Limit the number of read-only files the POSIX Env will have open.

Background compaction can create an unbounded number of
leveldb::RandomAccessFile instances. On 64-bit systems mmap is used and
file descriptors are only used beyond a certain number of mmap's.
32-bit systems to not use mmap at all. leveldb::RandomAccessFile does not
observe Options.max_open_files so compaction could exhaust the file
descriptor limit.

This change uses getrlimit to determine the maximum number of open
files and limits RandomAccessFile to approximately 20% of that value.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=143505556

util/env_posix.cc

@@ -10,6 +10,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <sys/mman.h>
+#include <sys/resource.h>
 #include <sys/stat.h>
 #include <sys/time.h>
 #include <sys/types.h>
@@ -23,15 +24,70 @@
 #include "util/logging.h"
 #include "util/mutexlock.h"
 #include "util/posix_logger.h"
+#include "util/env_posix_test_helper.h"
 
 namespace leveldb {
 
 namespace {
 
+static int open_read_only_file_limit = -1;
+static int mmap_limit = -1;
+
 static Status IOError(const std::string& context, int err_number) {
   return Status::IOError(context, strerror(err_number));
 }
 
+// Helper class to limit resource usage to avoid exhaustion.
+// Currently used to limit read-only file descriptors and mmap file usage
+// so that we do not end up running out of file descriptors, virtual memory,
+// or running into kernel performance problems for very large databases.
+class Limiter {
+ public:
+  // Limit maximum number of resources to |n|.
+  Limiter(intptr_t n) {
+    SetAllowed(n);
+  }
+
+  // If another resource is available, acquire it and return true.
+  // Else return false.
+  bool Acquire() {
+    if (GetAllowed() <= 0) {
+      return false;
+    }
+    MutexLock l(&mu_);
+    intptr_t x = GetAllowed();
+    if (x <= 0) {
+      return false;
+    } else {
+      SetAllowed(x - 1);
+      return true;
+    }
+  }
+
+  // Release a resource acquired by a previous call to Acquire() that returned
+  // true.
+  void Release() {
+    MutexLock l(&mu_);
+    SetAllowed(GetAllowed() + 1);
+  }
+
+ private:
+  port::Mutex mu_;
+  port::AtomicPointer allowed_;
+
+  intptr_t GetAllowed() const {
+    return reinterpret_cast<intptr_t>(allowed_.Acquire_Load());
+  }
+
+  // REQUIRES: mu_ must be held
+  void SetAllowed(intptr_t v) {
+    allowed_.Release_Store(reinterpret_cast<void*>(v));
+  }
+
+  Limiter(const Limiter&);
+  void operator=(const Limiter&);
+};
+
 class PosixSequentialFile: public SequentialFile {
  private:
   std::string filename_;
@@ -69,73 +125,51 @@ class PosixSequentialFile: public SequentialFile {
 class PosixRandomAccessFile: public RandomAccessFile {
  private:
   std::string filename_;
+  bool temporary_fd_;  // If true, fd_ is -1 and we open on every read.
   int fd_;
+  Limiter* limiter_;
 
  public:
-  PosixRandomAccessFile(const std::string& fname, int fd)
-      : filename_(fname), fd_(fd) { }
-  virtual ~PosixRandomAccessFile() { close(fd_); }
+  PosixRandomAccessFile(const std::string& fname, int fd, Limiter* limiter)
+      : filename_(fname), fd_(fd), limiter_(limiter) {
+    temporary_fd_ = !limiter->Acquire();
+    if (temporary_fd_) {
+      // Open file on every access.
+      close(fd_);
+      fd_ = -1;
+    }
+  }
+
+  virtual ~PosixRandomAccessFile() {
+    if (!temporary_fd_) {
+      close(fd_);
+      limiter_->Release();
+    }
+  }
 
