It seems to me that when I load a data set which exceeds the available memory, paraview will hard crash. Is that expected?
Yes, no way around that. I guess we could monitor the memory and abort some way.
There’s nothing really we can do in practice at our scale. We can’t even message because that might allocate memory in the I/O subsystems. VTK/ParaView generally aren’t exception-safe, so that’s just going to leak memory all over the place too.
I also want to bring up one of ParaView’s best tools - View/ Memory Inspector. This way, you can see how full your memory actually is.
Also, I believe there is nuance to the question. If it is a remote server, the assumption is that ParaView is attaching to a cluster. Allocate more nodes. If you are builtin server, it should not crash, but rather start moving into swap (and running SLOWLY).
If you crash, depending on what your setup is, there are things you can try. One is don’t load variables. Don’t use filters that create a lot of data (such as Clip). Increase the size of swap provided by your OS.
Thank you all for your comments. So, size matters.
However, I have now found an input which still leads to a hard crash. There is a difference between 5.8.1 and 5.9: 5.9 crashes, the older version works correctly.
Unable to reproduce your crash with 5.9.0 or master.
Also unrelated to the initial topic at hands.
Not so,I’m afraid. It is still about hard crashes of 5.9.0 (Windows10).
Initially the hard crashes were observed with fairly large meshes, which then could explain the crashes by the requested memory and failure to deliver that memory.
At this point however I find that I cannot get 5.9.0 started with any VTK data set. Are there some settings that it may be possible to reset?
Thanks. I removed the settings, and 5.9.0 works again.
Petr, did you by any chance keep the settings files?
Also, you can do the same thing through the GUI using Edit/ Reset to Default Settings.