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.

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.