Illumos/SmartOS/Solaris issue when using eventport and keepalive at the same time

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Illumos/SmartOS/Solaris issue when using eventport and keepalive at the same time

Peter Toth
Hi all,

There is a lock-up issue on Solaris derived operating systems when using "keepalive" and "use eventport" configuration directives at the same time.

The issue has been described previously back in 2015 and a patch was proposed here: https://forum.nginx.org/read.php?2,259798,259798#msg-259798

I have tested the patch and it is working.

Could anyone comment on whether the patch looks acceptable in its current form?
Any objections to applying this fix in the next release?

Thanks
P

_______________________________________________
nginx mailing list
[hidden email]
http://mailman.nginx.org/mailman/listinfo/nginx
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Illumos/SmartOS/Solaris issue when using eventport and keepalive at the same time

Maxim Dounin
Hello!

On Fri, Mar 17, 2017 at 11:46:46AM +1300, Peter Toth wrote:

> There is a lock-up issue on Solaris derived operating systems when using
> "keepalive" and "use eventport" configuration directives at the same time.
>
> The issue has been described previously back in 2015 and a patch was
> proposed here: https://forum.nginx.org/read.php?2,259798,259798#msg-259798
>
> I have tested the patch and it is working.
>
> Could anyone comment on whether the patch looks acceptable in its current
> form?
> Any objections to applying this fix in the next release?

There are problems with eventport implementation, known for at
least several years now, and these can be easily reproduced by
running our test suite with 'use eventport' added to test
configurations.  If I recall correctly, upstream keepalive is not
something important to trigger problems, any connection to an
upstream is basically enough.

Unfortunately, these problems are very low priority due to
minor popularity of Solaris.  Consider using /dev/poll instead,
which is the default, well tested and has no known problems.

The patch in question looks more like a hack to mitigate a
particular issue the author was facing.  It doesn't resolve the
underlying problems though, and likely there are other issues as
well.  There are no plans to commit this patch.

--
Maxim Dounin
http://nginx.org/
_______________________________________________
nginx mailing list
[hidden email]
http://mailman.nginx.org/mailman/listinfo/nginx
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Illumos/SmartOS/Solaris issue when using eventport and keepalive at the same time

Peter Toth
Great, thanks for clarifying this.

On Sat, Mar 18, 2017 at 2:09 AM, Maxim Dounin <[hidden email]> wrote:
Hello!

On Fri, Mar 17, 2017 at 11:46:46AM +1300, Peter Toth wrote:

> There is a lock-up issue on Solaris derived operating systems when using
> "keepalive" and "use eventport" configuration directives at the same time.
>
> The issue has been described previously back in 2015 and a patch was
> proposed here: https://forum.nginx.org/read.php?2,259798,259798#msg-259798
>
> I have tested the patch and it is working.
>
> Could anyone comment on whether the patch looks acceptable in its current
> form?
> Any objections to applying this fix in the next release?

There are problems with eventport implementation, known for at
least several years now, and these can be easily reproduced by
running our test suite with 'use eventport' added to test
configurations.  If I recall correctly, upstream keepalive is not
something important to trigger problems, any connection to an
upstream is basically enough.

Unfortunately, these problems are very low priority due to
minor popularity of Solaris.  Consider using /dev/poll instead,
which is the default, well tested and has no known problems.

The patch in question looks more like a hack to mitigate a
particular issue the author was facing.  It doesn't resolve the
underlying problems though, and likely there are other issues as
well.  There are no plans to commit this patch.

--
Maxim Dounin
http://nginx.org/
_______________________________________________
nginx mailing list
[hidden email]
http://mailman.nginx.org/mailman/listinfo/nginx


_______________________________________________
nginx mailing list
[hidden email]
http://mailman.nginx.org/mailman/listinfo/nginx
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Illumos/SmartOS/Solaris issue when using eventport and keepalive at the same time

James
In reply to this post by Maxim Dounin
On 17/03/2017 13:09, Maxim Dounin wrote:

Hello!

> There are problems with eventport implementation, known for at
> least several years now, and these can be easily reproduced by
> running our test suite with 'use eventport' added to test
> configurations.  If I recall correctly, upstream keepalive is not
> something important to trigger problems, any connection to an
> upstream is basically enough.
>
> Unfortunately, these problems are very low priority due to
> minor popularity of Solaris.  Consider using /dev/poll instead,
> which is the default, well tested and has no known problems.

I accept all that but learnt the hard way myself.  Perhaps the docs
could not recommend eventport with Solaris and warn against its use for
now.  eg:

http://nginx.org/en/docs/events.html


Спасибо!


James.
_______________________________________________
nginx mailing list
[hidden email]
http://mailman.nginx.org/mailman/listinfo/nginx
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Illumos/SmartOS/Solaris issue when using eventport and keepalive at the same time

Yaroslav Zhuravlev

On 18 Mar 2017, at 14:14, James <[hidden email]> wrote:

> On 17/03/2017 13:09, Maxim Dounin wrote:
>
> Hello!
>
>> There are problems with eventport implementation, known for at
>> least several years now, and these can be easily reproduced by
>> running our test suite with 'use eventport' added to test
>> configurations.  If I recall correctly, upstream keepalive is not
>> something important to trigger problems, any connection to an
>> upstream is basically enough.
>>
>> Unfortunately, these problems are very low priority due to
>> minor popularity of Solaris.  Consider using /dev/poll instead,
>> which is the default, well tested and has no known problems.
>
> I accept all that but learnt the hard way myself.  Perhaps the docs could not recommend eventport with Solaris and warn against its use for now.  eg:
>
> http://nginx.org/en/docs/events.html
>

Hi James,

Thanks, docs updated:

http://nginx.org/en/docs/events.html#eventport


>
> Спасибо!
>
>
> James.
> _______________________________________________
> nginx mailing list
> [hidden email]
> http://mailman.nginx.org/mailman/listinfo/nginx

--
yar
_______________________________________________
nginx mailing list
[hidden email]
http://mailman.nginx.org/mailman/listinfo/nginx
Loading...