Re: git commit: TS-1487: More startup order tweaking, based on reported crash.

[Igor Galić <i....@brainsware.org>]

----- Original Message -----
> Updated Branches:
>   refs/heads/master 95c8f3893 -> 6a70e5abf
> 
> 
> TS-1487: More startup order tweaking, based on reported crash.
> 
> http://git-wip-us.apache.org/repos/asf/trafficserver/blob/6a70e5ab/proxy/Main.cc
> ----------------------------------------------------------------------
> diff --git a/proxy/Main.cc b/proxy/Main.cc
> index e9c281c..438242e 100644
> --- a/proxy/Main.cc
> +++ b/proxy/Main.cc
> @@ -1559,12 +1559,19 @@ main(int /* argc ATS_UNUSED */, char **argv)
>      SplitDNSConfig::startup();
>  #endif
>  
> +    // Load HTTP port data. getNumSSLThreads depends on this.
> +    if (!HttpProxyPort::loadValue(http_accept_port_descriptor))
> +      HttpProxyPort::loadConfig();
> +    HttpProxyPort::loadDefaultIfEmpty();
>  
>      if (!accept_mss)
>        TS_ReadConfigInteger(accept_mss,
>        "proxy.config.net.sock_mss_in");
>  
>      NetProcessor::accept_mss = accept_mss;
>      netProcessor.start(0, stacksize);
> +
> +    sslNetProcessor.start(getNumSSLThreads(), stacksize);
> +
>  #ifndef INK_NO_HOSTDB
>      dnsProcessor.start(0, stacksize);
>      if (hostDBProcessor.start() < 0)
> @@ -1575,15 +1582,15 @@ main(int /* argc ATS_UNUSED */, char **argv)
>      clusterProcessor.init();
>  #endif
>  
> -    // Plugins can get callbacks for very early events, so
> initialize even earlier.
> +#ifndef INK_NO_LOG

again one of those options that's from TS_IS_MICRO_BUILD

I'm understand there's people out there who don't want logging, but
isn't the easiest way to achieve that to simply have that around the
logging functions themselves / or turning them into macros that evaluate
to nothingness?

Rather than have this (potentially) incomplete mess that we have here.

> +    // initialize logging (after event and net processor)
> +    Log::init(remote_management_flag ? 0 :
> Log::NO_REMOTE_MANAGEMENT);
> +#endif
> +
> +    // Init plugins as soon as logging is ready.
>      plugin_init(system_config_directory);        // plugin.config
>      pmgmt->registerPluginCallbacks(global_config_cbs);
>  
> -    // Load HTTP port data. getNumSSLThreads depends on this.
> -    if (!HttpProxyPort::loadValue(http_accept_port_descriptor))
> -      HttpProxyPort::loadConfig();
> -    HttpProxyPort::loadDefaultIfEmpty();
> -
>      cacheProcessor.set_after_init_callback(&CB_After_Cache_Init);
>      cacheProcessor.start();
>  
> @@ -1593,13 +1600,6 @@ main(int /* argc ATS_UNUSED */, char **argv)
>      if (num_of_udp_threads)
>        udpNet.start(num_of_udp_threads, stacksize);
>  
> -    sslNetProcessor.start(getNumSSLThreads(), stacksize);
> -
> -#ifndef INK_NO_LOG
> -    // initialize logging (after event and net processor)
> -    Log::init(remote_management_flag ? 0 :
> Log::NO_REMOTE_MANAGEMENT);
> -#endif
> -
>      //acc.init();
>      //if (auto_clear_authdb_flag)
>       // acc.clear_cache();
> 
> 

-- 
Igor Galić

Tel: +43 (0) 664 886 22 883
Mail: i.galic@brainsware.org
URL: http://brainsware.org/
GPG: 6880 4155 74BD FD7C B515  2EA5 4B1D 9E08 A097 C9AE

Re: git commit: TS-1487: More startup order tweaking, based on reported crash.

[James Peach <ja...@me.com>]

On Jul 27, 2013, at 2:30 PM, Leif Hedstrom <zw...@apache.org> wrote:

> On 7/27/13 2:40 PM, Igor Galić wrote:
>> 
>> 
>> initialize even earlier.
>> +#ifndef INK_NO_LOG
>> again one of those options that's from TS_IS_MICRO_BUILD
> 
> I personally think we should kill the MICRO_BUILD... Heck, even on my rinky dink Raspberry Pi, the normal build works just fine. I rather have us optimize our code such that the normal build works well everywhere :).

Yeh I wanted to kill it to. Next time I'm just gong to kill it unless someone ponies up a patch

J

Re: git commit: TS-1487: More startup order tweaking, based on reported crash.

[Leif Hedstrom <zw...@apache.org>]

On 7/27/13 2:40 PM, Igor Galić wrote:
>
>
> initialize even earlier.
> +#ifndef INK_NO_LOG
> again one of those options that's from TS_IS_MICRO_BUILD

I personally think we should kill the MICRO_BUILD... Heck, even on my 
rinky dink Raspberry Pi, the normal build works just fine. I rather have 
us optimize our code such that the normal build works well everywhere :).

-- Leif