Repository navigation
fork fleet: desk pool docs, a neutral example, the fork's pc pool on the desk link (D12) - #238
scuttletoby wants to merge 11 commits into
Conversation
0f9ca11 to
bb65599
Compare
e7c3fe6 to
f6b1bf0
Compare
bb65599 to
fbf95e3
Compare
f6b1bf0 to
821a225
Compare
…n the desk link (D12) The fork's own `pc` pool now launches through the desk link (fork-fleet-desk.py launch pc) instead of the never-made launch-pc wrapper. The runbook gets a made-up desk pool entry any project can copy, and the desk kit's README a five-step path from nothing to a working pool. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fbf95e3 to
ec41bd8
Compare
821a225 to
6a2c09b
Compare
|
Review round 1 at 6a2c09b (lens238): SAFE WITH NITS. Merging the
Checked fine: the example entry is neutral; |
Host step 3 installs the key tool and the README's step 2 names it; the example entry sits after the key line's paragraph; a host with an old pc dispatcher stops it and lets its runners finish first; the desk-key test drops the release's pc pool by name; the refresh test lists the desk link's files, not launch-pc. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Fix round 1 (commit 6678d64; D11b's round 1 merged in at 6bd6f2e and e7090ff; head e7090ff). Tests: desk-key 20 OK, refresh 43 OK (1 skipped), desk-seed 30 OK, desk 42 OK, desk-agent 31 OK, walls-wsl2 15 OK, install 28 OK, windows 11 OK (1 skipped, no pwsh), free_contract 13 OK, machine_file 17 OK.
The docs follow the fixed stack: per-pool seed path, the desktop's own limits, uninstall and drop-ins (merged from the lower branches). PR body updated. Still draft. |
|
Re-check lens 238 r1 at e7090ff: SAFE WITH NITS Earlier findings 1-5 FIXED: key tool installed in host step 3 ( The Nits:
def test_the_releases_pc_pool_is_a_desk_pool(self):
self.assertEqual(dk.desk_pools(dk.ff.load()), ['pc'])
Tests: desk-key 20, desk-seed 30, desk 42 OK; refresh |
…lines, another project's base (D12) - desk-key test: the release's pc pool is a desk pool (a launch typo shows in CI, not on the host) - desk README step 5: no "link is up" line; quote the real lines (runner started, pool answers again) - host step 3 and README step 2: another project points base at its own repository - README step 4: "the key the install printed" Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fix round 2Re-check r1 (SAFE WITH NITS), each nit:
Docs match round 2 of the stack below: this branch merges the round-2 D11b head (223795c). Its README and Tests at dcd20e2, run one file at a time: desk-key 21, desk-seed 41, install 34, desk 42, desk-agent 31, refresh 43 (1 skipped), walls-wsl2 15, windows 16 (2 skipped: no pwsh), free_contract 17, machine_file 17, all OK. |
Lens: re-check r2 (lens238r2), read at: dcd20e2 — SAFE WITH NITSRound-2 findings
Merging the New nit (low)
Nothing broken by the round; this PR's own diff is docs, tests and config only (the Tests: desk-key 21 OK, desk 42 OK; mutation run fails as expected. |
…sed pass (D12) The dispatcher logs "pool POOL answers again" only when the pool was left alone for more than one pass; one missed pass then an answer logs nothing. README step 5 and the desk link section now say so. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fix round 3Commit d8992da (head; D11b merged in as 833bd7c). Findings from lens238r2:
State question: this round changes docs only. Tests (local, at d8992da, whole stack merged): |
Lens: re-check r3 (lensd8minr3), read at: d8992da — SAFE WITH NITS
Tests: State: docs only; nothing new runs. |
…eet 0700 (D12) The desk kit's check and install hold /etc/fork-fleet to root's, 0700 and not a link (round 4 takes the install lock there); the by-hand block made it 0755. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fix round 4Re-check c6904850353:
Brief item "docs and READMEs must match round-4 behaviour":
Tests (head 4631043): desk-key 21 OK, refresh 43 OK (1 skipped), desk 42 OK, desk-seed 59 OK, gate-trust-docs all ok, fork-fleet 201 OK (7 skipped). |
Lens: re-check r4 (lensd8minr4), read at: 4631043 — SAFE WITH NITS
|
Fix round 5Merged #237 round 5 in b000dca (it carries #225's round 5). lensd8minr4
Tests (whole stack): desk-seed 63 OK, install 47 OK, desk-key 21 OK, refresh 43 OK (1 skipped), desk 42 OK. Still a draft. |
Lens: re-check (lensd8minr5), read at: b000dca — SAFERound-5 merge is clean:
|
Row D12 of the desktop pool plan, the last one: the docs that join D1 to D11 into one path from nothing to a working desk pool, a made-up example entry, and the fork's own
pcpool moved onto the desk link. Stacked on #237 (D11b).Why
The pieces are all there now (link, walls, agent, setup pack, Windows starter, free contract, seeds). What was missing:
pcpool using themUntil now, the
pcpool'slaunchpointed at a hand-made/opt/fork-fleet/launch-pcwrapper that runs commands over plain ssh. That wrapper was never part of the fork..github/fork-fleet.json: thepcpool'slaunchis nowpython3 -I /opt/fork-fleet/fork-fleet-desk.py launch pc, the same as any desk pool. Its_docsays so..github/fork-fleet.md:pcpool paragraph now says it is a desk pool, what turns it on (its dispatcher unit and the PC's desk key), and what happens while the PC sleeps (launchexits 3 at once, and its jobs go elsewhere).pcdispatcher on the wrapper first stops it and lets its runners finish (this config'scleanup skips a pool that is down, so it would never stop them); then it may delete the wrapper.
