myxce/src/scheduler/testsuite/test_maxtasks_wiring_sim_driver_parity.c

/*
 * test_maxtasks_wiring_sim_driver_parity.c - resreq-maxtasks-wiring spec,
 * task 3.2 remediation round (Req 1.4, 5.4, and design.md's "Project
 * Invariant Cross-Check" driver/simulation parity commitment).
 *
 * ── Why this is a SEPARATE file, not an addition to
 *    test_maxtasks_wiring_ingest.c ─────────────────────────────────────────
 * This task's driver/simulation-parity test previously lived in
 * test_maxtasks_wiring_ingest.c as
 * test_driver_and_simulation_style_jobs_push_apply_identical_task_count_
 * wiring(), an independent-review REJECTED design: it never executed
 * sim_loop.c's/driver.c's own `static` translators
 * (replay_event_to_core_event()/sim_process_replay_event_message()/
 * translate_one_pending_update()) -- it hand-built an xce_core_event_t by
 * MIRRORING the test author's own belief about those translators' field-
 * copy bodies, then fed that hand-built event into the already-covered
 * xce_core_push_events() path tests 1-3 of that file exercise. A real
 * future regression in either translator's own body (e.g. an incorrect
 * field name, a dropped assignment) would NOT have been caught -- the
 * mirror would just as happily reproduce the bug, or never notice a
 * divergence, since it never calls the translator at all.
 *
 * The reviewer identified that reaching the REAL, `static`
 * sim_process_replay_event_message() (and, through it, the REAL, `static`
 * replay_event_to_core_event()) is NOT impractical: test_sim_event_loop.c
 * ALREADY #include's ../src/sim_loop.c directly (a white-box #include,
 * not a link-only technique) and calls sim_process_replay_event_message()
 * with a hand-built, in-memory capnp ReplayEvent message -- no real
 * `.replay` file or full sched_sim_driver_run() boot required. This file
 * copies that exact technique (see test_sim_event_loop.c's own file header
 * comment for the fully-justified #include/stub set this file mirrors
 * verbatim).
 *
 * The reviewer also identified the genuine obstacle to folding this
 * directly into test_maxtasks_wiring_ingest.c: that file links the whole
 * real `scheduler_srcs` object list (needed for ITS OWN test 3, which
 * drives a real sched_init()'d allocator dispatch) -- white-box
 * #include-ing core_events.c/sim_loop.c into that SAME translation unit
 * would collide with meson's separate compilation of those same files into
 * the linked scheduler_srcs archive (duplicate-symbol link errors). This
 * file therefore lives on its own, using test_maxtasks_wiring_ingest.c's
 * task-3.1 fixture-helper NAMES (make_wire_queue()/make_push_job()) but
 * its own small, independent copies of them (this directory's own
 * established norm -- every white-box-#include test file in this
 * directory keeps its own self-contained fixture helpers rather than
 * sharing a header, e.g. test_sim_event_loop.c's own
 * build_fixture_scheduler() is not shared with test_sim_startup.c's
 * near-identical one either).
 *
 * Per design.md's own explicit language ("exercising this path through
 * sim_loop.c's own entry point") and this codebase's own cited history
 * (the scheduler-allocation task 5.2 incident: a fix applied only to the
 * real-driver path once caused a real driver/simulation divergence), the
 * SIMULATION side is this file's target -- driver.c's own real-driver-side
 * translate_one_pending_update() is already exercised for a JOBS_PUSH
 * payload by test_driver_translate.c's own test_translate_jobs_push()
 * (a real, non-mirrored proof on that side already exists); duplicating
 * that here would add risk (a second white-box #include-of-driver.c
 * translation unit alongside this one, its own separate link-surface to
 * verify) for no new coverage, so this file focuses on the one side that
 * genuinely lacked a real-code proof.
 *
 * ── What this test proves ───────────────────────────────────────────────
 * A real multi-task job (mintasks=4/maxtasks=6, distinct from every other
 * fixture value elsewhere in this feature's test suite) is carried inside
 * a hand-built, in-memory ReplayEvent{jobsPush} capnp message (built via
 * the SAME plain-C-struct + encode_ReplayEvent_ptr() technique
 * recording_events.c's own write_one_replay_event() uses for this exact
 * XCE_CORE_EVENT_JOBS_PUSH case in production -- not a hand-rolled capnp
 * builder walk). That message is written to a real temp file, reopened,
 * and fed through the REAL, `static` sim_process_replay_event_message()
 * (sim_loop.c) -- the same function sched_sim_driver_run()'s own read loop
 * calls for every ReplayEvent message in a genuine `.replay` file. The
 * xce_core_event_t THAT REAL FUNCTION populates (via its own internal,
 * equally-real call to replay_event_to_core_event()) is then pushed
 * through the identical xce_core_push_events()/core_build_job() ingest
 * chain this feature's fix lives in, and the resulting job's dispatched
 * standalone primary resreq is asserted to carry the job's correct real
 * task count (4/6), not parse_resreq_primary()'s own always-1 default.
 *
 * This directly catches a REAL regression in sim_loop.c's own translator
 * body: confirmed live (see "RED-phase verification" below) by temporarily
 * mutating replay_event_to_core_event()'s ReplayEvent_body_jobsPush case
 * and observing this test fail, then reverting.
 *
 * ── RED-phase verification performed while writing this file (not
 *    committed here) ─────────────────────────────────────────────────────
 * sim_loop.c's replay_event_to_core_event(), ReplayEvent_body_jobsPush
 * case, was temporarily mutated from:
 *   ev->u.jobs_push.njobs = src->body.jobs_push->count;
 * to:
 *   ev->u.jobs_push.njobs = 0;
 * This file's binary was rebuilt and run in isolation: the test FAILED at
 * the `TEST_ASSERT_EQUAL_INT(1, translated_ev.u.jobs_push.njobs)`
 * assertion (0 != 1) -- proving this test genuinely calls the REAL,
 * live sim_loop.c translator body, not a mirror of it (a mirror would be
 * unaffected by a mutation to the real function it never calls). The
 * mutation was then reverted and `git diff` against sim_loop.c confirmed
 * clean before this file was finalized.
 */
#include <errno.h>
#include <limits.h>
#include <stdatomic.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include <unity/unity.h>

