Bug#1109: Print all In new window Long delays (i.e. several minutes) waiting for redraw ONLY in seamless window mode
Package: x2goserver Version: 4.0.1.19-3 Severity: critical
Server OS: RHEL6.6 Client OS: Windows 7 SP1 64-bit Server Packages: x2goserver-4.0.1.19-3.el6.x86_64 x2goagent-3.5.0.32-3.el6.x86_64 nxagent-3.5.0.31-1.el6.x86_64 (not sure if this is a dependency) nx-libs-3.5.0.31-1.el6.x86_64 (not sure if this is a dependency) Client Version: 4.0.5.2
This only occurs when starting a terminal in seamless window mode, and then launching the "problem" applications from that terminal. I imagine that I'd observe the same behavior if launching the application directly when the x2go session is created.
Certain applications experience catastrophically long delays (i.e. unresponsive for several minutes) during startup and even more so during reconnect (after a session is suspended). In particular, this happens with 2 Synopsys EDA (Electronic Design Automation) applications that I heavily rely on, dve and verdi. Obviously, you won't be able to test these because they require an expensive license, but I can tell you that they are older-style X11 applications that user X server-side font rendering and probably don't use a lot of other modern X client-side rendering techniques.
Luckily, I was able to find a free application that is similarly unresponsive when starting up in seamless window mode, but that runs fine in virtual desktop mode: *xcircuit*. This happens to be a free CAD program for drawing circuit schematics. Obviously, I can't be 100% that this X11 application is failing in the same way as my proprietary EDA tools, but it is something the x2go developers can look at regardless.
http://opencircuitdesign.com/xcircuit/
In order to diagnose the problem, I ran my applications under xtrace with the '--relative-timestamps' option in order to see the kind of X11 protocol traffic that was being generated. In both cases (i.e. my Synopsys EDA applications and xcircuit), there appear to be MASSIVE amounts of notifications and events going from the X server to the X client. It seems like the application's window is made up of a TON on individual rectangle primitives. If I filter the log file dumped out by xtrace to just show timestamps, I see HUGE jumps/discontinuities where there are several seconds or even minutes of apparent inactivity between consecutive messages. According to top, x2goagent is NOT consuming an excessive amount of CPU during these period of unresponsiveness.
I don't know much about the details of the X11 protocol, but it almost seems as if there is some kind of message buffer overrunning somewhere causing things to get out of sync, only recovering after some kind of watchdog or timeout mechanism kicks in. If my theory is correct and this happens repeatedly, it can add up to MINUTES of unresponsiveness. However, I'm not sure why this would happen only in seamless window mode. Certainly, one difference between seamless window mode and virtual desktop mode is that when reconnecting after a suspend, seamless windows are probably going to see a lot more events from the server (e.g. expose events, window manipulation notifications, etc.) because of interactions with the locally running windows which may be in a totally different state compared to when the x2go session was suspended. In contrast, when running in a virtual desktop, the applications are contained with a "box" that isolates it from the state of the local windows on the x2go client side.
It is important to note that, for all of the applications that exhibit this behavior, once the application has started up (or after suffering long delays during an x2go reconnect), the application is perfectly functional and responsive.
On Thu, Oct 27, 2016 at 8:37 PM, Thomas Esposito <tmesposito00@gmail.com> wrote:
This looks like a problem of nx. Can you please try this:
- run /usr/bin/nxagent -R -ac :5 (you'll see nothing)
- launch the problematic application: export DISPLAY=:5 /usr/bin/problematic-app
- kill -HUP <pid of nxagent> (suspend session)
- wait some seconds
- kill -HUP <pid of nxagent> (resume session)
Does this also show the delays? (it doesn't for me with xcircuit, but I seem to have some font trouble here. I get lots of 'Error: font encoding file missing for font "Helvetica"' messages).
