1009500 - HTTP cache v2: respect OPEN_BYPASS_IF_BUSY in the new cache

Status: RESOLVED FIXED

(Core :: Networking: Cache, defect)

Description

[Honza Bambas (:mayhemer)]

There are still scenarios the consumer using this flag can get blocked like:

- channel 1 opens an entry
- is given a new empty entry
- entry is at WRITING state (bypasses any callbacks until write done/failed)
- channel 1 is suspended
- channel 2 opens the same entry, specifies OPEN_BYPASS_IF_BUSY 
- channel 1 is never resumed (the scenario waits for channel 2 to finish first)
=> channel 2 never gets it's cache entry and gets stuck

Comment 1

[Honza Bambas (:mayhemer)]

See bug 982598 comment 860 for logs.

Comment 2

[JW Wang [:jwwang] [:jw_wang]]

Do you mean bug 981153 comment 860?

Comment 3

[Honza Bambas (:mayhemer)]

Yep!

Comment 4

[Honza Bambas (:mayhemer)]

Attachment: v1

- callbacks get in only via CacheEntry::AsyncOpen
- all of them pass via CacheEntry::RememberCallback
- when the state is detected to be WRITING/REVAL and BYPASS flag is set, immediately (from AsyncOpen) invoke with null/NOT_FOUND error at that method (and actually don't remember the callback)
- test infra is added a new flag DONTFILL to bypass the metadata and data automatic write in OpenCallback js helper class
- the test itself opens an entry but never writes/validates/closes it, the second open with BYPASS must be invoked then

Comment 5

[Honza Bambas (:mayhemer)]

https://tbpl.mozilla.org/?tree=Try&rev=45248123e5ce

Comment 6

[Honza Bambas (:mayhemer)]

Rather this one: https://tbpl.mozilla.org/?tree=Try&rev=6d482e3d104f

Comment 7

[Honza Bambas (:mayhemer)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/e0fa93595164

Comment 8

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

https://hg.mozilla.org/mozilla-central/rev/fde411a82da7