<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
<META NAME="Generator" CONTENT="MS Exchange Server version 6.5.7651.53">
<TITLE>RE : renaming varnish concepts...</TITLE>
</HEAD>
<BODY>
<!-- Converted from text/plain format -->

<P><FONT SIZE=2>+1 for some house-cleaning.<BR>
<BR>
- Speaking of bans, is grace used for "banned" objets ?<BR>
<BR>
If i "ban" an object, the first request will regenerate it from the back-end. Will following requests be getting the stale version during regeneration ?<BR>
<BR>
- For the CLI, why not add a "compatible mode" (inactive by default) to remove the greeting ?<BR>
<BR>
Jean-François<BR>
<BR>
<BR>
-------- Message d'origine--------<BR>
De: varnish-misc-bounces@projects.linpro.no de la part de Poul-Henning Kamp<BR>
Date: mer. 28/01/2009 10:11<BR>
À: varnish-misc@projects.linpro.no<BR>
Objet : renaming varnish concepts...<BR>
<BR>
<BR>
A couple of places in varnish I have clearly missed the boat when it<BR>
comes to naming.<BR>
<BR>
And a couple of places, I should have thought more carefully.<BR>
<BR>
I wonder if we should correct these things in 2.1, in order to not<BR>
confuse future users more than necessary, or if we already have<BR>
sufficient users that it does not make sense to change things like<BR>
this now.<BR>
<BR>
2.1 is going to be a flag day anyway, because of the changes which<BR>
persistent storage will cause, so I'm thinking this might be a<BR>
good time for this sort of house-cleaning ?<BR>
<BR>
Input welcome.<BR>
<BR>
1.  Purge vs. Ban<BR>
-----------------<BR>
<BR>
The CLI and VCL commands are named "purge", but they don't, they<BR>
add a ban to the list of bans.<BR>
<BR>
I would actually like to rename purge to ban and add a real purge<BR>
function that gets rid of the current object (ie: one found in the<BR>
cache) and possibly its "Vary:" siblings.<BR>
<BR>
Purge does sound like it will be gone, whereas "ban" better explains<BR>
what happens when we use the delayed regexp checks.<BR>
<BR>
Obviously, if I co-opt "purge" to mean something different, backwards<BR>
compat is not possible, and all your purge scripts and VCLs with<BR>
purge facilities will break.<BR>
<BR>
<BR>
2.  CLI greeting<BR>
----------------<BR>
<BR>
When you connect to CLI port (-T argument) it does not announce<BR>
itself.  This was probably a mistake, as it makes it harder to<BR>
do error detection in scripts that talk to the CLI.<BR>
<BR>
The problem is that if I add the greeting, I will break all<BR>
existing scripts that talk to the CLI.<BR>
<BR>
<BR>
3.  Acceptor vs. waiter<BR>
-----------------------<BR>
<BR>
The subfunction we today call "acceptors" are not really, they are<BR>
waiters that monitor client connections for new requests.<BR>
<BR>
The real acceptor is a separate piece of code.<BR>
<BR>
This one is only visible in a command-line argument/parameter which<BR>
users generally only fiddle as part of debugging.<BR>
<BR>
--<BR>
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20<BR>
phk@FreeBSD.ORG         | TCP/IP since RFC 956<BR>
FreeBSD committer       | BSD since 4.3-tahoe   <BR>
Never attribute to malice what can adequately be explained by incompetence.<BR>
_______________________________________________<BR>
varnish-misc mailing list<BR>
varnish-misc@projects.linpro.no<BR>
<A HREF="http://projects.linpro.no/mailman/listinfo/varnish-misc">http://projects.linpro.no/mailman/listinfo/varnish-misc</A><BR>
<BR>
</FONT>
</P>

</BODY>
</HTML>