Uli
Where should I be running this from? I tried this from within my existing x2go session, but I get the following error:
_XSERVTransSocketINETCreateListener: ...SocketCreateListener() failed _XSERVTransMakeAllCOTSServerListeners: server already running Error: Aborting session with 'Cannot establish any listening sockets - Make sure an X server isn't already running'.
Then I tried it from within a VNC session. I was able to run the experiment you described, but I did not experience delays either with xcircuit or with my problematic app.
My server is entirely headless BTW. In fact, its a RHEL 6.6 virtual machine running on my company's VDI infrastructure.
Is there any value to giving you an xtrace dump with relative timestamps? When I do that, and then sort the log in order of increasing timestamp values (reported in seconds), I see BIG discontinuities where there can be DOZENS of seconds between consecutive timestamps.
I'm still betting that the problem has to do with the volume of X protocol messages. However, I'm not sure why this would be different in rootless vs. virtual desktop mode (I don't see the problem in virtual desktop mode).
Are you able to reproduce the delays that I'm seeing when running xcircuit directly within a rootless x2go session (i.e. NOT with your nxagent experiment). The whole point of using xciruit was because I wanted to find a FREE application that exhibits the problematic behavior, so that the developers could more easily reproduce the behavior.
On Wed, Nov 16, 2016 at 4:42 PM, Ulrich Sibiller <ulrich.sibiller@gmail.com> wrote:
On 2016-10-27 21:37, Thomas Esposito wrote:
Hi,
I'm experiencing the same effect than you, but the software applications and operating systems are a bit different. In my case the slow window redraws happen if:
The client operating system is Ubuntu Linux and the local desktop environment is either Unity, XFCE or Mate. If I'm running Ubuntu Linux on the client machine and the local desktop environment is either KDE or LXDE, I don't experience the slow window updates. I'm running the X2Go server on Azure and it's running CentOS 6.8. The desktop environment on the server is XFCE.
As you also noticed, the slow window updates only happen with the single application mode/seamless window mode.
I haven't tested much with Windows on the client side but the same slow window updates that you can see just by moving terminal windows on top of each other is not there (when you first open one terminal with single application mode and then open another terminal from the first).
I installed xcircuit on the X2Go server and it starts up and works quite ok, although I don't have any design file to test with. We are working on the same EDA field so I'm interested in finding a solution to this problem.
Bug reports talking about the slow window updates: http://bugs.x2go.org/cgi-bin/bugreport.cgi?bug=1010 http://bugs.x2go.org/cgi-bin/bugreport.cgi?bug=1081
Best regards,
Aki
I don't have a design file to test xcircuit with either. I only used this because it was the first free application that I stumbled upon (after some trial and error) that exhibits this behavior.
In my case, after launching xcircuit from a gnome-terminal in rootless mode, ALL X applications (including the gnome-terminal) running within the x2go session become COMPLETELY unresponsive (e.g. no window redraws/updates) for 5 MINUTES to the point where it almost looks like a total system lockup. However, during this 5 minute freeze of the x2go session, I have no problems running other programs on the host (e.g. from an ssh or VNC session). Also, while my x2go session is apparently frozen, x2go agent is sipping about 0.3% of the CPU. Eventually, after being frozen for 5 minutes, my x2go session recovers and I can interact with the xcircuit application with no apparent delays. Redraws are quick (including when windows are uncovered from beneath other windows) and everything is speedy and responsive.
This really seems like some kind of catastrophic network issue where data is getting lost (e.g. buffer overflows?), and the session won't recover until 1 (or more) timeouts expire. Maybe it doesn't have to do with the client OS at all and is instead a function of network conditions (which would be very difficult to reproduce).
I haven't been able to test this with an x2go client running on Linux because our RHEL repo only has the packages necessary to run the x2go server, not the client.
On Fri, Nov 18, 2016 at 4:38 AM, Aki Ketolainen <akik@mykolab.com> wrote:
-
Aki Ketolainen -
Thomas Esposito -
Ulrich Sibiller