Spec for numbering job-array elements from 1 (bug-1896): approved
requirements, a design reviewed over 20 rounds until clean (log in
research.md), and a 30-task plan. The design review found two live
apiserver defects, logged here: bracketed job_id/job_name filters are
never evaluated (bug-1898) and ControlJobByID re-dispatches earlier
selections (bug-1899).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
unsatisfy_() compared positional total/inuse/wanted arrays (entry i is
resource wanted->idxs[i]) with uncomitted[i], but the allocation
context's uncommitted vectors are dense and keyed by resource index.
From the second primary of a COMPOUND/ALTER job on, the check ignored
earlier primaries' claim on the same resource (over-admit) or charged
another resource's (wrong pend), with the matching wrong shortfall in
the pend reason.
unsatisfy_() now takes the vector and reads entry wanted->idxs[i]; an
index past its end (sim_loop.c's short all-zero vectors) or a NULL
vector counts as nothing uncommitted.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A job whose snapshot resreq fails to parse (any replay captured before
the bug-1888 fix) is kept by xce_sched_core_load_snapshot() with no
schedobj, but was still made a picker candidate -- by the loader and, in
simulation mode, by sim_bootstrap_picker()'s picker rebuild. The
allocator thread then dereferenced the NULL schedobj in
limit_job_election() and xce_scheduler --sim crashed.
Both push sites now skip such a job (it stays in the core so a restored
allocation is still charged), and sched_allocator() skips any picked job
without a schedobj as a backstop.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Feature validation GO (ASAN scheduler suite 96/96, smoke replay clean on
the allocation path). Roadmap entry marked implemented with follow-ups.
bug-1897 (found by the validation smoke run, not caused by this feature):
xce_scheduler --sim segfaults in limit_job_election on a job whose
snapshot resreq failed to parse (pre-bug-1888-fix replays).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Clean ASAN scheduler suite 96/96 with no expectation changes in existing
tests beyond wiring; scheduler invariant audit of the allocation path
clean (driver state-equivalence, core I/O isolation, picker/allocator
rendezvous, snapshot parity). issues/new_requirements.md item 4 marked
resolved.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Comments in scheduler sources and tests that presented the deleted
my_alloc()/parjob claim helpers as current code now name
sched_allocator_run() and its runner helpers; test helpers and tests
named after my_alloc are renamed. No logic or expectation changes.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
seqjob registers sched_allocator_run with seqjob_alloc_iface (no accepts
guard, plus its one-bucket plan iterator); its private my_alloc() is
gone. New fixtures pin the intended changes: a reason for every short
host, the bug-1893 shared reason, the host-stage OOM debit release, and
-EINVAL for missing inputs; seqjob's plan is asserted to be one bucket of
one slot. No other test expectation changed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
parjob registers sched_allocator_run with parjob_alloc_iface (accepts
maxTasks > 1, bug-1767, plus its plan iterator); its private my_alloc()
and claim helpers are gone. Characterization now pins the intended
changes: a reason for every short host (2/6/5) and the bug-1893 shared
reason. No other test expectation changed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
sched_finalize_job_alloc_() turns a satisfied plan's claims into the job
allocation exactly as parjob's commit tail does (host records from the
claim-time snapshots in claim order, shared records, jobid 0, no
duration) and publishes ctx->alloc only on success. Every out-of-memory
step frees what it built and releases the debits of claims not yet
turned into host allocations; each claim snapshot is freed exactly once.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The runner tries the plugin's plans in order and claims one distinct host
per bucket, with parjob's rollback and claim snapshots, nslots/nhosts per
candidate, a growable claims buffer and malformed-plan skipping. It
clears alloc_req.host before every candidate so each short host gets its
pend reasons (Req 3.6/3.7). Result assembly follows in task 3.3; until
then a satisfied plan's claims are released and the attempt pends.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
New alloc_runner.c: sched_allocator_run() validates its inputs (-EINVAL
without touching the pend reason), applies the plugin's optional accepts
guard, creates the pend-reason record, then checks shared resources: one
host-less NO_RESOURCE reason per unmet entry (skip only on vget < 0,
fixing bug-1893) and the met amounts counted once as uncommitted. Plan
search follows in task 3.2. Runner tests use a fake allocator contract;
the pend-reason factory audit now scans the runner.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
15 tests drive the real parjob plugin through job_alloc_fn: success,
short shared, pending on hosts, mid-plan rollback, task counts 0/1,
missing inputs, six OOM steps with their debits, COMPOUND. Values the
runner changes on purpose sit in *_today tests.
