pagespeed/system/apr_mem_cache_test.cc

/*
 * Copyright 2012 Google Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

// Author: jmarantz@google.com (Joshua Marantz)

// Unit-test the memcache interface.

#include "pagespeed/system/apr_mem_cache.h"

#include <unistd.h>
#include <cstdio>
#include <cstdlib>
#include <cstddef>

#include "apr_network_io.h"  // NOLINT
#include "apr_pools.h"  // NOLINT
#include "base/logging.h"
#include "pagespeed/kernel/base/google_message_handler.h"
#include "pagespeed/kernel/base/gtest.h"
#include "pagespeed/kernel/base/hasher.h"
#include "pagespeed/kernel/base/null_mutex.h"
#include "pagespeed/kernel/base/md5_hasher.h"
#include "pagespeed/kernel/base/mock_hasher.h"
#include "pagespeed/kernel/base/mock_timer.h"
#include "pagespeed/kernel/base/scoped_ptr.h"
#include "pagespeed/kernel/base/stack_buffer.h"
#include "pagespeed/kernel/base/statistics.h"
#include "pagespeed/kernel/base/string.h"
#include "pagespeed/kernel/base/string_util.h"
#include "pagespeed/kernel/base/thread_system.h"
#include "pagespeed/kernel/base/posix_timer.h"
#include "pagespeed/kernel/cache/cache_key_prepender.h"
#include "pagespeed/kernel/cache/cache_spammer.h"
#include "pagespeed/kernel/cache/cache_test_base.h"
#include "pagespeed/kernel/cache/fallback_cache.h"
#include "pagespeed/kernel/cache/lru_cache.h"
#include "pagespeed/kernel/thread/blocking_callback.h"
#include "pagespeed/kernel/util/platform.h"
#include "pagespeed/kernel/util/simple_stats.h"
#include "pagespeed/system/external_server_spec.h"
#include "pagespeed/system/tcp_server_thread_for_testing.h"

namespace net_instaweb {

namespace {

const int kTestValueSizeThreshold = 200;
const size_t kLRUCacheSize = 3 * kTestValueSizeThreshold;
const size_t kJustUnderThreshold = kTestValueSizeThreshold - 100;
const size_t kLargeWriteSize = kTestValueSizeThreshold + 1;
const size_t kHugeWriteSize = 2 * kTestValueSizeThreshold;

}  // namespace

class AprMemCacheTest : public CacheTestBase {
 protected:
  AprMemCacheTest()
      : timer_(new NullMutex, MockTimer::kApr_5_2010_ms),
        lru_cache_(new LRUCache(kLRUCacheSize)),
        thread_system_(Platform::CreateThreadSystem()),
        statistics_(thread_system_.get()) {
    AprMemCache::InitStats(&statistics_);
  }

  static void SetUpTestCase() {
    apr_initialize();
    atexit(apr_terminate);
    TcpServerThreadForTesting::PickListenPortOnce(&fake_memcache_listen_port_);
  }

