April 2026¶
Summary¶
A real testing sprint, not the quiet month the weekly logs suggested. The first three days were spent fighting Rocky 10's systemd namespace isolation on InfluxDB3 and a Proxmox molecule template-mapping bug (3 merged PRs, 2026-04-02/03) — the actual story behind these is CI-linter and test-matrix firefighting, which the commit titles alone undersell. Later in the month, solti-podman/solti-platforms shipped smart VM provisioning (auto-naming, auto-VMID, template selection) and fixed rootless Podman directory permissions.
What shipped (PRs)¶
- solti-monitoring #29 (merged 04-02) — Standardize venv management across the collection, aligning with other solti-* repos.
- solti-monitoring #30 (merged 04-02) — Rocky 10 InfluxDB3 fix: Rocky 9
only needed partial removal of
InaccessiblePathsin the systemd unit; Rocky 10 (systemd 257) requires complete removal, with auto-detection via journalctl. Also relaxed yamllint/ansible-lint configs and added a 2-minute metrics-collection wait to the Proxmox verify pipeline. - solti-monitoring #32 (merged 04-03) — Proxmox test template mapping
bug:
distribution_mapkeys didn't match actual template names (debian12-templatevsdebian-12-template— missing hyphen), breaking Debian 12/13 tests. Fixed, plus further Rocky 10 systemd diagnostics (PrivateMounts=no,MountAPIVFS=no, post-fix journal capture). Rocky 10 still failed after this PR — the service was trying to mount/run/systemd/inaccessible/sock, flagged as a known RHEL-derivative systemd-namespace issue requiring further investigation (tracked in open issue #31). - solti-platforms #4 (closed 04-06, not merged) — smart VM provisioning and UEFI improvements; superseded by solti-podman #25/#26 below.
- solti-podman #25 (merged 04-22) — Smart VM provisioning: template lookup by distribution name instead of VMID, auto-VMID allocation, auto-naming with prefix/suffix, Secure Boot for UEFI templates.
- solti-podman #26 (merged 04-23) — Rootless Podman directory
permissions (0750 → 0755) across 11 service roles, required for the
:Umount flag's uid/gid remapping to work. Root cause traced to a gitea deployment failure whereapp.iniwas inaccessible.
Commits (raw activity, full calendar month)¶
55 commits across 5 repos
| Repo | Commits |
|---|---|
| solti-podman | 34 |
| solti-monitoring | 14 |
| solti-platforms | 2 |
| jackaltx (root) | 3 |
| solti-matrix-bots | 1 |
| solti-matrix-mgr | 1 |
By repo (additional context beyond PRs)¶
jackaltx (root)¶
- Ignore collection tarballs (housekeeping)
- Reports for Matrix bot architecture analysis, OpenClaw removal verification, server state, watcher bot fix
- Claude Code Bot Haiku 4.5 upgrade report
solti-matrix-bots¶
- Initial commit — standardized Matrix bot deployment collection
Claude Code Sessions¶
No Claude Code session history available — local transcripts only go back to 2026-05-18. PRs #29/#30/#32 are all marked "Generated with Claude Code," so the original conversations describing why (the CI-linter sprint) existed at the time but are no longer recoverable — exactly the kind of loss this devlog is meant to prevent going forward.
Plans touched this month¶
- Claude Code Bot Phase 2 Test Plan (2026-04-18)
- Caddy Upgrade Plan (created early May, planned during this month)
Open issues (still unresolved as of this writing)¶
- solti-monitoring #31 — Rocky Linux systemd namespace issue not present in Podman deployment (opened 04-02, still unresolved after PR #32's attempt)
Weekly logs included¶
- Week of 2026-04-06
- Week of 2026-04-13
- Week of 2026-04-20
- Week of 2026-04-27
- Note: PRs #29/#30/#32 actually landed in the tail of Week of 2026-03-30, which spans the March/April boundary — they're reported here by calendar month, not by week-bucket.