Spec: the task showed that parjob records host reasons only for a bucket
search's first rejected candidate (the rejected host's request stays in
alloc_req.host and the host iterator pre-filters later hosts against
it). Decision: the runner clears alloc_req.host before every candidate,
for both allocators, so every short host gets its reasons (Req 3.6/3.7);
publishing reasons only when they change is a follow-up.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
12 tests drive the real seqjob plugin through job_alloc_fn and record
today's results, reasons and debits, with switch-stable OOM injection
(sched_make_allocs stub, --wrap of make_host_alloc/make_job_alloc).
Assertions the runner changes on purpose sit in *_today tests.
The characterization showed bug-1893's real effect: vget() returns 0 on
success, so the shared-reason loop (rc <= 0 skip) never records a shared
NO_RESOURCE reason. Bug record and spec wording corrected; the runner
skips only on rc < 0.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
xce_sched_allocator_iface_t (optional accepts guard + plan_init/next/fini)
and the sched_allocator_run() entry point in sched/alloc_iter.h; the
plugin's const allocator_iface reference, forward-declared in plugin.h
with the picker_iface ownership rule. No plugin uses them yet.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
xce_alloc_plan_iter_t.init takes xce_sched_core_t* instead of
xce_scheduler_t*; both built-in generators, both allocators' call sites
and the contract-test stub follow. The parjob purity test now shows an
identical plan sequence for a NULL and a poisoned core.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Requirements reviewed and revised (seqjob per-bucket reset, nslots/nhosts,
bug-1893 fix in the runner), design and 11-task plan generated and reviewed.
bug-1892: unsatisfy_() reads index-keyed uncommitted amounts by position.
bug-1893: allocators' shared pend-reason loop reads past the unmet list.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The apiserver and all components ran the frontend Go tests as one list,
'cd frontend/apiserver && go test ... && cd ../xcecli && go test ...'.
set -e ignores a failure of any command in an && list except the last,
so a failing apiserver go test only short-circuited the list: the build
reported success and the xcecli tests never ran. Run each suite as its
own command in a subshell (which also keeps the cd from leaking).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The apiserver's job store learned of a new job only from the
controller's asynchronous OnJobCreated / array push, so a 'jobs kill'
(or query) straight after 'jobs submit' could answer 'job not found'
for the job just accepted (2 of 4 immediate kills).
services.SubmitJob now records the accepted job right after the
controller's reply (which comes only after the job is committed and
applied): one record per element, base+i for arrays, stored with the new
DataStore.PutIfAbsent and marked JobInfo.Provisional (json:"-", not a
column). An OnJobCreated that already arrived is never overwritten; the
push that arrives later replaces the provisional record -- including in
dispatchJobArray, whose duplicate check no longer treats a provisional
element as already known.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
xcecli prints a submitted job's ID in the dashed
{cluster}-{apiserver}-{timestamp}-{sequence} form (format.FormatJobID,
since d6008ead), but run.sh still matched 'Job <[0-9]+>' -- so 'jobs
submit' always failed and 'jobs kill' was skipped (28/1/1). It also
extracted the ID with [0-9]{10,}, which would have picked out only the
timestamp. Match the dashed form and take the whole ID from between the
angle brackets; 'jobs kill' accepts it back.
'jobs submit' reads a job script from stdin whenever stdin is a pipe,
so the suite hung forever when run from a script or CI with an open
pipe; the submit now reads /dev/null.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The script required a pre-generated <repo>/ha/data (cluster.bin + TLS
certs) that ee8473bd removed, and called deploy-ha.sh with that dir as an
extra argument; deploy-ha.sh now generates both into the deploy dir and
takes <binary> <top-dir> <peers>. It failed at its first precondition in
every checkout.
Behind that, both committed-LSN boundaries still assumed bug-845's doubled
submits (1 + 2*N_JOBS): the commit wait never reached it and walcompare
compared an LSN past every member's WAL. Since that fix each job is one
WAL entry, so the boundary is 1 + N_JOBS.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Two xce-build.sh runs against the same build dir raced: one run's
setup_build rm -rf hit files the other was writing ('Directory not
empty'), or deleted a binary the other had just built and tested.
xce-build.sh now takes an exclusive per-build-dir lock right after cd-ing
into src/ and holds it until it exits; a second run against the same dir
waits, runs against different dirs are unaffected. The lock
(src/tools/build-lock.sh) is an mkdir lock holding the owner PID, outside
the build dir, keyed by a hash of its absolute path; it needs only
coreutils (the nix dev shell has no flock) and takes over a lock whose
owner died. src/tools/test-build-lock.sh tests it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
zarray_get_ex() checked only idx >= caps, so a negative index read
array->data[idx] out of bounds -- reachable through a resource whose
registration failed and kept idx == -1 (xcelimit.c add_vres()). It now
returns NULL for idx < 0.
repl_info_resource_one() bounded a shared resource's slot by
sharedInst.count / sharedInUse.count, the number of populated slots, not
the capacity: a populated slot above that count showed as value=(none).
