Uploaded https://codereview.chromium.org/2702083002/, which adds a StackTrace member to the URLRequestContextGetter base, to Print() when we hit the leak condition in Debug builds.

Does it actually matter that these leak on shutdown?

Adding a StackTrace is only really valuable if we plan on "fixing" the clients. 

Which in itself is not worthwhile unless if we prohibit leaks in the future.

FWIW, they seem to leak mid-session in some cases, not just during shutdown.

Agreed, though, that there's no point fixing leaks if we don't add stronger
checks to avoid regressions in future.

Thanks wez.

A mid-session leak is more of a problem, and may be worth looking at.

That sounds symptomatic of clients spinning up a temporary thread for the networking and then stopping it (and in the process leaking the context getter because it was referenced after the thread stop)

I can't seem to repro the mid-session leak right now, so I just grabbed the shutdown leak allocation stack:

[25027:25027:0222/132733.482186:WARNING:url_request_context_getter.cc(44)] URLRequestContextGetter leaking due to no owning thread. Created at: #0 0x7f539b9e876b base::debug::StackTrace::StackTrace()
#1 0x7f539b9e6dac base::debug::StackTrace::StackTrace()
#2 0x7f539b39c37e net::URLRequestContextGetter::URLRequestContextGetter()
#3 0x7f539e03503f ChromeURLRequestContextGetter::ChromeURLRequestContextGetter()
#4 0x7f539e0357db ChromeURLRequestContextGetter::Create()
#5 0x7f539e10c493 ProfileImplIOData::Handle::CreateMainRequestContextGetter()
#6 0x7f539e107389 ProfileImpl::CreateRequestContext()
#7 0x7f5395a01f42 content::StoragePartitionImplMap::Get()
#8 0x7f5394e232ff content::(anonymous namespace)::GetStoragePartitionFromConfig()
#9 0x7f5394e23280 content::BrowserContext::GetStoragePartition()
#10 0x7f5394e23439 content::BrowserContext::GetDefaultStoragePartition()
#11 0x7f539e1046d1 ProfileImpl::DoFinalInit()
#12 0x7f539e10664f ProfileImpl::OnLocaleReady()
#13 0x7f539e1038d9 ProfileImpl::OnPrefsLoaded()
#14 0x7f539e1035bd ProfileImpl::ProfileImpl()
#15 0x7f539e101c2a Profile::CreateProfile()
#16 0x7f539dc36a87 ProfileManager::CreateProfileHelper()
#17 0x7f539dc2ed7a ProfileManager::CreateAndInitializeProfile()
#18 0x7f539dc2e967 ProfileManager::GetProfile()
#19 0x7f53a02d4298 GetStartupProfile()
#20 0x7f539e3bcaa2 (anonymous namespace)::CreatePrimaryProfile()
#21 0x7f539e3bb13c ChromeBrowserMainParts::PreMainMessageLoopRunImpl()
#22 0x7f539e3ba580 ChromeBrowserMainParts::PreMainMessageLoopRun()
#23 0x7f5394e34af1 content::BrowserMainLoop::PreMainMessageLoopRun()
#24 0x7f5394313485 _ZN4base8internal13FunctorTraitsIMN7content22IndexedDBCallbacksImpl13InternalStateEFvvEvE6InvokeIPS4_JEEEvS6_OT_DpOT0_
#25 0x7f5394e3fc31 _ZN4base8internal12InvokeHelperILb0EiE8MakeItSoIRKMN7content15BrowserMainLoopEFivEJPS5_EEEiOT_DpOT0_
#26 0x7f5394e3fbd7 _ZN4base8internal7InvokerINS0_9BindStateIMN7content15BrowserMainLoopEFivEJNS0_17UnretainedWrapperIS4_EEEEEFivEE7RunImplIRKS6_RKSt5tupleIJS8_EEJLm0EEEEiOT_OT0_NS_13IndexSequenceIJXspT1_EEEE
#27 0x7f5394e3fb1c _ZN4base8internal7InvokerINS0_9BindStateIMN7content15BrowserMainLoopEFivEJNS0_17UnretainedWrapperIS4_EEEEEFivEE3RunEPNS0_13BindStateBaseE
#28 0x7f53941f5aeb base::internal::RunMixin<>::Run()
#29 0x7f53959ed06b content::StartupTaskRunner::RunAllTasksNow()
#30 0x7f5394e326c0 content::BrowserMainLoop::CreateStartupTasks()
#31 0x7f5394e43667 content::BrowserMainRunnerImpl::Initialize()
#32 0x7f5394e2e97f content::BrowserMain()
#33 0x7f5396648956 content::RunNamedProcessTypeMain()
#34 0x7f539664abbc content::ContentMainRunnerImpl::Run()
#35 0x7f5396647882 content::ContentMain()
#36 0x7f539cb1b94a ChromeMain
#37 0x7f539cb1b872 main
#38 0x7f5387c43f45 __libc_start_main

