author Mike Hommey <>
Wed, 24 Jan 2018 11:18:13 +0900
changeset 455913 d7d6780f2b42e0a434803bb2a734c7b8564713a1
parent 442424 ab03b4c381d4e1768d4a2af7cec65003d89a4b12
child 460170 7f0cedfb4bd85bfe1a523168019864c9c6c0e665
permissions -rw-r--r--
Bug 1432390 - Use zstandard and requests modules instead of spawning curl | zstd in docker.load_image. r=dustin The zstd command we spawn, if available at all, might be the wrong version: zstd changed its stream format in an incompatible way at some point, and the version shipped in e.g. Ubuntu 16.04 uses the old format, while the version taskcluster relies on uses the new format. Relying on gps's zstandard library allows to ensure we use the right version. Another advantage is that we can trivially pip install it in a virtualenv if it isn't available on the system running the command. If we're ridding ourselves of the subprocess spawning for zstd, we might as well cover curl as well. Especially considering the error handling when subprocesses are involved is not trivial, such that the current error handling code is actually broken and leads to dead-lock conditions, when, for example, curl is still waiting for the python side to read data, but the python side is not reading data anymore because an exception was thrown in the tar reading loop.

/* -*-  Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2; -*- */
/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at */

#ifndef StartupCache_h_
#define StartupCache_h_

#include "nsClassHashtable.h"
#include "nsComponentManagerUtils.h"
#include "nsTArray.h"
#include "nsZipArchive.h"
#include "nsITimer.h"
#include "nsIMemoryReporter.h"
#include "nsIObserverService.h"
#include "nsIObserver.h"
#include "nsIObjectOutputStream.h"
#include "nsIOutputStream.h"
#include "nsIFile.h"
#include "mozilla/Attributes.h"
#include "mozilla/MemoryReporting.h"
#include "mozilla/UniquePtr.h"

 * The StartupCache is a persistent cache of simple key-value pairs,
 * where the keys are null-terminated c-strings and the values are
 * arbitrary data, passed as a (char*, size) tuple.
 * Clients should use the GetSingleton() static method to access the cache. It
 * will be available from the end of XPCOM init (NS_InitXPCOM3 in XPCOMInit.cpp),
 * until XPCOM shutdown begins. The GetSingleton() method will return null if the cache
 * is unavailable. The cache is only provided for libxul builds --
 * it will fail to link in non-libxul builds. The XPCOM interface is provided
 * only to allow compiled-code tests; clients should avoid using it.
 * The API provided is very simple: GetBuffer() returns a buffer that was previously
 * stored in the cache (if any), and PutBuffer() inserts a buffer into the cache.
 * GetBuffer returns a new buffer, and the caller must take ownership of it.
 * PutBuffer will assert if the client attempts to insert a buffer with the same name as
 * an existing entry. The cache makes a copy of the passed-in buffer, so client
 * retains ownership.
 * InvalidateCache() may be called if a client suspects data corruption
 * or wishes to invalidate for any other reason. This will remove all existing cache data.
 * Additionally, the static method IgnoreDiskCache() can be called if it is
 * believed that the on-disk cache file is itself corrupt. This call implicitly
 * calls InvalidateCache (if the singleton has been initialized) to ensure any
 * data already read from disk is discarded. The cache will not load data from
 * the disk file until a successful write occurs.
 * Finally, getDebugObjectOutputStream() allows debug code to wrap an objectstream
 * with a debug objectstream, to check for multiply-referenced objects. These will
 * generally fail to deserialize correctly, unless they are stateless singletons or the
 * client maintains their own object data map for deserialization.
 * Writes before the final-ui-startup notification are placed in an intermediate
 * cache in memory, then written out to disk at a later time, to get writes off the
 * startup path. In any case, clients should not rely on being able to GetBuffer()
 * data that is written to the cache, since it may not have been written to disk or
 * another client may have invalidated the cache. In other words, it should be used as
 * a cache only, and not a reliable persistent store.
 * Some utility functions are provided in StartupCacheUtils. These functions wrap the
 * buffers into object streams, which may be useful for serializing objects. Note
 * the above caution about multiply-referenced objects, though -- the streams are just
 * as 'dumb' as the underlying buffers about multiply-referenced objects. They just
 * provide some convenience in writing out data.

