On 11/08/2009 12:53 AM, ios firefox wrote:
1. is TS's works asynchronous, like lighttpd?
or it works synchronous, one thread serves one request, when all
the threads in the pool exhausts, then refuse to work?
Both. There are <x> number of worker threads (net-threads), by default
we create 2.5 threads per core (CPU). This is configurable in
records.config. Within each worker thread, it's asynchronous like
lighttpd or squid (each thread has event driven state machines). This is
what makes TS scale well, there's no thread pool exhaustion like
Varnish, there's no single thread like Squid.
2. Can "Congestion Control" be used in revers proxy mode, when
congestion happend, return the user a user-defined page?
Congestion control is something we didn't use at Yahoo, and the code is
untested at a minimum, and maybe even missing partially (I think the
congestion control stuff depends on the ARM module, which we removed).
If someone is interested in working on this, we'd be more than happy to
help you get started, but it'd have to be owned by a group of people
that cares for it. We'll also have to figure what, if any, additional
code we'll have to opensource.
3. I am trying to find the command to delete certain cache object by
regex. But I only find the method which delete the total database. is
there a method to only delete certain object ?
This is the cache inspector. The documentation for this might be a
little wonky, and it used to only work with IE. But we made some changes
to this feature a while ago, where you can actually set it up in reverse
proxy mode (that avoids the issue of having to point your browser to it
as the forward proxy). In order to use the cache inspector, you have to
allow it, in records.config you'll have:
CONFIG proxy.config.http_ui_enabled INT 1
If I recall, you can setup a remap.config rule like this if you want to
use it in reverse proxy mode:
map http://yourmachinename.com/cacheinspector http://{cache}
(remember that yourmachinename.com can be anything, as long as the
requests Host: header matches that).
Now, this would allow it from anywhere, so you might want to put an ACL
on that in either remap.config, e.g.
map http://yourmachinename.com/cacheinspector http://{cache}
@action=allow @src_ip=10.0.0.0-10.255.255.255
Or use regular acl's (which are slow). Alternatively, you could bind a
second port, which is firewalled off from the the world, but allowed
from inside. E.g.
CONFIG proxy.config.http.server_other_ports STRING 81:X
And then change the map rule accordingly, e.g.
map http://yourmachinename.com:81/cacheinspector http://{cache}
(this would be my personal choice since it's easy to manage this once
and for all in a firewall ACL).
Cache inspection is very expensive if you have a very large cache, and
as far as I know, is only exposed via this type of GUI. It'd be useful
to make some generic interface for this from command line, so that it
could more easily be scripted. Patches are appreciated (but file Jira
tickets first :).
Cheers,
-- leif
