1169994 - DOM cache may unregister offline storage too soon

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[Jan Varga [:janv]]

Attachment: slowdown patch

After the vacuum stuff landed, closing of storage connections takes a bit more time. This causes that my directory lock stuff doesn't pass tests anymore. I investigated the issues a bit and I believe the offline storage is unregistered too soon. The same issue arises on current m-c when I artificially slowdown the closing even more (see the patch).

Steps to reproduce:
1. apply the slowdown patch
2. run ./mach mochitest-plain dom/cache dom/indexedDB
3. see the assertion in OriginInfo::~OriginInfo

I think Context::~Context should do a better job and release mOfflineStorage when the connection actually finished the closing procedure.

Comment 1

[Jan Varga [:janv]]

The assertion in OriginInfo destructor means that QM deleted an origin that still has live quota objects. I verified that these quota objects belong to cache.sqlite and cache.sqlite-wal

Comment 2

[Ben Kelly [:bkelly, not reviewing]]

Please post the patch that enables your directory locking.  The ~Context() destructor should not run while the Connection is executing on the IO thread.

Comment 3

[Ben Kelly [:bkelly, not reviewing]]

Oh, I see.  It happens without your patch.  Let me look.

Comment 4

[Ben Kelly [:bkelly, not reviewing]]

Shorter set of tests to run to trigger:

  ./mach mochitest dom/cache/test/mochitest/test_caches.html dom/indexedDB/test/test_add_put.html

Comment 5

[Jan Varga [:janv]]

(In reply to Ben Kelly [:bkelly] from comment #4)
> Shorter set of tests to run to trigger:
> 
>   ./mach mochitest dom/cache/test/mochitest/test_caches.html
> dom/indexedDB/test/test_add_put.html

Yeah, that narrows it down even more.

Comment 6

[Ben Kelly [:bkelly, not reviewing]]

Attachment: Fix Cache to close connection on right thread when init is canceled. r=ehsan

The Context's QuotaInitRunnable was not dropping its mData ref on the target thread.  This resulted in the connection getting proxy released when the init was canceled.

This patch fixes the runnable to drop the ref correctly.  I also changed the proxy release to a thread assert.

Comment 7

[Ben Kelly [:bkelly, not reviewing]]

 https://treeherder.mozilla.org/#/jobs?repo=try&revision=d0c864b8fa77

Comment 8

[Ben Kelly [:bkelly, not reviewing]]

New try since the outage caused the last to fail:

  https://treeherder.mozilla.org/#/jobs?repo=try&revision=474b3eaf4093

Comment 9

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/168cdb48c5eb

Comment 10

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/168cdb48c5eb