csshx
csshx copied to clipboard
brockgr
Rebuild Grid on Retile After Losing Slave
What steps will reproduce the problem?
1. Create a machine cluster of 5 hosts, 4 valid, 1 erroneous (or, simply open
multiple slaves)
2. Whether by error or intentionally, kill a particular slave manually.
3. Weep at the gap in your grid.
What is the expected output? What do you see instead?
Is "retile" support to restore window locations, or account for added/lost
slaves? I have a 10 machine cluster with 2 hosts currently offline for their
own reason. I open the cluster that includes them, and nearly immediately two
slave windows die off. No amount of retiling/changing the grid, etc. will allow
me to reclaim this lost space. Instead, I have to get the cluster connection
string and manually strip out the two hosts I don't want.
I would expect "retile" to react to the changes in the number of active slaves,
and follow all the same rules of useable space, and use it all/as much as
possible.
Just in case it makes some strange difference, I do specify --screen 2 on init,
pushing slaves off on my external display.
What version of csshX (do a "csshX -v") are you using? On what operating
system version?
Current HEAD (173) from SVN.
OSX 10.6 x64 capable mac book pro.
Original issue reported on code.google.com by [email protected] on 8 Nov 2010 at 9:33
This should probably be enhancement, whoops. Unless it is a bug because you
intended it to do this.
Original comment by [email protected] on 9 Nov 2010 at 8:08
Nope - this is a bug. I think it's a race condition somewhere.
My expected behavior is to leave a gap when a window is closed by the user, or
due to some error (so you know something happened). However this gap should be
cleared when a retile is requested.
I have seen this happen from time to time, I just need to practice reproducing
it.
Original comment by gavin.brock on 10 Nov 2010 at 2:33
- Changed state: Accepted
- Added labels: Priority-High
- Removed labels: Priority-Medium
Go figure, I don't remember EVER having a problem reproducing it, I go to do it
and it works like you expect, sigh.
Is there anything specific I can do (besides finding a specific reproducible
example) when it does happen again?
Original comment by [email protected] on 10 Nov 2010 at 2:51
I went back to my actions when I first wrote up this bug, and it happens 100%
consistently with a 10 machine (w/ 2 broken) cluster I have.
So, my steps were not correct, but my expected output details are correct for
my being able to consistently reproduce the issue.
Original comment by [email protected] on 10 Nov 2010 at 3:03
This may not be a race condition as much as perhaps a miscalculation. I had a
series of windows arranged quite like the screenshot on the home project page,
then closed all but one. I issued a [r]etile, and the first window enlarged to
full height, but retained it's (narrow) width.
I think my original example in here is pretty well related to that to.
Original comment by [email protected] on 11 Feb 2011 at 8:37