It now relies on zarray_get()'s own bounds check.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
bug-1580's evidence was a ':FAIL:' grep that cannot match Unity's
colored FAIL lines.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
sched_sim_driver_run() started the picker and allocator threads on
&sim->core and returned with them still running; only
sched_sim_driver_fini() stopped them, and 24 test files (35 call sites)
run the driver on a stack-local sim without calling it. The leftover
threads kept a pointer into the dead frame, and ASAN occasionally caught
one reading it (stack-use-after-return in cbuf_get) in
sched_sim_verify_test and sched_sim_record_output_test.
sched_sim_driver_run() now wraps the loop and stops both threads on
every return (new sched_sim_driver_stop_threads(), shared with fini). If
the allocator fails to start, the running picker is stopped too. The
stop no longer pushes a jobid 0 to wake the allocator: cbuf_get/put wait
in pthread_cond_wait, a cancellation point, and the fake end-of-session
was visible to the allocator.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A promoted vote endpoint carries a peer index relative to the votes
array (which leaves out the primary); a late-joiner's heartbeat endpoint
carries the absolute index. repl_ack_decode_and_dispatch recorded state
reports under the raw index, so with ctrl-1 primary, ctrl-3's vote
endpoint (relative 1) and a rejoined ctrl-2's heartbeat endpoint
(absolute 1) shared standby_reported_*[1]. Each handler then acted on
the other's position over its own endpoint: the snapshot meant for the
empty, rejoining ctrl-2 went to the caught-up ctrl-3, and ctrl-2 never
caught up.
Resolve the absolute index with ha_repl_ack_abs_peer_idx (as the
WAL-ack branch already did) for the state report and for
ha_walack_on_ack, and size catchup_inflight by MAX_HA_CANDIDATES so
every absolute index fits.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
With Nagle on, the small TLS record that ends a burst -- a packet header,
the last short chunk of an HA snapshot, a reply -- waited for the peer's
delayed ACK, ~40 ms on Linux: seen on 5 of 40 snapshot transfers. Set
TCP_NODELAY on every netio TCP socket, outbound (socket_create) and
accepted (socket_accept).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
SSL_get_error() reports SSL_ERROR_SSL whenever the calling thread's
OpenSSL error queue is non-empty. io_worker serves every endpoint from
one thread, so an error left by another connection -- a ctrlstatus
client's EOF queues 'unexpected eof while reading' in OpenSSL 3 -- turned
the next plain would-block SSL_read on a healthy connection into a fatal
error. handle_error() then tore that connection down; mid snapshot
transfer this left a rejoining standby with a truncated snap.part until
its liveness timeout forced a reconnect and a fresh transfer.
Clear the queue before each SSL_read/SSL_write/SSL_do_handshake so the
classification is about that call only.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
New specs from /kiro-discovery (2026-09-29), based on the LSF 10.1 Admin
Guide section "Job dependency and job priority". Dependency only; job priority
is deferred to a later discovery. Both are added to roadmap.md.
- job-dependency: -w expressions by job ID (done, ended, exit(id[,op code]),
started, post_done, post_err, bare id; && || ! and parentheses). One libxce
grammar, used from Go through cgo. The apiserver validates early (syntax,
parent exists, job:view). The controller re-validates, stores the
expression, and keeps an in-memory reverse index rebuilt on replay. The
scheduler evaluates in sched_job_is_ready(); unsatisfiable dependencies stay
PEND. Also closes the scheduler exit-code field-completeness gap. Adds
xcecli -w / modify -w -wn and harness jobsub -w / jobmod.
- job-dependency-insight: bjdepinfo-style query, dependency info in job
details, and optional orphan termination (-ti or a cluster grace period).
- controller-job-cleanup brief: job-dependency will replace the
job_has_dependents() stub.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
All required tasks are done and feature-level GO is recorded (d19e9f27);
only optional task 8.5 (observing transient catching-up live) remains.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Its false-ERROR-on-reconnect symptom was already fixed by bug-1854 and
bug-1886; today's HA soak shows zero scheduler ERRORs on reconnect.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
scheduler-sim-diagnostics, scheduler-sim-live-allocation, xce-test-cli-ux
and eliminate-frontend-makefile had every sub-task checked but their
parent boxes and spec.json phase still read tasks-generated.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
