Connections handling

[Krzysztof Chodak]

Rails 3.2.2 with remote SQL Server 2005. When I lost sight of DB (VPN down) my rails server (thin development environment) crashes. Anyone else having it?

[Clifford Heath]

Just off the top of my head, it sounds like FreeTDS might not be rigorously checking for errors on the socket.
If this results in it writing on a socket that's been closed, it can result in a signal that will kill the app.

Try adding a catcher for SIGPIPE and see if it gets called.

Clifford Heath.

On 08/03/2012, at 6:02 AM, Krzysztof Chodak wrote:

> Rails 3.2.2 with remote SQL Server 2005. When I lost sight of DB (VPN down) my rails server (thin development environment) crashes. Anyone else having it?
>

> --
> You received this message because you are subscribed to the Google Groups "Rails SQLServer Adapter" group.
> To view this discussion on the web visit https://groups.google.com/d/msg/rails-sqlserver-adapter/-/EGSJGo4iiHQJ.
> To post to this group, send email to rails-sqlse...@googlegroups.com.
> To unsubscribe from this group, send email to rails-sqlserver-a...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/rails-sqlserver-adapter?hl=en.

[Krzysztof Chodak]

I need to add that I am on Windows 7 32B so I guess SIGPIPE won't work for me, right?

On Wednesday, March 7, 2012 9:33:20 PM UTC+1, Clifford Heath wrote:

Just off the top of my head, it sounds like FreeTDS might not be rigorously checking for errors on the socket.
If this results in it writing on a socket that's been closed, it can result in a signal that will kill the app.

Try adding a catcher for SIGPIPE and see if it gets called.

Clifford Heath.

On 08/03/2012, at 6:02 AM, Krzysztof Chodak wrote:

> Rails 3.2.2 with remote SQL Server 2005. When I lost sight of DB (VPN down) my rails server (thin development environment) crashes. Anyone else having it?
>
> --
> You received this message because you are subscribed to the Google Groups "Rails SQLServer Adapter" group.
> To view this discussion on the web visit https://groups.google.com/d/msg/rails-sqlserver-adapter/-/EGSJGo4iiHQJ.

> To post to this group, send email to rails-sqlserver-adapter@googlegroups.com.
> To unsubscribe from this group, send email to rails-sqlserver-adapter+unsub...@googlegroups.com.

[Clifford Heath]

On 08/03/2012, at 7:36 AM, Krzysztof Chodak wrote:
> I need to add that I am on Windows 7 32B so I guess SIGPIPE won't work for me, right?

Ahh, probably not. But there'll likely be an equivalent thing, I forget my Windows socket programming.
Nothing likes being asked to write data on a closed socket though...

> On Wednesday, March 7, 2012 9:33:20 PM UTC+1, Clifford Heath wrote:
> Just off the top of my head, it sounds like FreeTDS might not be rigorously checking for errors on the socket.
> If this results in it writing on a socket that's been closed, it can result in a signal that will kill the app.
> Try adding a catcher for SIGPIPE and see if it gets called.
>
> Clifford Heath.
>
> On 08/03/2012, at 6:02 AM, Krzysztof Chodak wrote:
>
> > Rails 3.2.2 with remote SQL Server 2005. When I lost sight of DB (VPN down) my rails server (thin development environment) crashes. Anyone else having it?
> >
> > --
> > You received this message because you are subscribed to the Google Groups "Rails SQLServer Adapter" group.
> > To view this discussion on the web visit https://groups.google.com/d/msg/rails-sqlserver-adapter/-/EGSJGo4iiHQJ.

> > To post to this group, send email to rails-sqlse...@googlegroups.com.
> > To unsubscribe from this group, send email to rails-sqlserver-a...@googlegroups.com.

> > For more options, visit this group at http://groups.google.com/group/rails-sqlserver-adapter?hl=en.
>
>
>
>
>
>
>

> --
> You received this message because you are subscribed to the Google Groups "Rails SQLServer Adapter" group.

> To view this discussion on the web visit https://groups.google.com/d/msg/rails-sqlserver-adapter/-/iMiuy_txKXUJ.
> To post to this group, send email to rails-sqlse...@googlegroups.com.
> To unsubscribe from this group, send email to rails-sqlserver-a...@googlegroups.com.

[Ken Collins]

We have some tests in TinyTDS for this. In general we return INT_CANCEL for the server down message so we can keep on trucking.

https://github.com/rails-sqlserver/tiny_tds/blob/master/ext/tiny_tds/client.c#L57

This is supposed to compliment the adapters connection handling too by checking the state of the connection and dropping it if needed and creating a new one. Here lately I have been appending this info I recently found on the bottom of connection related issues. It covers keep alive settings and may be useful.

http://blogs.msdn.com/b/sql_protocols/archive/2006/03/09/546852.aspx

It causes good connection to break during transient network failures.
So configuring the keep-alive values too small is not recommended

[Krzysztof Chodak]

Now I got crash due to timeout:

C:/Ruby193/lib/ruby/gems/1.9.1/gems/tiny_tds-0.5.1-x86-mingw32/lib/tiny_tds/client.rb:68:in `connect': Adaptive Server connection timed out (TinyTds::Error)

but why it hangs whole server (now switched to thin) instead of fail to handle requests?

