rdmpw3265m-changelog-20260916-1737-post-reboot-resume
rdmpw3265m changelog — post-reboot resume
[2026-09-16 17:34:29 EDT → 2026-09-16 17:37:29 EDT · rdmpw3265m · claude session 9a48ec92]
Context
rdmpw3265m booted about 16:43 EDT. rdmsm4x was up 24 min at 17:35, so both rebooted. No work was cut off here: the fleet lead session d40d3ead had finished its turn at 10:44 EDT, and /init (476df9de) had finished before the reboot. herdr had already auto-resumed 4 idle sessions. Budget advice was proceed (Anthropic 7d at 45%, vendor figure).
Post-reboot verification (measured)
- tyrelld pid 2082, tyrellbar 2110, herdr 2097 running. localhost:43117/api/status answers, role=client.
- All 6 hosts are online on Tailscale and answer ssh. jdmbair13m5 is back (it had been offline since 09-15 21:34), and its Data volume is 99% full with 12Gi free.
- Dedupe with peers: rdmsm4x/b3123f57 has TASK-20260916-04 only. rdmpw3275m/e5c848ec is finished, with no overlap.
Changes
- ISSUE-20260905-12: added the ssh ControlMaster stanza on jdmbair13m5, the last unmultiplexed spoke. Backup is ~/.ssh/config.bak-20260916-173627. Verified by protocol (ssh -O check plus the -v mux line) on 5/5 spokes after the reboot.
- Gauge correction on that ticket: after the reboot the hub labels the privileged half "[postauth]", not "[priv]". This morning's recommended grep now returns 0. Real count: 12 connections (12+12 procs; launchctl sshd instances = 12).
~/dev/apps/CLAUDE.md: fetched from rdmsm4x (51,935 B, sha1 c7fd62fd, matches). This repairs the brokenapps/bitROO/CLAUDE.mdsymlink.~/dev/README-CANONICAL-DEV.md: added a dated correction.~/dev-archive-20260831/exists on neither this host nor rdmsm4x. Backup:_backups/claude-md-20260916/README-CANONICAL-DEV.md.20260916(22 → 24 lines).~/dev/CLAUDE.md: updated both of the notes above.
My own error
In the ticket comment I wrote times (17:40) without reading a clock, and "17:38" in CLAUDE.md. Both are corrected: a second ticket comment, and CLAUDE.md now uses the file mtime 17:36:01.
Still open
- Fleet health probe needs
ssh -O checkper spoke. mem0 over HTTP (FEAT-20260905-04) is the structural fix.