Re: patches

Duane Wessels wrote:
>
> Here's the list of your patches that I added to 2.3:
>
[...]
> X Squid-2.2.DEVEL3: Assertion failure on FTP PUT to directories
> X Squid-2.2.DEVEL3: Blank content types logged in store.log

Phew. The list has been a bit too long lately ;-)

> _ Squid-2.3.DEVEL2: cache_swap_log name based on cache_dir name
> _ Squid-2.3.DEVEL2: allow-miss cache_peer option
> _ Squid-2.2.STABLE2: UNIX domain IPC communication
> _ Squid-2.2.STABLE2: Use estimated RTT for parent selection on timeouts
> _ Squid-2.3.DEVEL2: ICP timeout selection
> _ Squid-2.2.DEVEL3: Don't allow netdb selection to bypass never_direct
> _ Squid-2.2.STABLE3: nonhierarchical_direct squid.conf directive
> _ Squid-2.3.DEVEL2: Improved TCP dead peer detection & failover
> _ Squid-2.2.DEVEL3: Improved ICP dead peer detection

Is this a list of patches not addded?

> didn't add any that modify or improve peer selection because I think
> its already too complicated and ugly (which is my fault). Before we
> add more and more confusing options and special cases, I want to
> have a well-written document that describes what the procedure should
> be.

I didn't expect you to, except for the nonhierarchical_direct which
changes Squid to behave more in the way one expects when not using ICP.
The fact that it adds one configure option in not what is important,
what is important is that it changes Squid to try to use parents for
hierarchical requests even if ICP isn't used.

I agree that the peer selection is messy and ugly, and in many ways
inconsistent between the different selection methods. The whole peer
selection should probably been torn apart and put together again in
parallell with a design document describing what it should do (not what
it currently does). Much of the ugliness comes from firts being written
for ICP alone and then being cludged to support other methods without
any clear goal.

/Henrik
Received on Wed Oct 20 1999 - 16:02:29 MDT