Groupcast commands in zigbee

Yes, I check the return status of the API.
In the first case which I said (back to back group remove and scene remove), si_send _packet API on Linux Gateway returns failure.
In the second case (scene recall), I get gs_generic_cnf: Status SUCCESS. But after that I can't send any more commands.

When you send command back to back, do you mean you send command continuously in short period?

Yes, immediately after I get gs_generic_cnf for group remove command I send scene remove command.

Try not to send command immediately. Send a delay such as 100 ms to send next command.

Ok, adding a delay works.

BUt I have another problem which was not observed before.
I send scene recall commands in groupcast mode
After a scene recall command my module becomes unresponsive. I dont understand where I am going wrong.

I cannot judge why your device gets stuck from your descriptions. I would suggest you to set breakpoint to debug your device to know why it gets stuck.

I found the problem.
If I change the following macro
#define ZCL_GEN_SCENE_EXT_LEN 8
from 31 to 8, my scene_recall callback runs in a while(true) loop
So I changed it back to 31. Now it works.

But I dont understand the reason of it being run in while(1) loop.

Acoording to this:
// The maximum length of the scene extension field:
// 2 + 1 + 1 for On/Off cluster (onOff attibute)
// 2 + 1 + 1 for Level Control cluster (currentLevel attribute)
// 2 + 1 + 11 for Color Control cluster (currentX/currentY/EnhancedCurrentHue/CurrentSaturation/colorLoopActive/colorLoopDirection/colorLoopTime attributes)
// 2 + 1 + 1 for Door Lock cluster (Lock State attribute)
// 2 + 1 + 2 for Window Covering cluster (LiftPercentage/TiltPercentage attributes)

If I am just using on/off and level control, 8 bytes should be enough.
But then it results in that problem.
Can you suggest what is going wrong?

I think the max length (31) is reserved for (2 + 1 + 11) bytes of Color Control cluster plus ZCL_GEN_SCENE_NAME_LEN=16.

But why can't I modify it if I am not using color control cluster?

I am not sure about this part. Maybe it has to reserve max length after all. You have to trace source code to know the reason.

Ok,

About the previous suggestion you gave:

"Try not to send command immediately. Send a delay such as 100 ms to send next command."

Is there any other option? 

because 100ms delay in between a train of commands will increase system latency in our application.

I can tell you the failure message again:

"Busy -  please wait for previous operation to complete"

"gs_remove_scene: Error: Could not send msg."

Can you please check again and suggest if  there is any other option than introducing a delay?

Maybe you can try to decrease the in-between delay to 20 ms.