[Krzysztof Chodak]

Now i got full blown ruby crash two times an hour

windbg gives me the following

Probably caused by : tiny_tds.so ( tiny_tds!Init_tiny_tds+3260f )

using tiny_tds (0.5.1 x86-mingw32)

where could I get the source to analyze?

[Krzysztof Chodak]

Ok, found some info in  https://github.com/rails-sqlserver/tiny_tds

I will see what I can do myself

[Krzysztof Chodak]

tiny_tds compiled

ruby still crashing (Probably caused by : tiny_tds.so ( tiny_tds!Init_tiny_tds+31ecc ))

any hints on how to get ruby_debug19 working on ruby 1.9.3p125 (2012-02-16) [i386-mingw32]?

ruby_debug.c:29:19: error: conflicting types for 'rb_iseq_compile_with_option'

C:/Ruby193/include/ruby-1.9.1/ruby-1.9.3-p125/vm_core.h:505:7: note: previous declaration of 'rb_iseq_compile_with_option' was here

was trying with

gem install ruby-debug-base19x --pre

and other hints from http://stackoverflow.com/questions/8835522/unable-to-install-gem-ruby-debug19-on-win7-with-installd-ruby-1-9-2-and-1-9-3

but with no luck

[Krzysztof Chodak]

Hmm, after reboot it builds fine

Debugger ready however it crashes too

I should expect that as it is whole ruby.exe that is crushing

What next? Anyone? As you I have gone some path alone but now i really need advise

Neem some sleep now...

[Ken Collins]

I am not sure where I can help. Can you give me some context within running the tests in the project? Did you find my previous reply helpful? Not? In what way one or the other?

- Ken

[Krzysztof Chodak]

tiny_tds tests are passed

I use remote 2005 sql through juniper vpn mainly issuing procedures returning more than one recordset (+ standard activerecord queries)

I can only guess that crashes are due to connectivity issues

I will try to move database to be accessed locally and see how it works then

[Ken Collins]

OK, did you find my keep-alive info helpful? Can you follow up and let me know how/if that plays out?

- Ken

[Krzysztof Chodak]

30 sec keep-alive was on all the time. 

I will try to catch the crash on wire using wireshark to check whether connection is alive during the crash or no.

[Krzysztof Chodak]

Now I know what is happening on the wire: server resets the connection. The question is why?

What is strange is that there are some server responses saying that incomming packets were malformed.

Maybe it is dropping connection after defined number of malformed requestes (here 9)?
I have put pcap file (with first couple of login frames skipped) below - maybe you could have a look:
https://docs.google.com/open?id=0B8Pbr6Jr1eaCa1R0NnVfdnpUb1M4a0pVVGl3WEZTZw 

For sure there is a problem in library with handling socket resets but there could be another one with malforming requests?

[Ken Collins]

I have no idea what a pcap file is. Have you tried looking at the FreeTDS log using DTSDUMP env var (http://www.freetds.org/userguide/logging.htm)? This could be a FreeTDS issue, but I have consistently seem them re-verify that solid base the network protocol is on. That said, a search on their forums and/or a question with logs from the dump file above would be a good way to start a conversation. If indeed there is something I can do in TinyTDS, I am all ears.

 - Ken

[Krzysztof Chodak]

pcap file is a file containing network packets: here filtered to 1433 port traffic

wireshark is a program used for capturing but also for analyzing traffic

It should be similar to dtsdump output however it has all lower level info like socket signals (here RST) etc

I have sent inquiry to freetds mailing list

Thank you for your support

[Ken Collins]

Cool, sorry I could not be more helpful. Your at a level that is just not in my comfort zone. I see your thread on the FreeTDS list too and will monitor it there.

- Ken

[Krzysztof Chodak]

Why you are not comfortable? :)
I highly recommend trying wireshark - it is a very generic and
straightforward tool to debug any network related issues or trace what
the application is ever doing on the wires.
I was very surprised to see plain text password being sent using TDS
protocol. I was expecting at least a hash...

On Thu, Mar 22, 2012 at 21:29, Ken Collins <k...@metaskills.net> wrote:
>
> Cool, sorry I could not be more helpful. Your at a level that is just not in my comfort zone. I see your thread on the FreeTDS list too and will monitor it there.
>
>  - Ken
>

> --
> You received this message because you are subscribed to the Google Groups "Rails SQLServer Adapter" group.

[Ken Collins]

FreeTDS has kerberos and SSH options for logins.

 - Ken

[Krzysztof Chodak]

Ken, it looks like this old story already reported in freetds threat is still happening

I have commented on https://github.com/rails-sqlserver/tiny_tds/issues/79

Is it freetds bug still unresolved or what? I am keen to help to solve it somehow: got at least two application crashing from it