   virtual Status Read(uint64_t offset, size_t n, Slice* result,
                       char* scratch) const {
+    int fd = fd_;
+    if (temporary_fd_) {
+      fd = open(filename_.c_str(), O_RDONLY);
+      if (fd < 0) {
+        return IOError(filename_, errno);
+      }
+    }
+
     Status s;
-    ssize_t r = pread(fd_, scratch, n, static_cast<off_t>(offset));
+    ssize_t r = pread(fd, scratch, n, static_cast<off_t>(offset));
     *result = Slice(scratch, (r < 0) ? 0 : r);
     if (r < 0) {
       // An error: return a non-ok status
       s = IOError(filename_, errno);
     }
-    return s;
-  }
-};
-
-// Helper class to limit mmap file usage so that we do not end up
-// running out virtual memory or running into kernel performance
-// problems for very large databases.
-class MmapLimiter {
- public:
-  // Up to 1000 mmaps for 64-bit binaries; none for smaller pointer sizes.
-  MmapLimiter() {
-    SetAllowed(sizeof(void*) >= 8 ? 1000 : 0);
-  }
-
-  // If another mmap slot is available, acquire it and return true.
-  // Else return false.
-  bool Acquire() {
-    if (GetAllowed() <= 0) {
-      return false;
-    }
-    MutexLock l(&mu_);
-    intptr_t x = GetAllowed();
-    if (x <= 0) {
-      return false;
-    } else {
-      SetAllowed(x - 1);
-      return true;
+    if (temporary_fd_) {
+      // Close the temporary file descriptor opened earlier.
+      close(fd);
     }
+    return s;
   }
-
-  // Release a slot acquired by a previous call to Acquire() that returned true.
-  void Release() {
-    MutexLock l(&mu_);
-    SetAllowed(GetAllowed() + 1);
-  }
-
- private:
-  port::Mutex mu_;
-  port::AtomicPointer allowed_;
-
-  intptr_t GetAllowed() const {
-    return reinterpret_cast<intptr_t>(allowed_.Acquire_Load());
-  }
-
-  // REQUIRES: mu_ must be held
-  void SetAllowed(intptr_t v) {
-    allowed_.Release_Store(reinterpret_cast<void*>(v));
-  }
-
-  MmapLimiter(const MmapLimiter&);
-  void operator=(const MmapLimiter&);
 };
 
 // mmap() based random-access
@@ -144,12 +178,12 @@ class PosixMmapReadableFile: public RandomAccessFile {
   std::string filename_;
   void* mmapped_region_;
   size_t length_;
-  MmapLimiter* limiter_;
+  Limiter* limiter_;
 
  public:
   // base[0,length-1] contains the mmapped contents of the file.
   PosixMmapReadableFile(const std::string& fname, void* base, size_t length,
-                        MmapLimiter* limiter)
+                        Limiter* limiter)
       : filename_(fname), mmapped_region_(base), length_(length),
         limiter_(limiter) {
   }
@@ -332,7 +366,7 @@ class PosixEnv : public Env {
         mmap_limit_.Release();
       }
     } else {
-      *result = new PosixRandomAccessFile(fname, fd);
+      *result = new PosixRandomAccessFile(fname, fd, &fd_limit_);
     }
     return s;
   }
@@ -532,10 +566,42 @@ class PosixEnv : public Env {
   BGQueue queue_;
 
   PosixLockTable locks_;
-  MmapLimiter mmap_limit_;
+  Limiter mmap_limit_;
+  Limiter fd_limit_;
 };
 
-PosixEnv::PosixEnv() : started_bgthread_(false) {
+// Return the maximum number of concurrent mmaps.
+static int MaxMmaps() {
+  if (mmap_limit >= 0) {
+    return mmap_limit;
+  }
+  // Up to 1000 mmaps for 64-bit binaries; none for smaller pointer sizes.
+  mmap_limit = sizeof(void*) >= 8 ? 1000 : 0;
+  return mmap_limit;
+}
+
+// Return the maximum number of read-only files to keep open.
+static intptr_t MaxOpenFiles() {
+  if (open_read_only_file_limit >= 0) {
+    return open_read_only_file_limit;
+  }
+  struct rlimit rlim;
+  if (getrlimit(RLIMIT_NOFILE, &rlim)) {
+    // getrlimit failed, fallback to hard-coded default.
+    open_read_only_file_limit = 50;
+  } else if (rlim.rlim_cur == RLIM_INFINITY) {
+    open_read_only_file_limit = std::numeric_limits<int>::max();
+  } else {
+    // Allow use of 20% of available file descriptors for read-only files.
+    open_read_only_file_limit = rlim.rlim_cur / 5;
+  }
+  return open_read_only_file_limit;
+}
+
+PosixEnv::PosixEnv()
+    : started_bgthread_(false),
+      mmap_limit_(MaxMmaps()),
+      fd_limit_(MaxOpenFiles()) {
   PthreadCall("mutex_init", pthread_mutex_init(&mu_, NULL));
   PthreadCall("cvar_init", pthread_cond_init(&bgsignal_, NULL));
 }
@@ -610,6 +676,16 @@ static pthread_once_t once = PTHREAD_ONCE_INIT;
 static Env* default_env;
 static void InitDefaultEnv() { default_env = new PosixEnv; }
 