fork-fleet-desk-key.pywith the other desk files.waits: falseandslotsmean for a desktop.fork-kit/desk/README.md: "From nothing to a working pool" in five steps (project config, host once: host step 3then the desk link, desktop install, add the key, check). Step 5 quotes the dispatcher's real log lines: a runner
started on the desktop (
RUNNER: started for JOB, attempt N), andpool POOL answers againonly after the pool wasleft alone for more than one pass. There is no "link is up" line. Step 2 and host step 3 say that another project points
baseat its ownrepository.
What happens when this runs again, stops half-way, or meets existing, old or empty state?
routenever counts it and no job waits for it.pcruns with the new config, it calls the desk link instead of the oldlaunch-pcwrapper. The wrapper is then unused; the docs say to stop the oldpcdispatcher and let its runners finish first, then delete it. Until the PC's desk agent is installed and its key added,launchexits 3 and jobs go to other pools, the same as when the PC is off.pcpool with no dispatcher stays off, as before.General vs project
desktopandfleet-desktop-example) and the paragraph onwaits/slots.pcpool entry in.github/fork-fleet.json, which is this repository's own fleet config. No personal names, addresses or machine ids are in it.Upstream fit
Only fork-only files (
.github/fork-fleet.*,fork-kit/desk/README.md) change. Nothing in upstream T3 changes.It also lists the stack in
FORK_CHANGES.md(new on main since #223):deskin the host kits row with #225, and#229, #230, #236, #237 and #238 in the fleet runners row.
Tests
No new test file. Round 2 adds one test to
fork-fleet-desk-key.test.py: the release'spcpool is a desk pool, so atypo in its
launchfails in CI instead of on the host. The changed config is also checked by the existing tests:.github/scripts/fork-fleet.test.pyloads.github/fork-fleet.json, andcheck_configfinds no problems.fork-fleet-desk-seed.test.pyandfork-fleet-desk-key.test.pyread the pool list from it.Local run at round 2 (the whole stack merged in):
.github/scripts/fork-fleet-desk-key.test.py: 21 OK.github/scripts/fork-fleet-refresh.test.py: 43 OK (1 skipped).github/scripts/fork-fleet-desk-seed.test.py: 41 OK.github/scripts/fork-fleet-desk.test.py: 42 OK.github/scripts/fork-fleet-desk-agent.test.py: 31 OK.github/scripts/fork-fleet-walls-wsl2.test.py: 15 OKfork-kit/desk/tests/install.test.py: 34 OKfork-kit/desk/tests/windows.test.py: 16 OK (2 skipped: no pwsh here)fork-kit/common/tests/free_contract.test.py: 17 OKfork-kit/common/tests/machine_file.test.py: 17 OKRound 1: the desk-key test drops the release's own desk pool by name (
pc), not with the function under test, andasserts
vmstays;fork-fleet-refresh.test.py's list of root-run files names the desk link's files instead oflaunch-pc.Round 2: the test above; README steps 2, 4 and 5 and host step 3 reworded (see "What" above).
Red-checks
The new desk-key test guards the release config, which is already right, so it passes on round 1 too. With a typo in
the
pcpool'slaunch(fork-fleet-dsk.py), in a copy of the tree, it fails. Everything else is config and docs,with no new rule to break.
Round 3
fork-fleet.md("The desk link"):answers againis logged only after more than one missed pass; one missed pass then an answer logs nothing..torn.json, run deadline, disk budget) are in the README andfork-fleet.md.Round 4
/etc/fork-fleetas root's, 0700 and not a link (install -d -m 700), the same rule the kit'scheckand install use (4631043)./etc/fork-fleet; install refuses a linked or non-root-owned/opt/fork-fleet,/etc/fork-fleetor desk key; stricter runner-user checks; the owed-restart mark; Wine names its process after the.exe(fork-kit/common/FREE.md); the seed refuses link members it cannot make and odd sizes, retry times are approximate, a new copy on the host is tried at once, a lost list keeps the torn record, an unwritable.torn.jsonexits 1, a copy lives about 25 h, and the disk budget ismax_mb x (25 / every_hours, rounded up, + 2).Round 5
fork-fleet.mdparagraphs that do come from fork fleet: desk setup pack kit (D8) #225 and fork fleet: a desktop fetches its desk pool's seeds (D11b) #237 and say the round-5 rules (a new host copy skips the wait only once per wait; a runner user's main group must be the group named after it; a linked or non-rootunits,fork-fleet-wallsor other manifest folder is refused;checkcovers both halves of the desk key).Tar._find_link_target,unpack); odd sizes are refused (a member, and from round 5 an extended header, with a size below 0; a non-file with a size); retry times are approximate (5 min margin plus the timer's random delay); a lost list keeps the torn record (runreturns before reading it); an unwritable.torn.jsonexits 1 (TornLog.unsaved); a copy lives about 25 h (KEEP_OLD24 h plus the hourly prune); the disk budget formula matches. One part no longer holds: "a new copy on the host is tried at once" is now "at once, but only once per wait".🤖 Generated with Claude Code