#include "controller.capnp.h"
#include "kvmgr.h"
#include "resreq.h"
#include "scheduler.h"
#include "xstring.h"

/* Pull in the real functions under test into this TU (white-box #include),
 * the EXACT same set test_sim_event_loop.c's own file header comment
 * justifies file-by-file (see that file for the full per-file rationale
 * this comment does not repeat): the core/session/decision machinery
 * sim_loop.c's event loop drives, the core.h-scoped picker registration
 * path (plugin.c/builtin_fcfs.c), recording.c/recording_session.c/
 * sim_record.c (sim_loop.c unconditionally references their symbols),
 * repl.c and its siblings (sim_loop.c #includes repl.h, whose task 12.1
 * inline wrappers unconditionally reference xce_repl_start()/
 * xce_repl_post_session_check()), sim.c/sim_loop.c/sim_picker.c (the code
 * under test plus its bug-1567 sim_bootstrap_picker() sibling),
 * sim_verify.c (sched_sim_driver_run() references sim_decision_eq()/
 * sim_log_mismatch() as externs even though this file never drives that
 * function), and snapshot.c/snapshot_wire.c/snapshot_load.c (needed by
 * sim.c's own snapshot load). */
#include "../src/core.c"
#include "../src/core_clock.c"
#include "../src/core_events.c"
#include "../src/core_session.c"
#include "../src/decision.c"
#include "../src/plugin.c"
#include "../src/plugins/builtin_fcfs.c"
#include "../src/recording.c"
#include "../src/recording_session.c"
#include "../src/repl.c"
#include "../src/repl_fmt.c"
#include "../src/repl_rollup.c"
#include "../src/repl_break.c"
#include "../src/repl_info.c"
#include "../src/repl_params.c"
#include "../src/sim.c"
#include "../src/sim_loop.c"
#include "../src/sim_picker.c"
#include "../src/sim_record.c"
#include "../src/sim_verify.c"
#include "../src/snapshot.c"
#include "../src/snapshot_load.c"
#include "../src/snapshot_wire.c"

/* Link-satisfying stubs -- identical rationale/bodies to
 * test_sim_event_loop.c's own copies (schedobj.c/resource.c are not linked
 * here, same as that file). */

void kill_sched_job(void* arg) { (void)arg; }
void kill_sched_queue(void* arg) { (void)arg; }
void kill_sched_worker(void* arg) { (void)arg; }
void kill_sched_user(void* arg) { (void)arg; }
void kill_sched_project(void* arg) { (void)arg; }
void kill_sched_resval(xce_val_t* val) { (void)val; }

int sched_reconcile_job_resource_usage(xce_sched_core_t*  c,
                                       xce_job_account_t* a,
                                       xce_job_rusages_t* r)
{
    (void)c; (void)a; (void)r; return 0;
}

void sched_worker_release_host_ctx(xce_sched_worker_t* w) { (void)w; }

int sched_fill_limit_requests(xce_scheduler_t* sched, xvec_idval_t* vec,
                              xlist_t* limits)
{
    (void)sched;
    (void)vec;
    (void)limits;
    return 0;
}