  // Establishes a connection to a memcached instance; either one
  // on localhost:$MEMCACHED_PORT or, if non-empty, the one in
  // cluster_spec_.
  bool ConnectToMemcached(bool use_md5_hasher) {
    // See install/run_program_with_memcached.sh where this environment
    // variable is established during development testing flows.
    if (cluster_spec_.empty()) {
      const char* kPortString = getenv("MEMCACHED_PORT");
      int port;
      if (kPortString == nullptr || !StringToInt(kPortString, &port)) {
        LOG(ERROR) << "AprMemCache tests are skipped because env var "
                   << "$MEMCACHED_PORT is not set.  Set that to the port "
                   << "number where memcached is running to enable the "
                   << "tests. ALL DATA ON THE SERVER WILL BE ERASED! See "
                   << "install/run_program_with_memcached.sh as a way to "
                   << "run separate instance of memcached for testing.";
        // Does not fail the test.
        return false;
      }
      cluster_spec_.servers = { ExternalServerSpec("localhost", port) };
    }
    Hasher* hasher = &mock_hasher_;
    if (use_md5_hasher) {
      hasher = &md5_hasher_;
    }
    servers_.reset(new AprMemCache(cluster_spec_, 5, hasher, &statistics_,
                                   &timer_, &handler_));
    // As memcached is not restarted between tests, we need some other kind of
    // isolation. One option would be to flush memcached, if apr_memcache
    // supported that. We do not want to modify our fork even further, so we
    // prepend the test name and current time to all keys that go to memcached.
    const ::testing::TestInfo* const test_info =
        ::testing::UnitTest::GetInstance()->current_test_info();
    PosixTimer timer;
    const GoogleString memcache_prefix =
        StrCat(test_info->test_case_name(), ".",
               test_info->name(), "_",
               Integer64ToString(timer.NowUs()), "_");
    prefixed_memcache_.reset(
        new CacheKeyPrepender(memcache_prefix, servers_.get()));

    cache_.reset(new FallbackCache(prefixed_memcache_.get(), lru_cache_.get(),
                                   kTestValueSizeThreshold,
                                   &handler_));

    // apr_memcache actually lazy-connects to memcached, it seems, so
    // if we fail the Connect call then something is truly broken.  To
    // make sure memcached is actually up, we have to make an API
    // call, such as GetStatus.
    GoogleString buf;
    return servers_->Connect() && servers_->GetStatus(&buf);
  }

  // Attempts to initialize the connection to memcached.  It reports a
  // test failure if there is a memcached configuration specified in
  // cluster_spec_ or via $MEMCACHED_PORT, but we fail to connect to it.
  //
  // Consider three scenarios:
  //
  //   Scenario                                Test-status     Return-value
  //   --------------------------------------------------------------------
  //   cluster_spec_ empty                      OK              false
  //   cluster_spec_ non-empty, memcached ok    OK              true
  //   cluster_spec_ non-empty, memcached fail  FAILURE         false
  //
  // This helps developers ensure that the memcached interface works, without
  // requiring people who build & run tests to start up memcached.
  bool InitMemcachedOrSkip(bool use_md5_hasher) {
    bool initialized = ConnectToMemcached(use_md5_hasher);
    EXPECT_TRUE(initialized || cluster_spec_.empty())
        << "Please start memcached on " << cluster_spec_.ToString();
    return initialized;
  }

  virtual CacheInterface* Cache() { return cache_.get(); }

  GoogleMessageHandler handler_;
  MD5Hasher md5_hasher_;
  MockHasher mock_hasher_;
  MockTimer timer_;
  scoped_ptr<LRUCache> lru_cache_;
  scoped_ptr<AprMemCache> servers_;
  scoped_ptr<CacheKeyPrepender> prefixed_memcache_;
  scoped_ptr<FallbackCache> cache_;
  scoped_ptr<ThreadSystem> thread_system_;
  SimpleStats statistics_;
  ExternalClusterSpec cluster_spec_;
  static apr_port_t fake_memcache_listen_port_;
};

apr_port_t AprMemCacheTest::fake_memcache_listen_port_ = 0;

// Simple flow of putting in an item, getting it, deleting it.
TEST_F(AprMemCacheTest, PutGetDelete) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  CheckPut("Name", "Value");
  CheckGet("Name", "Value");
  CheckNotFound("Another Name");

  CheckPut("Name", "NewValue");
  CheckGet("Name", "NewValue");

  cache_->Delete("Name");
  CheckNotFound("Name");
  EXPECT_EQ(0, lru_cache_->size_bytes()) << "fallback not used.";
}

TEST_F(AprMemCacheTest, MultiGet) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }
  TestMultiGet();
  EXPECT_EQ(0, lru_cache_->size_bytes()) << "fallback not used.";
}

