We are using MariaDB in a Galera Cluster setup, and the 10.5.10 to 10.5.11 upgrade brought huge changes to the wsrep_sst_common and wsrep_sst_mariabackup scripts. These changes make them entirely FreeBSD-incompatible. The most important changes (to us) are - reliance on --tmpdir option to mktemp - is_local_ip always returns true in a jail The former makes the SST script fail because of invalid option. This is easily (and probably compatible with Linux) fixed by removing the --tmpdir option and rather prefixing mktemp with 'TMPDIR=$tmpdir ;' inside the same $() around line 935 of wsrep_sst_mariabackup: < xtmpdir=$(mktemp '-d' --tmpdir "$tmpdir") > xtmpdir=$(TMPDIR="$tmpdir" ; mktemp '-d') The latter I had to hack by removing the commonname option passed to socat on line 387 of the same file: < CN_option='localhost' > CN_option='' This is obviously not a fix, as I don't fully understand what's going on - but I did not have more time to spend on this as our cluster was down :( Anything you can do to help bring attention to this from upstream? It's really broken that they're making linux-ism assumptions all over the place.. And the change set between these two releases seems pretty big in these scripts. Alternatively, patches need to be put in place in the port itself, though I imagine this is a less desirable option long-term.
Raised upstream as https://jira.mariadb.org/browse/MDEV-26441 but I'm not confident it'll get fixed any time soon.
Is this sufficiently addressed by https://jira.mariadb.org/browse/MDEV-26360? Perhaps create another PR for the common-name issue?