we don't return attrs in listing results
+
+    Y_UNUSED(data);
+    Y_UNUSED(len);
+    Y_UNUSED(visitor);
+#endif
+
+    return MakeError(S_OK);
+}
+
+}   // namespace NCloud::NFileStore::NFuse

cloud/filestore/libs/vfs_fuse/fs_directory_content_format.h

@@ -0,0 +1,34 @@
+#pragma once
+
+#include "fs_directory_handle.h"
+
+namespace NCloud::NFileStore::NFuse {
+
+////////////////////////////////////////////////////////////////////////////////
+
+class TDirectoryBuilder
+{
+private:
+    TBufferPtr Buffer;
+
+public:
+    explicit TDirectoryBuilder(size_t size) noexcept;
+
+    void Add(
+        fuse_req_t req,
+        const TString& name,
+        const fuse_entry_param& entry,
+        size_t offset);
+
+    TBufferPtr Finish();
+};
+
+////////////////////////////////////////////////////////////////////////////////
+
+using TNodeIdVisitor = std::function<bool(ui64)>;
+NProto::TError ResetAttrTimeout(
+    char* data,
+    ui64 len,
+    const TNodeIdVisitor& visitor);
+
+}   // namespace NCloud::NFileStore::NFuse