Right, the shutdown leak is 'expected' because we're no longer cleanly shutting down.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/c59596d13ec0e105c72dd69661b8bcb505d9afa7

commit c59596d13ec0e105c72dd69661b8bcb505d9afa7
Author: wez <wez@chromium.org>
Date: Thu Feb 23 00:04:56 2017

Print creator's call-stack on URLRequestContextGetter leak in Debug.

This adds a StackTrace member to the URLRequestContextGetter base, to
allow the call-stack that created the instance to be printed in case
of it having to be leaked because of the owning thread having gone away.

BUG= 693965 

Review-Url: https://codereview.chromium.org/2702083002
Cr-Commit-Position: refs/heads/master@{#452289}

[modify] https://crrev.com/c59596d13ec0e105c72dd69661b8bcb505d9afa7/net/url_request/url_request_context_getter.cc
[modify] https://crrev.com/c59596d13ec0e105c72dd69661b8bcb505d9afa7/net/url_request/url_request_context_getter.h

Re #7: I don't think we can depend on the necessary bits to detect and ignore when this happens during browser shutdown, so perhaps we should just revert it?

That said, I thought we still [try to] shutdown the browser cleanly; it's just renderers that we exit without teardown?

Nope, like I mentioned on our hangout when discussing this earlier, but perhaps not clearly, the policy applies to shutdown in general. That is, the consequence of that discussion has been to affect _browser_ shutdown and not just _renderer_ shutdown (in my experience of the CLs I've seen)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/55fa5c6ec96a1f955cf0c3a85b6689239e8ab08e

commit 55fa5c6ec96a1f955cf0c3a85b6689239e8ab08e
Author: wez <wez@chromium.org>
Date: Thu Feb 23 20:24:26 2017

Revert of Print creator's call-stack on URLRequestContextGetter leak in Debug. (patchset #1 id:1 of https://codereview.chromium.org/2702083002/ )

Reason for revert:
This output is too noisy on browser shutdown, where is apparently now expected that we leak things.

Original issue's description:
> Print creator's call-stack on URLRequestContextGetter leak in Debug.
>
> This adds a StackTrace member to the URLRequestContextGetter base, to
> allow the call-stack that created the instance to be printed in case
> of it having to be leaked because of the owning thread having gone away.
>
> BUG= 693965 
>
> Review-Url: https://codereview.chromium.org/2702083002
> Cr-Commit-Position: refs/heads/master@{#452289}
> Committed: https://chromium.googlesource.com/chromium/src/+/c59596d13ec0e105c72dd69661b8bcb505d9afa7

TBR=rsleevi@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG= 693965 

Review-Url: https://codereview.chromium.org/2708923008
Cr-Commit-Position: refs/heads/master@{#452605}

[modify] https://crrev.com/55fa5c6ec96a1f955cf0c3a85b6689239e8ab08e/net/url_request/url_request_context_getter.cc
[modify] https://crrev.com/55fa5c6ec96a1f955cf0c3a85b6689239e8ab08e/net/url_request/url_request_context_getter.h