TEST_F(AprMemCacheTest, MultiGetWithoutServer) {
  cluster_spec_.servers = {ExternalServerSpec("localhost", 99999)};
  ASSERT_FALSE(ConnectToMemcached(true)) << "localhost:99999 should not exist";

  Callback* n0 = AddCallback();
  Callback* not_found = AddCallback();
  Callback* n1 = AddCallback();
  IssueMultiGet(n0, "n0", not_found, "not_found", n1, "n1");
  WaitAndCheckNotFound(n0);
  WaitAndCheckNotFound(not_found);
  WaitAndCheckNotFound(n1);
}

TEST_F(AprMemCacheTest, BasicInvalid) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  // Check that we honor callback veto on validity.
  CheckPut("nameA", "valueA");
  CheckPut("nameB", "valueB");
  CheckGet("nameA", "valueA");
  CheckGet("nameB", "valueB");
  set_invalid_value("valueA");
  CheckNotFound("nameA");
  CheckGet("nameB", "valueB");
  EXPECT_EQ(0, lru_cache_->size_bytes()) << "fallback not used.";
}

TEST_F(AprMemCacheTest, SizeTest) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  for (int x = 0; x < 10; ++x) {
    for (int i = kJustUnderThreshold/2; i < kJustUnderThreshold - 10; ++i) {
      GoogleString value(i, 'a');
      GoogleString key = StrCat("big", IntegerToString(i));
      CheckPut(key, value);
      CheckGet(key, value);
    }
  }
  EXPECT_EQ(0, lru_cache_->size_bytes()) << "fallback not used.";
}

TEST_F(AprMemCacheTest, StatsTest) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  GoogleString buf;
  ASSERT_TRUE(servers_->GetStatus(&buf));
  EXPECT_TRUE(buf.find("memcached server localhost:") != GoogleString::npos);
  EXPECT_TRUE(buf.find(" pid ") != GoogleString::npos);
  EXPECT_TRUE(buf.find("\nbytes_read: ") != GoogleString::npos);
  EXPECT_TRUE(buf.find("\ncurr_connections: ") != GoogleString::npos);
  EXPECT_TRUE(buf.find("\ntotal_items: ") != GoogleString::npos);
  EXPECT_EQ(0, lru_cache_->size_bytes()) << "fallback not used.";
}

TEST_F(AprMemCacheTest, HashCollision) {
  if (!InitMemcachedOrSkip(false)) {
    return;
  }
  CheckPut("N1", "V1");
  CheckGet("N1", "V1");

  // Since we are using a mock hasher, which always returns "0", the
  // put on "N2" will overwrite "N1" in memcached due to hash
  // collision.
  CheckPut("N2", "V2");
  CheckGet("N2", "V2");
  CheckNotFound("N1");
  EXPECT_EQ(0, lru_cache_->size_bytes()) << "fallback not used.";
}

TEST_F(AprMemCacheTest, JustUnderThreshold) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }
  const GoogleString kValue(kJustUnderThreshold, 'a');
  const char kKey[] = "just_under_threshold";
  CheckPut(kKey, kValue);
  CheckGet(kKey, kValue);
  EXPECT_EQ(0, lru_cache_->size_bytes()) << "fallback not used.";
}

// Basic operation with huge values, only one of which will fit
// in the fallback cache at a time.
TEST_F(AprMemCacheTest, HugeValue) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }
  const GoogleString kValue(kHugeWriteSize, 'a');
  const char kKey1[] = "large1";
  CheckPut(kKey1, kValue);
  CheckGet(kKey1, kValue);
  EXPECT_LE(kHugeWriteSize, lru_cache_->size_bytes());

  // Now put in another large value, causing the 1st to get evicted from
  // the fallback cache.
  const char kKey2[] = "large2";
  CheckPut(kKey2, kValue);
  CheckGet(kKey2, kValue);
  CheckNotFound(kKey1);

