769262 - pipeline read-timeout rescheduling problem

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect)

Description

[Patrick McManus [:mcmanus]]

A transaction at the front of a pipeline can be canceled and restarted in a non-pipelined context if it hangs for a very long time - the concern is that the pipelining created a server or intermediary hang somewhere.

Two fixes are necessary to this logic:

1 - the pipeline position of the rescheduled transaction needs to be reset to 0 so that this timer does not apply the next time around. Before we switched SOLO transactions away from using nsHttpPipeline that happened as part of the pipeline scheduling, but now the old value is just inherited so it needs to be reset explicitly.

2 - NET_TIMEOUT needs to be added to the list of conditions for which pipeline response(0) will be retried.

Comment 1

[Patrick McManus [:mcmanus]]

Attachment: patch 0

Comment 2

[Honza Bambas (:mayhemer)]

Comment on attachment 637507 [details] [diff] [review]
patch 0

Review of attachment 637507 [details] [diff] [review]:
-----------------------------------------------------------------

r=honzab.

Comment 3

[Patrick McManus [:mcmanus]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/10c1a7434e67

the checkin comment accidentally says bug 763312

Comment 4

[Patrick McManus [:mcmanus]]

changeset:   97971:10c1a7434e67
user:        Patrick McManus <[e-mail]>
date:        Fri Jun 29 14:24:29 2012 -0400
summary:     bug 763312 pipeline position 0 read time rescheduling fixes r=honzab