namespace mozilla {

namespace scache {

struct CacheEntry
  UniquePtr<char[]> data;
  uint32_t size;

  CacheEntry() : size(0) { }

  // Takes possession of buf
  CacheEntry(UniquePtr<char[]> buf, uint32_t len) : data(Move(buf)), size(len) { }


  size_t SizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) {
    return mallocSizeOf(this) + mallocSizeOf(data.get());

// We don't want to refcount StartupCache, and ObserverService wants to
// refcount its listeners, so we'll let it refcount this instead.
class StartupCacheListener final : public nsIObserver
  ~StartupCacheListener() {}

class StartupCache : public nsIMemoryReporter

friend class StartupCacheListener;


  // StartupCache methods. See above comments for a more detailed description.

  // Returns a buffer that was previously stored, caller takes ownership.
  nsresult GetBuffer(const char* id, UniquePtr<char[]>* outbuf, uint32_t* length);

  // Stores a buffer. Caller keeps ownership, we make a copy.
  nsresult PutBuffer(const char* id, const char* inbuf, uint32_t length);

  // Removes the cache file.
  void InvalidateCache();

  // Signal that data should not be loaded from the cache file
  static void IgnoreDiskCache();

  // In DEBUG builds, returns a stream that will attempt to check for
  // and disallow multiple writes of the same object.
  nsresult GetDebugObjectOutputStream(nsIObjectOutputStream* aStream,
                                      nsIObjectOutputStream** outStream);

  static StartupCache* GetSingleton();
  static void DeleteSingleton();

  // This measures all the heap memory used by the StartupCache, i.e. it
  // excludes the mapping.
  size_t HeapSizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const;

  size_t SizeOfMapping();

  nsresult ResetStartupWriteTimer();
  bool StartupWriteComplete();
  virtual ~StartupCache();

  nsresult LoadArchive();
  nsresult Init();
  void WriteToDisk();
  void WaitOnWriteThread();

  static nsresult InitSingleton();
  static void WriteTimeout(nsITimer *aTimer, void *aClosure);
  static void ThreadedWrite(void *aClosure);

  nsClassHashtable<nsCStringHashKey, CacheEntry> mTable;
  nsTArray<nsCString> mPendingWrites;
  RefPtr<nsZipArchive> mArchive;
  nsCOMPtr<nsIFile> mFile;

  nsCOMPtr<nsIObserverService> mObserverService;
  RefPtr<StartupCacheListener> mListener;
  nsCOMPtr<nsITimer> mTimer;

  bool mStartupWriteInitiated;

  static StaticRefPtr<StartupCache> gStartupCache;
  static bool gShutdownInitiated;
  static bool gIgnoreDiskCache;
  PRThread *mWriteThread;
#ifdef DEBUG
  nsTHashtable<nsISupportsHashKey> mWriteObjectMap;

// This debug outputstream attempts to detect if clients are writing multiple
// references to the same object. We only support that if that object
// is a singleton.
#ifdef DEBUG
class StartupCacheDebugOutputStream final
  : public nsIObjectOutputStream
  ~StartupCacheDebugOutputStream() {}


  StartupCacheDebugOutputStream (nsIObjectOutputStream* binaryStream,
                                   nsTHashtable<nsISupportsHashKey>* objectMap)
  : mBinaryStream(binaryStream), mObjectMap(objectMap) { }


  bool CheckReferences(nsISupports* aObject);

  nsCOMPtr<nsIObjectOutputStream> mBinaryStream;
  nsTHashtable<nsISupportsHashKey> *mObjectMap;
#endif // DEBUG

} // namespace scache
} // namespace mozilla

#endif //StartupCache_h_