  // Finally, delete the second value explicitly.  Note that value will be
  // in the fallback cache, but we will not be able to get to it because
  // we've removed the sentinel from memcached.
  CheckGet(kKey2, kValue);
  cache_->Delete(kKey2);
  CheckNotFound(kKey2);
}

TEST_F(AprMemCacheTest, LargeValueMultiGet) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }
  const GoogleString kLargeValue1(kLargeWriteSize, 'a');
  const char kKey1[] = "large1";
  CheckPut(kKey1, kLargeValue1);
  CheckGet(kKey1, kLargeValue1);
  EXPECT_EQ(kLargeWriteSize + STATIC_STRLEN(kKey1),
            lru_cache_->size_bytes());

  const char kSmallKey[] = "small";
  const char kSmallValue[] = "value";
  CheckPut(kSmallKey, kSmallValue);

  const GoogleString kLargeValue2(kLargeWriteSize, 'b');
  const char kKey2[] = "large2";
  CheckPut(kKey2, kLargeValue2);
  CheckGet(kKey2, kLargeValue2);
  EXPECT_LE(2 * kLargeWriteSize, lru_cache_->size_bytes())
      << "Checks that both large values were written to the fallback cache";

  Callback* large1 = AddCallback();
  Callback* small = AddCallback();
  Callback* large2 = AddCallback();
  IssueMultiGet(large1, kKey1, small, kSmallKey, large2, kKey2);
  WaitAndCheck(large1, kLargeValue1);
  WaitAndCheck(small, "value");
  WaitAndCheck(large2, kLargeValue2);
}

TEST_F(AprMemCacheTest, MultiServerFallback) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  // Make another connection to the same memcached, but with a different
  // fallback cache.
  LRUCache lru_cache2(kLRUCacheSize);
  FallbackCache mem_cache2(prefixed_memcache_.get(), &lru_cache2,
                           kTestValueSizeThreshold,
                           &handler_);

  // Now when we store a large object from server1, and fetch it from
  // server2, we will get a miss because they do not share fallback caches..
  // But then we can re-store it and fetch it from either server.
  const GoogleString kLargeValue(kLargeWriteSize, 'a');
  const char kKey1[] = "large1";
  CheckPut(kKey1, kLargeValue);
  CheckGet(kKey1, kLargeValue);

  // The fallback caches are not shared, so we get a miss from mem_cache2.
  CheckNotFound(&mem_cache2, kKey1);

  CheckPut(&mem_cache2, kKey1, kLargeValue);
  CheckGet(&mem_cache2, kKey1, kLargeValue);
  CheckGet(cache_.get(), kKey1, kLargeValue);
}

TEST_F(AprMemCacheTest, KeyOver64kDropped) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  // We set our testing byte thresholds too low to trigger the case where
  // the key-value encoding fails, so make an alternate fallback cache
  // with a threshold over 64k.

  // Make another connection to the same memcached, but with a different
  // fallback cache.
  const int kBigLruSize = 1000000;
  const int kThreshold =  200000;    // fits key and small value.
  LRUCache lru_cache2(kLRUCacheSize);
  FallbackCache mem_cache2(prefixed_memcache_.get(), &lru_cache2,
                           kThreshold, &handler_);

  const GoogleString kKey(kBigLruSize, 'a');
  CheckPut(&mem_cache2, kKey, "value");
  CheckNotFound(&mem_cache2, kKey.c_str());
}

// Even keys that are over the *value* threshold can be stored in and
// retrieved from the fallback cache.  This is because we don't even
// store the key in memcached.
//
// Note: we do not expect to see ridiculously large keys; we are just
// testing for corner cases here.
TEST_F(AprMemCacheTest, LargeKeyOverThreshold) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  const GoogleString kKey(kLargeWriteSize, 'a');
  const char kValue[] = "value";
  CheckPut(kKey, kValue);
  CheckGet(kKey, kValue);
  EXPECT_EQ(kKey.size() + STATIC_STRLEN(kValue), lru_cache_->size_bytes());
}

