Hi guys,
Back and I suppose cutting vacation a tad short. It's important to have a
stream per client unless you want everything to serialize to a particular
server. That is very unwise, in fact it's plain wrong. If you want crappy
performance, then yes one should serialize. I don't see the point and a
serialization patch is not tenable here. More responses as I wade through
the thread backwards.
Andy
On Fri, 30 Aug 2013, Alja Mrak Tadel wrote:
> On 08/29/13 13:58, Alja Mrak Tadel wrote:
>> Hi Brian,
>>
>> I have disable this. I'm running tests with this change.
>
> Just to confirm, the proxy server is now using only one connection to the
> origin.
>
> Alja
>
>>
>> On 08/29/13 12:48, Brian Bockelman wrote:
>>> Hi Alja,
>>>
>>> Maybe it's a dumb question - but what happens if we disable this? The
>>> new XrdCl is supposed to be thread safe, so I don't see the advantage
>>> of having a stream per client -- I'd prefer a stream per server.
>>>
>>> Brian
>>>
>>> On Aug 29, 2013, at 1:32 PM, Alja Mrak Tadel <[log in to unmask]> wrote:
>>>
>>>> Here is the backtrace:
>>>>
>>>> (gdb) bt
>>>> #0 XrdPssSys::P2URL (retc=@0x7ffff5b4c32c, pbuff=0x7ffff5b4b320 "",
>>>> pblen=4096, path=<value optimized out>, Split=0, Cgi=0x7ffff554777c
>>>> "oss.lcl=1", CgiLn=9, Ident=0x66a414 ":37@xrootd-proxy", doN2N=1) at
>>>> /home/alja/xrd/newcl/xrootd-src/src/XrdPss/XrdPss.cc:893
>>>> #1 0x00007ffff5544107 in XrdPssFile::Open (this=0x7fffcc001960,
>>>> path=0x7fffcc000c00 "/store/user/alja/data.root", Oflag=0, Mode=384,
>>>> Env=...) at /home/alja/xrd/newcl/xrootd-src/src/XrdPss/XrdPss.cc:611
>>>> #2 0x00007ffff79b8d21 in XrdOfsFile::open (this=0x7fffcc001010,
>>>> path=0x7fffcc000c00 "/store/user/alja/data.root", open_mode=0,
>>>> Mode=384, client=0x7fffcc000980, info=0x0) at
>>>> /home/alja/xrd/newcl/xrootd-src/src/XrdOfs/XrdOfs.cc:626
>>>> #3 0x0000000000428fd2 in XrdXrootdProtocol::do_Open (this=<value
>>>> optimized out>) at
>>>> /home/alja/xrd/newcl/xrootd-src/src/XrdXrootd/XrdXrootdXeq.cc:1169
>>>> #4 0x00007ffff7787529 in XrdLink::DoIt (this=0x66a358) at
>>>> /home/alja/xrd/newcl/xrootd-src/src/Xrd/XrdLink.cc:396
>>>> #5 0x00007ffff778b425 in XrdScheduler::Run (this=0x7ffff7ddca60) at
>>>> /home/alja/xrd/newcl/xrootd-src/src/Xrd/XrdScheduler.cc:307
>>>> #6 0x00007ffff778b619 in XrdStartWorking (carg=<value optimized
>>>> out>) at /home/alja/xrd/newcl/xrootd-src/src/Xrd/XrdScheduler.cc:84
>>>> #7 0x00007ffff77573af in XrdSysThread_Xeq (myargs=0x6494a0) at
>>>> /home/alja/xrd/newcl/xrootd-src/src/XrdSys/XrdSysPthread.cc:86
>>>> #8 0x00007ffff731b851 in start_thread () from /lib64/libpthread.so.0
>>>> #9 0x00007ffff66c190d in clone () from /lib64/libc.so.6
>>>>
>>>> (gdb) n
>>>> 894 if ((idP = index(idBuff, '@'))) {*(idP+1) = 0; theID =
>>>> idBuff;}
>>>> (gdb) l
>>>> 889 // If we have an Ident then usethe fd number as the userid.
>>>> This allows us to
>>>> 890 // have one stream per open connection.
>>>> 891 //
>>>> 892 if (Ident && (Ident = index(Ident, ':')))
>>>> 893 {strncpy(idBuff, Ident+1, 7); idBuff[7] = 0;
>>>> 894 if ((idP = index(idBuff, '@'))) {*(idP+1) = 0; theID =
>>>> idBuff;}
>>>> 895 }
>>>> 896
>>>> 897 // Format the header into the buffer and check if we
>>>> overflowed. Note that there
>>>> 898 // can be a maximum of 8 substitutions, so that's how many we
>>>> provide.
>>>> (gdb) p idBuff
>>>> $7 = "37@xroo"
>>>>
>>>> Alja
>>>>
>>>> ########################################################################
>>>> Use REPLY-ALL to reply to list
>>>>
>>>> To unsubscribe from the XROOTD-DEV list, click the following link:
>>>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
>>>
>>
>> ########################################################################
>> Use REPLY-ALL to reply to list
>>
>> To unsubscribe from the XROOTD-DEV list, click the following link:
>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
>
> ########################################################################
> Use REPLY-ALL to reply to list
>
> To unsubscribe from the XROOTD-DEV list, click the following link:
> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
>
########################################################################
Use REPLY-ALL to reply to list
To unsubscribe from the XROOTD-DEV list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