int sched_account_jobs(xce_sched_core_t* core)
{
    (void)core;
    return 0;
}

int sched_resolve_job_picker(xce_sched_core_t* core, xce_sched_plugin_t** out)
{
    (void)core;

    if (out == NULL) {
        return -EINVAL;
    }

    /* No picker plugin is ever registered in this file's fixture -- this
     * test's Observable is the ingest-time resreq wiring, not picker
     * hook-firing (task 4.3's own test suite covers that exhaustively) --
     * so -EINVAL (core_events.c's own documented behavior on an
     * unresolved picker, "WARNs and proceeds with picker=NULL") is always
     * the correct, deliberate return here. */
    *out = NULL;
    return -EINVAL;
}

void setUp(void) { return; }
void tearDown(void) { return; }

#define JOBID_SIM_DRIVER_PARITY ((uint64_t)910701ULL)

/* ---------------------------------------------------------------------- */
/* Fixture helpers                                                         */
/* ---------------------------------------------------------------------- */

/* make_tmp_dir -- identical technique/rationale to
 * test_sim_event_loop.c's own helper of the same name. */
static void make_tmp_dir(char* out, size_t out_len, const char* suffix)
{
    char tmpl[PATH_MAX];

    snprintf(tmpl, sizeof(tmpl), "/tmp/sched_maxtasks_simparity%s_XXXXXX",
             suffix);
    TEST_ASSERT_NOT_NULL(mkdtemp(tmpl));
    TEST_ASSERT_TRUE(strlen(tmpl) < out_len);
    strcpy(out, tmpl);
}

/* make_wire_queue / make_push_job -- same shape and rationale as
 * test_maxtasks_wiring_ingest.c's own helpers of the same names (this
 * file's own small, independent copies -- see this file's header comment
 * for why this directory's convention is per-file fixtures, not a shared
 * header, for white-box-#include test TUs). Never torn down, same
 * directory-wide convention. */
static xce_queue_t* make_wire_queue(const char* name, int32_t priority)
{
    xce_queue_t* q = (xce_queue_t*)calloc(1, sizeof(*q));

    TEST_ASSERT_NOT_NULL(q);

    q->queue    = strdup(name);
    q->status   = QUEUE_STAT_OPEN;
    q->priority = priority;

    return q;
}

static xce_job_t* make_push_job(uint64_t jobid, char* resreq_str, char* user,
                                char* queue, char* project, uint32_t mintasks,
                                uint32_t maxtasks)
{
    xce_job_t* j = (xce_job_t*)calloc(1, sizeof(*j));

    TEST_ASSERT_NOT_NULL(j);

    j->jobid      = jobid;
    j->state      = JOB_STAT_PEND;
    j->sub_time   = 1000;
    j->queue_time = 1000;

    j->submit = (job_submit_data_t*)calloc(1, sizeof(*(j->submit)));
    TEST_ASSERT_NOT_NULL(j->submit);

    j->submit->queue        = queue;
    j->submit->user         = user;
    j->submit->project_name = project;
    j->submit->resreq       = resreq_str;
    j->submit->mintasks     = mintasks;
    j->submit->maxtasks     = maxtasks;

    return j;
}

/* append_jobspush_event -- appends one ReplayEvent{jobsPush} message to the
 * already-open stream `f`, using the SAME plain-C-struct
 * (xce_jobs_data_t wrapper) + encode_ReplayEvent_ptr()/write_capnp_message()
 * technique BOTH recording_events.c's own write_one_replay_event()
 * (production write-side, XCE_CORE_EVENT_JOBS_PUSH case) and
 * test_sim_event_loop.c's own append_jobstatechange_event()/
 * append_replay_session() (test-side precedent) already establish --
 * NOT a hand-rolled capnp builder walk. */
static void append_jobspush_event(FILE* f, uint64_t ts, xce_job_t** jobs,
                                  int njobs)
{
    xce_jobs_data_t      jobs_wrap;
    replay_event_t       src;
    struct capn          c;
    struct capn_segment* cs;
    ReplayEvent_ptr      p;

    memset(&jobs_wrap, 0, sizeof(jobs_wrap));
    jobs_wrap.n_total = njobs;
    jobs_wrap.count   = njobs;
    jobs_wrap.jobs    = jobs;

    memset(&src, 0, sizeof(src));
    src.ts              = ts;
    src.body_which      = ReplayEvent_body_jobsPush;
    src.body.jobs_push  = &jobs_wrap;

    capn_init_malloc(&c);
    cs = capn_root(&c).seg;
    encode_ReplayEvent_ptr(cs, &p, &src);
    capn_setp(capn_root(&c), 0, p.p);