TEST_F(AprMemCacheTest, HealthCheck) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  const int kNumIters = 5;  // Arbitrary number of repetitions.
  for (int i = 0; i < kNumIters; ++i) {
    for (int j = 0; j < AprMemCache::kMaxErrorBurst; ++j) {
      EXPECT_TRUE(servers_->IsHealthy());
      servers_->RecordError();
    }
    EXPECT_FALSE(servers_->IsHealthy());
    timer_.AdvanceMs(AprMemCache::kHealthCheckpointIntervalMs - 1);
    EXPECT_FALSE(servers_->IsHealthy());
    timer_.AdvanceMs(2);
  }
  EXPECT_TRUE(servers_->IsHealthy());
}

TEST_F(AprMemCacheTest, ThreadSafe) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  GoogleString large_pattern(kLargeWriteSize, 'a');
  large_pattern += "%d";
  CacheSpammer::RunTests(5 /* num_threads */,
                         200 /* num_iters */,
                         10 /* num_inserts */,
                         false, true, large_pattern.c_str(),
                         prefixed_memcache_.get(),
                         thread_system_.get());
}

// Tests that a very low timeout out value causes a simple Get to fail.
// Warning: if this turns out to be flaky then just delete it; it will
// have served its purpose.
//
// Update 12/9/12: this test is flaky on slow machines.  This test should
// only be run interactively to check on timeout behavior.  To run it,
// set environemnt variable (APR_MEMCACHE_TIMEOUT_TEST).
TEST_F(AprMemCacheTest, OneMicrosecondGet) {
  if (getenv("APR_MEMCACHE_TIMEOUT_TEST") == NULL) {
    LOG(WARNING)
        << "Skipping flaky test AprMemCacheTest.OneMicrosecond, set "
        << "$APR_MEMCACHE_TIMEOUT_TEST to run it";
    return;
  }

  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  // With the default timeout, do a Put, which will work.
  CheckPut("Name", "Value");
  CheckGet("Name", "Value");

  // Set the timeout insanely low and now watch the fetch fail.
  servers_->set_timeout_us(1);
  CheckNotFound("Name");
  EXPECT_EQ(1, statistics_.GetVariable("memcache_timeouts")->Get());
}

TEST_F(AprMemCacheTest, OneMicrosecondPut) {
  if (getenv("APR_MEMCACHE_TIMEOUT_TEST") == NULL) {
    LOG(WARNING)
        << "Skipping flaky test AprMemCacheTest.OneMicrosecond, set "
        << "$APR_MEMCACHE_TIMEOUT_TEST to run it";
    return;
  }

  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  // With the default timeout, do a Put, which will work.
  CheckPut("Name", "Value");
  CheckGet("Name", "Value");

  // Set the timeout insanely low and now watch the fetch fail.
  servers_->set_timeout_us(1);
  CheckPut("Name", "Value");
  EXPECT_EQ(1, statistics_.GetVariable("memcache_timeouts")->Get());
}

TEST_F(AprMemCacheTest, OneMicrosecondDelete) {
  if (getenv("APR_MEMCACHE_TIMEOUT_TEST") == NULL) {
    LOG(WARNING)
        << "Skipping flaky test AprMemCacheTest.OneMicrosecond, set "
        << "$APR_MEMCACHE_TIMEOUT_TEST to run it";
    return;
  }

  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  // With the default timeout, do a Put, which will work.
  CheckPut("Name", "Value");
  CheckGet("Name", "Value");

  // Set the timeout insanely low and now watch the fetch fail.
  servers_->set_timeout_us(1);
  CheckDelete("Name");
  EXPECT_EQ(1, statistics_.GetVariable("memcache_timeouts")->Get());
}

