[PATCH 02/15] Add a BAN_Shutdown() routine that is called before STV_Close(), and makes sure that the ban lists will not change until we exit.

Poul-Henning Kamp phk at phk.freebsd.dk
Mon Nov 12 10:40:15 CET 2012

In message <1352287938-7247-2-git-send-email-martin at varnish-software.com>, Mart
in Blix Grydeland writes:
>This is to give persistent stevedores a chance to clean up their ban
>lists without having to worry about the lists changing under them
>(normally the ban callbacks are called under the ban mutex, but not so
>during exit).

I don't have a problem with the idea, but I don't like having BAN_Insert()
return a failure we don't know how to handle.  I think BAN_Insert() should
take ownership of the ban even when it fails, but returning a status code
is probably a good idea, provided it gets used for something somewhere.

I think it should be used in the CLI::ban case, so that a cluster-controller
can keep track of which bans have been entered on which varnish instances.

>Also remove the duplicate drop tail bans code. That is now only done
>in the main thread loop.

This breaks r01030.vtc so it should probably be done in its own commit.

Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.

More information about the varnish-dev mailing list