commit d529d12ded4c70e57a7cda59e2d089366098b6cc
author Chase Phillips <cmp@chromium.org> Thu Feb 21 20:37:24 2019
committer Commit Bot <commit-bot@chromium.org> Thu Feb 21 20:37:24 2019
tree c05390646fbbbb279e6e484668ffee549ca6b2d5
parent 1c9ff1df8ff498821171ca6a9378e065d0ed6ade

Blob storage: Fix ChromeBlobStorageContext deleter to use DeleteOnIOThread

The original one-off deleter for ChromeBlobStorageContext attempts to
delete |this| directly when not called on the IO thread and the IO
thread isn't available.  This causes problems with objects it owns
since some objects require being deleted on the IO thread.

This manifested in flaky tests like EmptyBlob and Bug84933Test which
started failing in a flaky way when IndexedDBDispatcherHost (which holds
a scoped_refptr<ChromeBlobStorageContext>) was moved from the IO thread
to the IDB task runner.  The flaky crashes looked like:

[8496:100:0221/084948.427:FATAL:memory_dump_manager.cc(247)] Check failed:
(*mdp_iter)->task_runner &&
(*mdp_iter)->task_runner->RunsTasksInCurrentSequence().
MemoryDumpProvider "BlobStorageContext" attempted to unregister itself
in a racy way. Please file a crbug.
Backtrace:
	base::debug::CollectStackTrace [0x00007FF6F62CCAA2+18]
	base::debug::StackTrace::StackTrace [0x00007FF6F62CC002+18]
	logging::LogMessage::~LogMessage [0x00007FF6F62E7885+101]
	base::trace_event::MemoryDumpManager::UnregisterDumpProviderInternal [0x00007FF6F634BC2B+827]
	storage::BlobStorageContext::~BlobStorageContext [0x00007FF6F61D7362+34]
	storage::BlobStorageContext::`scalar deleting destructor' [0x00007FF6F61DA8F0+16]
	content::ChromeBlobStorageContext::`scalar deleting destructor' [0x00007FF6F46EF4F8+40]
	content::IndexedDBDispatcherHost::~IndexedDBDispatcherHost [0x00007FF6F48FC6BE+382]
	content::IndexedDBDispatcherHost::`scalar deleting destructor' [0x00007FF6F48FDFA0+16]

Remove the one-off deleter for ChromeBlobStorageContext in favor of
DeleteOnIOThread which always tries to delete the object on the IO
thread and, if the IO thread isn't available, gives up since that
indicates a leak on shutdown, which is generally okay in circumstances
like this.

Bug: 717812, 934243, 934250
Change-Id: Ibfda4b0b8f17d3eeb95773bb2a75f0e9a16e84b5
Reviewed-on: https://chromium-review.googlesource.com/c/1481597
Reviewed-by: Daniel Murphy <dmurph@chromium.org>
Reviewed-by: Marijn Kruisselbrink <mek@chromium.org>
Commit-Queue: Chase Phillips <cmp@chromium.org>
Cr-Commit-Position: refs/heads/master@{#634328}