Moshe

19 0.00 0.00 Client connections accepted
9 0.00 0.00 Client requests received

I am running the following options:

DAEMON_OPTS=”-a :80 \
-T localhost:6082 \
-f /etc/varnish/default.vcl \
-u varnish -g varnish \
-s file,/var/lib/varnish/varnish_storage.bin,5G \
-p thread_pool_add_delay=2 \
-p thread_pools=2 \
-p thread_pool_max=4000 \
-p session_linger=50 \
-p sess_workspace=16384
”

Also sess_workspace=16k does not work it only liked the number

I was using the code here: http://varnish-cache.org/wiki/VCLExamplePurging . But as the VCL is quite the same as the one you wrote in this article, is the sentence correct: «That means that if you now send a

PURGE / HTTP/1.0
Host: http://www.example.com

to Varnish over port 80 (restricted to client with IP 192.0.2.14), your / document from the http://www.example.com website will be purged.»

With your answer, I would guess that any «/» document will be purged, was it on http://www.example.com or not. If so, can someone fix the wiki? Along with your remark about the fact that it’s a purging command, so an error should be sent right away, or a GET should be used (that’s the core of your posting here).

Thanks a lot!

To the second question: Using just req.http.host is faster, though I doubt you’d notice the difference.

As for ‘G’-purges, those are ‘Gone’: They are just placeholders because an object points to them, but they will never be tested against. They will be removed once there are no objects referencing them. (This is why you always have at least one item on the purge list, even right after you start up).

n_purge should contain the number of purges you have active, and you can see that if you subtract the retired purges from the added purges (4844 – 4801), you get n_purge (43). I can’t explain why you only see 39 purges, but if varnishstat and purge.list disagree on the number of purges, purge.list is probably correct.

I’ve got the code form the URL you’ve mentioned at the top of your article. Is the hostname used in such a case? Because in purge.list, I never see any hostnames.
Your example above states:
purge(”req.http.host == ” req.http.host ” && req.url ~ ” req.http.X-Purge-Regex);
so I think there’s an issue in my code as I don’t have the host (of course I have multiple hostnames, else the question is silly).

Another question:
if you want to purge a whole host, is it better to do:
purge(”req.http.host ~ example.com$”);
or:
purge(”req.http.host ~ example.com$ && req.url ~ ^/.*”);

I think the first one should be faster.

And finally, I want to track the number of purging rules (to see if there’s not too much rules). If I run purge.list, I can see 35 rules (I exclude the «G» rules, I think these are duplicates, else it would be 39 rules), but in the stats I can read:
n_purge 43 . N total active purges
n_purge_add 4844 0.00 N new purges added
n_purge_retire 4801 0.00 N old purges deleted
n_purge_obj_test 5206879 3.06 N objects tested
n_purge_re_test 34006296 19.97 N regexps tested against
n_purge_dups 1569 0.00 N duplicate purges removed

What data contains the real active rules?

We have been depending on the varnish_allocated munin plugin[1] though it wasn’t reporting any new figures since changing to malloc as a storage type. I have since tweaked it to grep for SMA bytes allocated.

Now I’m considering developing it to further to report on the SMA bytes free. Though I’m not sure it is would be valuable information.

Thanks!

[1] http://muninexchange.projects.linpro.no/?search=&cid=37&os4=on&os7=on&os3=on&os2=on&os5=on&os8=on&os1=on&os6=on&pid=154

I discovered there’ll be a talk about it at FOSDEM, though, so I’ll postpone my decision until after that. It does seem rather like what I want, though.