[AGENT++] Single-threaded build under autoconf

[Ralph Siemsen]

Hi Frank,

Yes, the whole agent is single threaded now, and it seems to
be quite stable - survived an overnight stress test.

Cheers,
-Ralph


On Thu, Aug 28, 2014 at 12:55:13AM +0200, Frank Fock wrote:
> Hi Ralph,
> 
> Is the whole agent single threaded then? Otherwise,
> single threaded AGENT++ will not help much, because you
> will get locking/cocnurrent access issues in the instrumentation
> anyway.
> 
> The _NO_THREADS flag still works. You can override it in the
> generated .h file.
> 
> Nevertheless, I will try to fix the autoconf opions to work
> consistently between SNMP++ and AGENT++ for the next
> releases.
> 
> Best regards,
> Frank
> 
> 
> Am 28.08.2014 00:18, schrieb Ralph Siemsen:
> >Hi Frank,
> >
> >I'm using single-threaded as a quick fix for locking problems
> >in code that I have inherited. I spent a week with valgrind
> >trying to fix the locks (in our code, not snmp++/agent++),
> >but ended up going in circles. Unfortunately I do not have
> >time to rewrite the whole thing.
> >
> >Also, we are running this on a small embedded system with
> >a single CPU core. Here threads do not really help, in fact
> >the extra system calls for mutexes, and context switches,
> >actually reduce performance somewhat.
> >
> >Any thoughts about agent++ single-threading? With the older
> >versions (before autotools) the _NO_THREAD option was picked
> >up from the snmp++ configuration. But with autotools that does
> >not seem to work anymore.
> >
> >Cheers,
> >-Ralph
> >
> >On Wed, Aug 27, 2014 at 10:54:09PM +0200, Frank Fock wrote:
> >>Hi Ralph,
> >>
> >>First of all, many thanks for reporting those issues. They will be fixed
> >>with the next release. However, the IPX support will be dropped
> >>(also from the configuration) as I do not think that anybody is using
> >>that anymore. The MAC address support needs to be reorganized
> >>to be independent of the obsolete code, because there some users
> >>that need it.
> >>
> >>Second, why do you need to compile a single threaded agent?
> >>
> >>Best regards,
> >>Frank
> >>
> >>
> >>Am 27.08.2014 22:42, schrieb Ralph Siemsen:
> >>>I'm trying to build snmp++-3.3.4 without thread support.
> >>>
> >>>  ./configure --disable-threads
> >>>
> >>>This results in a syntax error in the generated config_snmp_pp.h.
> >>>
> >>>snmp++-3.3.4/include/snmp_pp/config_snmp_pp.h:79:5: error: #if with no expression
> >>>
> >>>79:  #if
> >>>80:  #define HAVE_PTHREAD 1
> >>>81:  #endif
> >>>
> >>>It looks like the root cause is in configure.ac, when threads are disabled,
> >>>then WITH_THREADS=0 is set, but WITH_PTHREAD variable remains undefined,
> >>>which causes the generated header file to be invalid.
> >>>
> >>>The following patch to configure.ac seems to fix it:
> >>>
> >>>--- snmp++-3.3.4/configure.ac   2014-01-24 08:30:38.000000000 -0500
> >>>+++ snmp++-3.3.4.new/configure.ac       2014-08-27 10:14:55.914449056 -0400
> >>>@@ -186,7 +186,7 @@
> >>>          ],
> >>>          [])
> >>>      ])
> >>>-], [WITH_THREADS=0]) # endif : --enable-threads
> >>>+], [WITH_THREADS=0;WITH_PTHREAD=0]) # endif : --enable-threads
> >>>  dnl Checks for libraries.
> >>>
> >>>You may of course prefer to format that a little more clearly, I just
> >>>packed it all on the same line to minimize the noise in the diff.
> >>>
> >>>~~~~~
> >>>
> >>>Also in the same configure.ac, the MacAddress / IpxAddress options look
> >>>suspiciously incorrect. Running configure --help shows:
> >>>
> >>>   --enable-macaddress     enable support of class MacAddress (default: off)
> >>>   --disable-macaddress    disable support of class MacAddress
> >>>   --enable-macaddress     enable support of class IpxAddress (default: off)
> >>>   --disable-macaddress    disable support of class IpxAddress
> >>>
> >>>The names of the options for IpxAddress seem to be duplicates? I do not use
> >>>these options, but noticed it while fixing the first problem.
> >>>
> >>>~~~~~
> >>>
> >>>Lastly, I am also trying to build single-threaded version of agent++-4.0.2.
> >>>There is no --disable-threads option in its configure script, but it does
> >>>check for pthread support on non-win32 systems. This causes the compile to
> >>>be done with gcc -pthread option, regardless of how snmp++ has been built.
> >>>
> >>>I was able to get it working by ripping out the pthread section from the
> >>>configure.ac script, but that clearly is not a good solution. Would you
> >>>have a suggestion on how best to handle this?
> >>>
> >>>Thanks,
> >>>-Ralph
> >>>
> >>>_______________________________________________
> >>>AGENTPP mailing list
> >>>AGENTPP at agentpp.org
> >>>https://oosnmp.net/mailman/listinfo/agentpp
> >>-- 
> >>---
> >>AGENT++
> >>Maximilian-Kolbe-Str. 10
> >>73257 Koengen, Germany
> >>https://agentpp.com
> >>Phone: +49 7024 8688230
> >>Fax:   +49 7024 8688231
> >>
> >>_______________________________________________
> >>AGENTPP mailing list
> >>AGENTPP at agentpp.org
> >>https://oosnmp.net/mailman/listinfo/agentpp
> >_______________________________________________
> >AGENTPP mailing list
> >AGENTPP at agentpp.org
> >https://oosnmp.net/mailman/listinfo/agentpp
> 
> -- 
> ---
> AGENT++
> Maximilian-Kolbe-Str. 10
> 73257 Koengen, Germany
> https://agentpp.com
> Phone: +49 7024 8688230
> Fax:   +49 7024 8688231
> 
> _______________________________________________
> AGENTPP mailing list
> AGENTPP at agentpp.org
> https://oosnmp.net/mailman/listinfo/agentpp