quickfixj-users

[Quickfixj-users] Is anybody seeing dropped incoming SSL messages?

[Toli K. <to...@ma...>]

Hi,

I'm using QFJ to connect to a broker via SSL.

Every once in a while, i get in a situation where the underlying MINA
layer is seeing incoming FIX messages, but they don't seem to ever
make their way up to the application layer.
Connecting to the same broker without SSL, i'm seeing the same exact
incoming messages, but they don't ever get dropped.

For example, during connection in the Mina output i'm seeing a Logon
(35=A) followed by a Trading Session Status (35=h).
The Logon comes through, but every once in a while the
TradingSessionStatus doesn't.
The broker logs shows them sending me a heartbeat and a test request,
but i never see that (even in Mina layer) so eventually i'm timed out
and disconnected.

Is it possible that there's some race condition in the SSL decoding
code? i can't seem to trace where the entrypoint from Mina into QFJ
code is.

The log with QFJ and Mina messages attached.

-- 
Toli Kuznets
http://www.marketcetera.com: Open-Source Trading Platform
download.run.trade.

Re: [Quickfixj-users] Is anybody seeing dropped incoming SSLmessages?

[Brad H. <Bra...@gb...>]

Hi Toli,

Perhaps this is fixed by http://www.quickfixj.org/jira/browse/QFJ-179
for 1.1.1?=20

Brad.

-----Original Message-----
From: qui...@li...
[mailto:qui...@li...] On Behalf Of Toli
Kuznets
Sent: Tuesday, 5 June 2007 11:20 AM
To: qui...@li...
Subject: Re: [Quickfixj-users] Is anybody seeing dropped incoming
SSLmessages?

QuickFIX/J Documentation: http://www.quickfixj.org/documentation/
QuickFIX/J Support: http://www.quickfixj.org/support/


In the spirit of answering my own questions, I have figured out my
problem - it was a "user error" of sorts.
Turns out I had a deadlock: since this was a quick-and-dirty sample
app, i was stupid to send a MarketDataRequest from inside the
onLogon() method. As a result, i had a lock on the DataDictionary and
was waiting on SocketIoProcessor (mina's SSL processor) while the
SocketIoProcessor was receiving a new message and was in turn waiting
on the DataDictionary.

So rest assured, SSL (if used correctly) does not drop any incoming
messages.

Here's a corollary question: it's obvious that sending messages out in
onXXX callbacks can lead to deadlocks. Any advice/best practices on
how to handle that situation, aside from handing work out to other
threads/runnables?

thanks!

Re: [Quickfixj-users] Is anybody seeing dropped incoming SSLmessages?

[Toli K. <to...@ma...>]

Brad,

I was running from SVN HEAD, so looks like there still may be some
nascent deadlock issues possible, even given the current codebase.

my only changes were the ones i made for QFJ-188, but i don't think
that introduced any new synchronization problems.

On 6/4/07, Brad Harvey <Bra...@gb...> wrote:
> QuickFIX/J Documentation: http://www.quickfixj.org/documentation/
> QuickFIX/J Support: http://www.quickfixj.org/support/
>
> Hi Toli,
>
> Perhaps this is fixed by http://www.quickfixj.org/jira/browse/QFJ-179
> for 1.1.1?
>
> Brad.


-- 
Toli Kuznets
http://www.marketcetera.com: Open-Source Trading Platform
download.run.trade.

Re: [Quickfixj-users] Is anybody seeing dropped incomingSSLmessages?

[Steve B. <st...@te...>]

Hi Toli,

The Session implementation in the trunk doesn't synchronize the
data dictionary access. I had also modified the Session to release
locks on itself before any app callback. The ATApplication implementation
used for acceptance testing has asserts to check that no locks are
held. This should have eliminated the type of deadlock you are describing.
Can you take a look at the version of Session you are using to verify
that the data dictionary access is not synchronized?

Steve

> -----Original Message-----
> From: qui...@li... [mailto:quickfixj-
> use...@li...] On Behalf Of Toli Kuznets
> Sent: Tuesday, June 05, 2007 3:17 AM
> To: qui...@li...
> Subject: Re: [Quickfixj-users] Is anybody seeing dropped
> incomingSSLmessages?
> 
> QuickFIX/J Documentation: http://www.quickfixj.org/documentation/
> QuickFIX/J Support: http://www.quickfixj.org/support/
> Brad,
> 
> I was running from SVN HEAD, so looks like there still may be some
> nascent deadlock issues possible, even given the current codebase.
> 
> my only changes were the ones i made for QFJ-188, but i don't think
> that introduced any new synchronization problems.
> 
> On 6/4/07, Brad Harvey <Bra...@gb...> wrote:
> > QuickFIX/J Documentation: http://www.quickfixj.org/documentation/
> > QuickFIX/J Support: http://www.quickfixj.org/support/
> >
> > Hi Toli,
> >
> > Perhaps this is fixed by http://www.quickfixj.org/jira/browse/QFJ-179
> > for 1.1.1?
> >
> > Brad.
> 
> 
> --
> Toli Kuznets
> http://www.marketcetera.com: Open-Source Trading Platform
> download.run.trade.
> 
> -------------------------------------------------------------------------
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Quickfixj-users mailing list
> Qui...@li...
> https://lists.sourceforge.net/lists/listinfo/quickfixj-users

Re: [Quickfixj-users] Is anybody seeing dropped incoming SSL messages?

[Toli K. <to...@ma...>]

In the spirit of answering my own questions, I have figured out my
problem - it was a "user error" of sorts.
Turns out I had a deadlock: since this was a quick-and-dirty sample
app, i was stupid to send a MarketDataRequest from inside the
onLogon() method. As a result, i had a lock on the DataDictionary and
was waiting on SocketIoProcessor (mina's SSL processor) while the
SocketIoProcessor was receiving a new message and was in turn waiting
on the DataDictionary.

So rest assured, SSL (if used correctly) does not drop any incoming messages.

Here's a corollary question: it's obvious that sending messages out in
onXXX callbacks can lead to deadlocks. Any advice/best practices on
how to handle that situation, aside from handing work out to other
threads/runnables?

thanks!

(stack dump with deadlock attached for the curious).

On 6/4/07, Toli Kuznets <to...@ma...> wrote:
> Hi,
>
> I'm using QFJ to connect to a broker via SSL.
>
> Every once in a while, i get in a situation where the underlying MINA
> layer is seeing incoming FIX messages, but they don't seem to ever
> make their way up to the application layer.
> Connecting to the same broker without SSL, i'm seeing the same exact
> incoming messages, but they don't ever get dropped.
>
> For example, during connection in the Mina output i'm seeing a Logon
> (35=A) followed by a Trading Session Status (35=h).
> The Logon comes through, but every once in a while the
> TradingSessionStatus doesn't.
> The broker logs shows them sending me a heartbeat and a test request,
> but i never see that (even in Mina layer) so eventually i'm timed out
> and disconnected.
>
> Is it possible that there's some race condition in the SSL decoding
> code? i can't seem to trace where the entrypoint from Mina into QFJ
> code is.
>
> The log with QFJ and Mina messages attached.
>
> --
> Toli Kuznets
> http://www.marketcetera.com: Open-Source Trading Platform
> download.run.trade.
>
>


-- 
Toli Kuznets
http://www.marketcetera.com: Open-Source Trading Platform
download.run.trade.