Fix crash in NTP servers GUI dialog on late network configuration

[rvykydal]

Resolves: rhbz#1938168

[rvykydal]

The crash happens when trying to update an iterator from the Gtk model for list of NTP servers after the iterator becomes invalid (by clearing the model):

• User enters DateTime Spoke -> on spoke refresh the model refresh (of ntp server state) is started: for iterator ITR_X from the model:
• THREAD_Y starts to check server pointed by ITR_X with ntp_server_workig() method
• User enters NTP configuration dialog -> model is reinitalized / cleared (servers/items are added from scratch)
• THREAD_Y finishes the ntp_server_working() check and tries to update the ITR_X which no longer exists => crash

The issue is hit when the ntp_server_working takes too long (5 secs), ie longer than the time between entering the spoke and opening the NTP configuration dialog. For example when networking is configured later in the installation (like just before NTP configuration).

The crash showed with this
rvykydal@212eb1c
RVDBG debug info added:

anaconda.crash.log

If the ntp_server_working function is fast, eg network is activated early the refresh on spoke entry is fast enough to prevent the crash:

anaconda.ok.log

Seems to be related also to this change #2468
In upstream the code was completely rewritten in commit 15d2b2f

I think the safe fix for rhel8-branch would be checking the validity of iterator before using it as in this PR. In case of invalid itr there is no action so I think even the result should be correct.

We could probably also remove the model refresh (

anaconda/pyanaconda/ui/gui/spokes/datetime_spoke.py

Line 651 in e8714d7

gtk_call_once(self._config_dialog.refresh_servers_state)

) from spoke refresh method because the refreshed values are used/available (I think) only when the dialog is entered, and entering the dialog does reinitialization of the model anyway. Perhaps the refresh became useless (malign) after the #2468. I haven't found any reason for its existence but we should think twice here.

[VladimirSlavik]

Looks good to me. Thank you! And sorry for breaking it.

I'm actually amazed that we keep an iterator in the asynchronous call instead of just the server hostname. But I guess it's understandable how this came to be.

[poncovka]

Looks good to me. Thanks!

[jkonecny12]

LGTM! Great work!

[JeongHyeon-Kim]

Thank for solving issue and record