Re: loss of coordination in SMP mode upon reconfigure in trunk?

Hi,
I've changed the config removing references to aufs, and set cache_mem
0 in order to force hits from disk; then I launched a GET loop forcing
hits:

1363686304.923 0 2.2.2.1 TCP_HIT/200 8212 GET
http://wiki.squid-cache.org/wiki/squidtheme/img/squid-bubbles.png -
HIER_NONE/- image/png

$ squidclient mgr:storedir
HTTP/1.1 200 OK
Server: squid/3.3.3-BZR
Mime-Version: 1.0
Date: Tue, 19 Mar 2013 09:45:21 GMT
Content-Type: text/plain
Expires: Tue, 19 Mar 2013 09:45:21 GMT
Last-Modified: Tue, 19 Mar 2013 09:45:21 GMT
Connection: close

by kid1 {
Store Directory Statistics:
Store Entries : 27
Maximum Swap Size : 0 KB
Current Store Swap Size: 0.00 KB
Current Capacity : 0.00% used, 0.00% free

} by kid1

by kid2 {
Store Directory Statistics:
Store Entries : 27
Maximum Swap Size : 0 KB
Current Store Swap Size: 0.00 KB
Current Capacity : 0.00% used, 0.00% free

} by kid2

by kid3 {
Store Directory Statistics:
Store Entries : 27
Maximum Swap Size : 262144 KB
Current Store Swap Size: 168328.50 KB
Current Capacity : 64.21% used, 35.79% free

Store Directory #0 (rock): /home/kinkie/squid/install/var/cache/rock
FS Block Size 1024 Bytes

Maximum Size: 262144 KB
Current Size: 168328.50 KB 64.21%
Maximum entries: 4194
Current entries: 2693 64.21%
Pending operations: 1 out of 0
Flags:
} by kid3

After a reconfigure:

squidclient mgr:storedir
HTTP/1.1 200 OK
Server: squid/3.3.3-BZR
Mime-Version: 1.0
Date: Tue, 19 Mar 2013 09:46:02 GMT
Content-Type: text/plain
Expires: Tue, 19 Mar 2013 09:46:02 GMT
Last-Modified: Tue, 19 Mar 2013 09:46:02 GMT
Connection: close

by kid1 {
Store Directory Statistics:
Store Entries : 27
Maximum Swap Size : 0 KB
Current Store Swap Size: 0.00 KB
Current Capacity : 0.00% used, 0.00% free

} by kid1

by kid2 {
Store Directory Statistics:
Store Entries : 27
Maximum Swap Size : 0 KB
Current Store Swap Size: 0.00 KB
Current Capacity : 0.00% used, 0.00% free

} by kid2

by kid3 {
Store Directory Statistics:
Store Entries : 27
Maximum Swap Size : 0 KB
Current Store Swap Size: 0.00 KB
Current Capacity : 0.00% used, 0.00% free

} by kid3

But still:
1363686365.503 1 2.2.2.1 TCP_HIT/200 8212 GET
http://wiki.squid-cache.org/wiki/squidtheme/img/squid-bubbles.png -
HIER_NONE/- image/png

Hits seem to be correctly recorded in mgr:info
All kids report to mgr:info -> memory allocation statistics

So it seems to be only an issue with reporting..

On Mon, Mar 18, 2013 at 8:15 PM, Alex Rousskov
<rousskov_at_measurement-factory.com> wrote:
> On 03/18/2013 11:56 AM, Alex Rousskov wrote:
>> On 03/18/2013 06:47 AM, Kinkie wrote:
>>> I'm seeing a strange behavior with a trunk build.
>>>
>>> Key configuration:
>>> workers 2
>>> cache_dir rock /cache/rock 256 max-size=64000
>>> cache_dir aufs /cache/aufs/${process_number} 40 16 256
>>>
>>>
>>> then with squidclient mgr:storedir
>>> I can see all kids report their own cache_dir stats
>>>
>>> After a reconfigure, only the rock dir is reported by cachemgr.
>>> Also, mgr:info will report results only from kids 3 and 4 (disker and
>>> coordinator)
>>>
>>> Should I report a bug?
>>
>> If this is specific to [a]ufs, then please indicate that in the bug
>> report summary line. If you can reproduce these problems without [a]ufs
>> lines, please open a bug report without them.
>
> And, as somebody pointed out to me privately, please note that bug 3774
> is similar: http://bugs.squid-cache.org/show_bug.cgi?id=3774
>
> Bug 3774 config is using a single worker, but, technically, it is still
> SMP because diskers are kids, just like workers. It may be important
> that diskers disappear from mgr:storedir when there is only one worker
> because that would kind of match what Kinkie is describing from aufs
> point of view (since aufs is not SMP-aware).
>
>
> HTH,
>
> Alex.
>

-- 
    /kinkie