Skip to content

docs: the 2026-09-14 explorer deploys (#182-#188) and the C6a/C6a2 roll scripts - #189

Merged
minuszka merged 2 commits into
mainfrom
docs/c6a-rollouts-20260914
Sep 15, 2026
Merged

minuszka merged 2 commits into
mainfrom
docs/c6a-rollouts-20260914

Conversation

@minuszka

@minuszka minuszka commented Sep 15, 2026 •

Copy link
Copy Markdown
Owner

Records the three explorer deploys that shipped the C6a formation-end work on 2026-09-14, and adds the operator scripts the two C6a fleet rolls actually ran.

Explorer deploys, 2026-09-14 (verified against journalctl -u deftrack-devnet and the VPS deploy.log/deploy-186.log, which log in CEST -- UTC below)

PR restart (UTC) tip at restart scope
#182 (bundled with #183-#185) 15:44:21Z 13370 formationEndHeight/isSchedulable, RPC-pool test, docs -- the C6a first roll
#186 17:50:05Z 13419 health endpoint: idle ticks now also move heartbeatAt, closing the false 503 after a >5 min block gap that the #181 deploy measurement had found
#188 19:10:57Z 13450 llmq_50_60/llmq_60_75 formationEndHeight 13536 -> 13488, matching Core #244's activation height; ran inside the fleet-roll window (19:09:40-19:27:16Z)

#187 (test/sync-overlap-activity) was test-only; no deploy belonged to it.

The C6a closure

Both Experiments runs are closed: fleet-rollout-c6a-243-2026-09-14 at 13439, fleet-rollout-c6a-end13488-2026-09-14 at 2026-09-15 08:57Z, endHeight 13781. Frozen outcome over 13439-13781: 19 formed / 3 failed / 1 pending, 0 penalised; ChainLock 343/343; fleet 160/160 on one chain, all 14583e7c; InstantSend 3/3 locked at close. The independent review's F1-F5 findings are summarized in the new log section, with links to the review and closure folders.

The 12 roll scripts (explorer ops/)

script runs on does
ops/c6a-seed-roll.sh, ops/c6a2-seed-roll.sh explorer VPS canary: seed + devnet2 (defcond-nobdb), --check preflight, post-checks running md5 and the re-read #231 key
ops/c6a-fleet-stage.sh, ops/c6a2-fleet-stage.sh jump host stages the fleet binary on every host: md5 as the host sees it, ldd, -version; installs nothing
ops/c6a-fleet-roll.sh + ops/c6a-fleet-roll-remote.sh, ops/c6a2-fleet-roll.sh + ops/c6a2-fleet-roll-remote.sh jump host / each host preflight, then host by host; stops at the first host that does not come back whole
ops/c6a-roll-at-height.sh, ops/c6a2-roll-at-height.sh jump host, nohup arms the roll for a target height and its hard-stop, polls two fleet hosts, fires the fleet-roll script
ops/c6a-s-check.sh, ops/c6a2-s-check.sh VPS, read-only end-height/retirement check: commitments by type and quorum height, PoSe, Sentinel, explorer rounds and health

Byte-verified with sha256sum against the copies actually executed on the VPS (/root/roll-c6a/, /root/roll-c6a2/) and the jump host (/root/roll-c6a/, /root/roll-c6a2/) -- 10 of 12 match exactly.

The two s-check.sh scripts differ from the executed copy, deliberately: npm run verify:shell (repo-wide ShellCheck gate, default severity, so info-level counts) flagged SC2086 in both -- unquoted $tip/$h, always numeric values from getblockcount/seq, never a word-splitting/globbing risk in this read-only script. Owner decision: add narrow # shellcheck disable=SC2086 directives with a reason, rather than leave CI red or widen the repo's shellcheck scope. One of the four spots needed a one-line split (a leading assignment moved onto its own line) because ShellCheck's next-line directive does not reach a second ;-joined command sharing that assignment's line -- verified locally before pushing. No quoting, logic, or output changed; sha256sum of the checked-in files (410c442f..., da8a9f29...) differs from the executed ones (ef0ce5a7..., 54f8af1d...) only by these added comment/newline bytes.

Result of the two rolls these ran: 162/162 daemons on the C6a end-height binary, one chain, fleet roll windows 15:03-15:22Z (13354->13361) and 19:09:40-19:27:16Z (13450->13459), both 16/16 hosts.

No addresses, keys or host names; the inventory lives on the jump host. The md5 constants name the C6a/C6a2 builds and have to be updated for any other roll.

🤖 Generated with Claude Code

…ll scripts

Records the three explorer deploys that shipped the C6a formation-end
work (#182 with #183-#185, #186, #188), verified against
journalctl/deploy.log on the VPS, and the closure of both C6a
Experiments runs (endHeight 13781, independent review F1-F5).

Adds the 12 operator scripts the two rolls actually ran, versioned so
the next roll starts from them rather than from a scratch directory;
byte-verified against the copies on the VPS and the jump host.
npm run verify:shell flags SC2086 (unquoted $tip/$h in numeric
contexts) in c6a-s-check.sh and c6a2-s-check.sh. Both are read-only
diagnostic scripts that never ran with those quotes, so the fix stays
comment-only plus one line split (a leading assignment moved onto its
own line so its disable directive actually binds -- ShellCheck's
next-line directive does not reach a second ';'-joined command on the
same physical line, verified locally). No quoting, logic, or output
changed; the checked-in copy still differs from the executed one only
by these added comment/newline bytes.
@minuszka
minuszka merged commit 4fca3cb into main Sep 15, 2026
6 checks passed
@minuszka
minuszka deleted the docs/c6a-rollouts-20260914 branch September 15, 2026 11:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant