Same issue here (I'd posted a comment on the thread Kevin started as well).
I have some notes that indicate that things were working correctly with appliance RPM
"ovirt-engine-appliance-4.5-20240606072424.el9.x86_64" (2024-June-6 nightly,
tested on 2024-June-7). Unfortunately it looks like the nightly repo keeps only the last
10 builds and these were 2024-Aug-8 through 2024-Aug-14) so I cannot reproduce exactly
that scenario.
Based on some testing against the latest available nightly
"ovirt-engine-appliance-4.5-20240817071039.1.el9.x86_64.rpm" (2024-Aug-18) the
behaviour of engine-config changes after the "dnf update *" executed on the
appliance VM during the build.
Using "he_pause_before_engine_setup", I can confirm that engine-config --list
states that there is no database available.
Immediately after performing the dnf update, the same command fails with the "Index 1
out of bounds for length 1" error.
I compared behaviour against the latest appliance release
("ovirt-engine-appliance-4.5-20231201120201.1.el9.x86_64.rpm", modified for the
CS9 repos being moved to vault). In 20231201, we find:
- at he_pause_before_engine_setup, 'engine-config --list' reports no database
available.
- after 'dnf update *', 'engine-config --list' reports no database
available.
- after engine-setup, 'engine-config --list' reports all key-values in the
database.
With the baseline behaviour from 20231201, it definitely looks like simply _updating_
packages on the most recent nightly appliance breaks engine-config. This seems weird to
me - and it seems like some sort of change between 2024-June and today has introduced an
issue.
I am tempted to attach a Java debugger but I don't have any of the development
toolchain for this project available. Does anyone know whether it is worthwhile
troubleshooting at this level?
Would there be someone who knows engine-config codebase well enough to take a look? There
were a couple of recent changes as far as I can tell (79c665d and 9a9b4c6).