Merge ~bryce/ubuntu/+source/postgresql-common:fix-lp1690432-hirsute into ubuntu/+source/postgresql-common:ubuntu/devel
Status: | Merged |
---|---|
Approved by: | Bryce Harrington |
Approved revision: | 76a556b784f2fa03e187fce1e8999a2bc969b27d |
Merge reported by: | Bryce Harrington |
Merged at revision: | 76a5b5d3fdd3fa738251b65b7b6df5fd94c58eaf |
Proposed branch: | ~bryce/ubuntu/+source/postgresql-common:fix-lp1690432-hirsute |
Merge into: | ubuntu/+source/postgresql-common:ubuntu/devel |
Diff against target: | 0 lines |
Related bugs: |
Reviewer | Review Type | Date Requested | Status |
---|---|---|---|
Sergio Durigan Junior (community) | Approve | ||
git-ubuntu developers | Pending | ||
Canonical Server packageset reviewers | Pending | ||
git-ubuntu developers | Pending | ||
Review via email: mp+398752@code.launchpad.net |
Description of the change
Only add postgres to group ssl-cert on new installs if postgres isn't already a member.
While this doesn't really "fix" a bug, it enables a way to work around a buggy situation. If postgres is installed in a place where the system config is unwriteable for some reason, then the postgres postinst fails when it tries to add postgres to group ssl-cert. This problem came up in creating the Docker image for postgres (maybe due to some glibc issue? still investigating). But the problem can crop up in other circumstances, such as when a user is maintaining their config with oneconf, but oneconf is broken for some reason.
There is an earlier test in the postinst for if postgres is in the postgres group, so I've used the same approach in implementing the test for the ssl-cert group. For this reason I think this stands a good chance of being taken by Debian, and I'll forward it their way if this looks like an acceptable solution to the problem.
PPA: https:/
I've verified in a privileged container that the package installs properly, however the original bug only appears in an unprivileged container. Unfortunately due to LP: #1916485 apt-get update fails when run a second time in an unprivileged container, so testing the fixed postgresql-common there doesn't seem to be feasible.