// Two following tests are identical and ensure that no keys are leaked between
// tests through shared running Memcached server.
TEST_F(AprMemCacheTest, TestsAreIsolated1) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  CheckNotFound("SomeKey");
  CheckPut("SomeKey", "SomeValue");
}

TEST_F(AprMemCacheTest, TestsAreIsolated2) {
  if (!InitMemcachedOrSkip(true)) {
    return;
  }

  CheckNotFound("SomeKey");
  CheckPut("SomeKey", "SomeValue");
}

namespace {
// Used in HangingMultigetTest.
class FakeMemcacheServerThread : public TcpServerThreadForTesting {
 public:
  FakeMemcacheServerThread(apr_port_t fake_memcache_listen_port,
                           ThreadSystem* thread_system)
      : TcpServerThreadForTesting(fake_memcache_listen_port, "fake_memcache",
                                  thread_system) {}

  virtual ~FakeMemcacheServerThread() { ShutDown(); }

 private:
  void HandleClientConnection(apr_socket_t* sock) override {
    static const char kMessage[] = "blah\n";
    apr_size_t message_size = STATIC_STRLEN(kMessage);
    char buf[kStackBufferSize];
    apr_size_t size = sizeof(buf) - 1;
    apr_socket_recv(sock, buf, &size);
    apr_socket_send(sock, kMessage, &message_size);
    apr_socket_close(sock);
  }
};
}  // namespace

TEST_F(AprMemCacheTest, HangingMultigetTest) {
  // Test that we do not hang in the case of corrupted responses from memcached,
  // as seen in bug report 1048
  // https://github.com/pagespeed/mod_pagespeed/issues/1048
  scoped_ptr<FakeMemcacheServerThread> thread(new FakeMemcacheServerThread(
      fake_memcache_listen_port_, thread_system_.get()));
  ASSERT_TRUE(thread->Start());
  apr_port_t port = thread->GetListeningPort();
  ExternalClusterSpec spec;
  spec.servers = {ExternalServerSpec("localhost", port)};
  scoped_ptr<AprMemCache> cache(
      new AprMemCache(spec, 3 /* maximal number of client connections */,
                      &mock_hasher_, &statistics_, &timer_, &handler_));
  static const char k1[] = "hello";
  static const char k2[] = "hi";
  BlockingCallback cb1(thread_system_.get());
  BlockingCallback cb2(thread_system_.get());
  CacheInterface::MultiGetRequest* request =
      new CacheInterface::MultiGetRequest;  // will be owned by MultiGet()
  request->push_back(CacheInterface::KeyCallback(k1, &cb1));
  request->push_back(CacheInterface::KeyCallback(k2, &cb2));
  cache->Connect();
  // Capture stderr, make sure we get the proper string.
  // This test depends on a custom fprintf in apr_memcache2.
  // TODO(jcrowell) do this more nicely, don't depend on the print from
  // multiget, as the real test is that this should not hang.
  int stderr_backup;
  char buffer[4096];
  fflush(stderr);
  int err_pipe[2];
  stderr_backup = dup(STDERR_FILENO);
  ASSERT_NE(-1, stderr_backup);
  ASSERT_EQ(0, pipe(err_pipe));
  ASSERT_NE(-1, dup2(err_pipe[1], STDERR_FILENO));
  close(err_pipe[1]);
  // Make the multiget request.
  cache->MultiGet(request);
  cb1.Block();
  cb2.Block();
  fflush(stderr);
  int bytes_read = read(err_pipe[0], buffer, sizeof(buffer));
  ASSERT_NE(-1, bytes_read);
  // And give back stderr.
  ASSERT_NE(-1, dup2(stderr_backup, STDERR_FILENO));
  // Now check to make sure that we had the proper output.
  StringPiece output(buffer, bytes_read);
  EXPECT_TRUE(
      strings::StartsWith(
          output, "Caught potential spin in apr_memcache multiget!"))
      << output;
}


}  // namespace net_instaweb