+void EnvPosixTestHelper::SetReadOnlyFDLimit(int limit) {
+  assert(default_env == NULL);
+  open_read_only_file_limit = limit;
+}
+
+void EnvPosixTestHelper::SetReadOnlyMMapLimit(int limit) {
+  assert(default_env == NULL);
+  mmap_limit = limit;
+}
+
 Env* Env::Default() {
   pthread_once(&once, InitDefaultEnv);
   return default_env;

util/env_posix_test_helper.h

@@ -0,0 +1,28 @@
+// Copyright 2017 The LevelDB Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file. See the AUTHORS file for names of contributors.
+
+#ifndef STORAGE_LEVELDB_UTIL_ENV_POSIX_TEST_HELPER_H_
+#define STORAGE_LEVELDB_UTIL_ENV_POSIX_TEST_HELPER_H_
+
+namespace leveldb {
+
+class EnvPosixTest;
+
+// A helper for the POSIX Env to facilitate testing.
+class EnvPosixTestHelper {
+ private:
+  friend class EnvPosixTest;
+
+  // Set the maximum number of read-only files that will be opened.
+  // Must be called before creating an Env.
+  static void SetReadOnlyFDLimit(int limit);
+
+  // Set the maximum number of read-only files that will be mapped via mmap.
+  // Must be called before creating an Env.
+  static void SetReadOnlyMMapLimit(int limit);
+};
+
+}  // namespace leveldb
+
+#endif  // STORAGE_LEVELDB_UTIL_ENV_POSIX_TEST_HELPER_H_

util/env_test.cc

@@ -6,10 +6,13 @@
 
 #include "port/port.h"
 #include "util/testharness.h"
+#include "util/env_posix_test_helper.h"
 
 namespace leveldb {
 
 static const int kDelayMicros = 100000;
+static const int kReadOnlyFileLimit = 4;
+static const int kMMapLimit = 4;
 
 class EnvPosixTest {
  private:
@@ -19,6 +22,11 @@ class EnvPosixTest {
  public:
   Env* env_;
   EnvPosixTest() : env_(Env::Default()) { }
+
+  static void SetFileLimits(int read_only_file_limit, int mmap_limit) {
+    EnvPosixTestHelper::SetReadOnlyFDLimit(read_only_file_limit);
+    EnvPosixTestHelper::SetReadOnlyMMapLimit(mmap_limit);
+  }
 };
 
 static void SetBool(void* ptr) {
@@ -97,8 +105,42 @@ TEST(EnvPosixTest, StartThread) {
   ASSERT_EQ(state.val, 3);
 }
 
+TEST(EnvPosixTest, TestOpenOnRead) {
+  // Write some test data to a single file that will be opened |n| times.
+  std::string test_dir;
+  ASSERT_OK(Env::Default()->GetTestDirectory(&test_dir));
+  std::string test_file = test_dir + "/open_on_read.txt";
+
+  FILE* f = fopen(test_file.c_str(), "w");
+  ASSERT_TRUE(f != NULL);
+  const char kFileData[] = "abcdefghijklmnopqrstuvwxyz";
+  fputs(kFileData, f);
+  fclose(f);
+
+  // Open test file some number above the sum of the two limits to force
+  // open-on-read behavior of POSIX Env leveldb::RandomAccessFile.
+  const int kNumFiles = kReadOnlyFileLimit + kMMapLimit + 5;
+  leveldb::RandomAccessFile* files[kNumFiles] = {0};
+  for (int i = 0; i < kNumFiles; i++) {
+    ASSERT_OK(Env::Default()->NewRandomAccessFile(test_file, &files[i]));
+  }
+  char scratch;
+  Slice read_result;
+  for (int i = 0; i < kNumFiles; i++) {
+    ASSERT_OK(files[i]->Read(i, 1, &read_result, &scratch));
+    ASSERT_EQ(kFileData[i], read_result[0]);
+  }
+  for (int i = 0; i < kNumFiles; i++) {
+    delete files[i];
+  }
+  ASSERT_OK(Env::Default()->DeleteFile(test_file));
+}
+
 }  // namespace leveldb
 
 int main(int argc, char** argv) {
+  // All tests currently run with the same read-only file limits.
+  leveldb::EnvPosixTest::SetFileLimits(leveldb::kReadOnlyFileLimit,
+                                       leveldb::kMMapLimit);
   return leveldb::test::RunAllTests();
 }