Difference between revisions of "X applications"

From CRIU
Jump to: navigation, search
(empty)
Line 1: Line 1:
 
[[Category:Plans]]
 
[[Category:Plans]]
 
[[Category:Empty articles]]
 
[[Category:Empty articles]]
 +
 +
== Problem ==
 +
 +
Doing C/R of an X-application is now impossible due to a simple fact -- all X-apps hold a connections to X-server and, sometimes, have a video-card-related device opened as a file or even mapped as a virtual memory region. Both have problems checkpointning and restoring.
 +
 +
== Connection to X-server ==
 +
 +
A connection to the X-server is typically a TCP (or Unix) socket that criu [[TCP connection|can handle]] by itself. And, in theory, we could dump an X-application and restore it on the same box making the X-server "think", that nothing has changed. Unfortunately this is not what "checkpoint and restore an X-app" is expected to be. The expectation is when after checkpoint the app window(s) disappear from the screen and after restore they pop-up on it again. Do to _this_ we need to talk to X-server no both sides -- checkpoint node and restore one.
 +
 +
On the checkpoint node we need to get the list of associated with the application resources that X-server maintains in memory to serve the application X-needs.
 +
 +
On the restore node we need to put the resources in question back to the X-server ... to make the app window pop-up on the screen again (in simple words).
 +
 +
Both tasks require some (maybe deep) knowledge about the X-server and X-protocol we do not currently have in our team.
 +
 +
== Video card device ==
 +
 +
Criu can handle some devices, in particular ttys and virtual /dev/null|zero|etc. ones. For a "more real" device  the situation is close to what has just been described -- a device _may_ have some state regarding the application referring it via file descriptor or a memory mapping. This state also has to be handled somehow.

Revision as of 14:58, 25 March 2019


Problem

Doing C/R of an X-application is now impossible due to a simple fact -- all X-apps hold a connections to X-server and, sometimes, have a video-card-related device opened as a file or even mapped as a virtual memory region. Both have problems checkpointning and restoring.

Connection to X-server

A connection to the X-server is typically a TCP (or Unix) socket that criu can handle by itself. And, in theory, we could dump an X-application and restore it on the same box making the X-server "think", that nothing has changed. Unfortunately this is not what "checkpoint and restore an X-app" is expected to be. The expectation is when after checkpoint the app window(s) disappear from the screen and after restore they pop-up on it again. Do to _this_ we need to talk to X-server no both sides -- checkpoint node and restore one.

On the checkpoint node we need to get the list of associated with the application resources that X-server maintains in memory to serve the application X-needs.

On the restore node we need to put the resources in question back to the X-server ... to make the app window pop-up on the screen again (in simple words).

Both tasks require some (maybe deep) knowledge about the X-server and X-protocol we do not currently have in our team.

Video card device

Criu can handle some devices, in particular ttys and virtual /dev/null|zero|etc. ones. For a "more real" device the situation is close to what has just been described -- a device _may_ have some state regarding the application referring it via file descriptor or a memory mapping. This state also has to be handled somehow.