(ITS#8225) writetimeout should ignore consumers

[hyc@openldap.org]

Full_Name: Howard Chu
Version: 2.4
OS: 
URL: ftp://ftp.openldap.org/incoming/
Submission from: (NULL) (70.87.222.79)
Submitted by: hyc


We found a case where a site was using regular syncrepl and replicating changes
to very large groups (hundreds of thousands of members, tens of megabytes total
entry size) and the socket send queues on the provider were getting filled
because the consumers weren't reading or processing fast enough. At some point
the writetimeout on the provider kicked in and these syncrepl connections got
closed, forcing the consumers to reconnect and start the whole process all over
again. This sequence of events basically meant the consumers never made any
progress, because the refresh on next connect attempt had to send such a large
volume of data all over again. syslogs showed consumers having their connection
closed multiple times within short (5 minute) spans of time.

The idletimeout already ignores consumer connections; writetimeout should as
well. Sites can use keepalive to handle truly hung consumers, as opposed to just
overloaded consumers.