    TEST_ASSERT_EQUAL_INT(0, write_capnp_message(f, &c));
    capn_free(&c);
}

/* ---------------------------------------------------------------------- */
/* The test                                                                */
/* ---------------------------------------------------------------------- */

/*
 * See this file's own header comment for the full narrative. Summary: a
 * real multi-task job's ReplayEvent{jobsPush} wire message is decoded and
 * translated by the REAL, `static` sim_process_replay_event_message()
 * (which internally calls the REAL, `static` replay_event_to_core_event())
 * -- not a hand-built mirror of either -- and the resulting real
 * xce_core_event_t is pushed through the identical
 * xce_core_push_events()/core_build_job() ingest chain this feature's fix
 * lives in, proving sim_loop.c's OWN translator code applies this
 * feature's task-count wiring correctly.
 */
static void
test_sim_process_replay_event_jobs_push_wires_real_task_count_into_resreq_primary(
    void)
{
    char                   path[PATH_MAX];
    FILE*                  f;
    xce_sched_sim_driver_t sim;
    xce_queue_t*           queue;
    xce_core_event_t       queue_event;
    xce_job_t*             job;
    xce_job_t*             job_ptrs[1];
    struct capn            c;
    capn_ptr               root;
    xce_core_event_t       translated_ev;
    int                    record_rc = 0;
    int                    translated;
    xce_sched_job_t*       sjob;
    xce_primary_resreq_t*  primary;
    int                    rc;

    /* A minimal sim driver -- same rationale as
     * test_sim_event_loop.c's own Test 1: only sim->core is exercised by
     * sim_process_replay_event_message() (via sim_advance_virtual_clock()),
     * so xce_sched_core_init() plus a frozen (speed=0) clock is a
     * sufficient, self-contained fixture. */
    memset(&sim, 0, sizeof(sim));
    rc = xce_sched_core_init(&(sim.core));
    TEST_ASSERT_EQUAL_INT(0, rc);

    atomic_store_explicit(&(sim.core.speed_num), 0, memory_order_release);
    atomic_store_explicit(&(sim.core.speed_den), 1, memory_order_release);

    /* Register a real queue the job references -- hand-built ingest, not
     * part of this test's own Observable (the jobsPush translator below);
     * same technique test_maxtasks_wiring_ingest.c's own fixtures use for
     * an identical, incidental purpose. */
    queue = make_wire_queue("normal", 1);
    memset(&queue_event, 0, sizeof(queue_event));
    queue_event.type                 = XCE_CORE_EVENT_QUEUE_UPDATE;
    queue_event.u.queue_update.queue = queue;
    rc = xce_core_push_events(&(sim.core), &queue_event, 1);
    TEST_ASSERT_TRUE_MESSAGE(rc >= 0, "fixture: real QUEUE_UPDATE ingest "
                                      "must succeed");

    /* A real multi-task job (mintasks=4/maxtasks=6, deliberately distinct
     * from every other fixture value elsewhere in this feature's test
     * suite, so this is a genuinely independent data point). */
    job = make_push_job(JOBID_SIM_DRIVER_PARITY, "type == any", "alice",
                        "normal", "proj1", 4, 6);
    job_ptrs[0] = job;

    /* ── Build a REAL ReplayEvent{jobsPush} capnp wire message on disk,
     * then reopen and decode it -- the SAME on-disk round trip
     * sched_sim_driver_run()'s own read loop performs for every message in
     * a genuine `.replay` file. ─────────────────────────────────────── */
    make_tmp_dir(path, sizeof(path), "_jobspush");
    strcat(path, "/event.replay");

    f = fopen(path, "w");
    TEST_ASSERT_NOT_NULL(f);
    append_jobspush_event(f, 42000ULL, job_ptrs, 1);
    fclose(f);

    f = fopen(path, "r");
    TEST_ASSERT_NOT_NULL(f);
    TEST_ASSERT_EQUAL_INT(0, capn_init_fp(&c, f, 0));
    TEST_ASSERT_TRUE(msg_root_has_shape(&c, &root, ReplayEvent_word_count,
                                        ReplayEvent_pointer_count));

    /* ── The proof-of-real-code step: drive this message through the
     * REAL, `static` sim_process_replay_event_message() (sim_loop.c) --
     * not a mirror of it. record_rc is always 0 here: sim.record_file is
     * NULL for this whole fixture (no --sim-output involved). ────────── */
    translated = sim_process_replay_event_message(&sim, &c, root,
                                                   &translated_ev, &record_rc);
    fclose(f);

    TEST_ASSERT_EQUAL_INT(0, record_rc);
    TEST_ASSERT_EQUAL_INT_MESSAGE(
        1, translated, "sim_loop.c's real sim_process_replay_event_message() "
                       "must translate a jobsPush ReplayEvent into a "
                       "usable xce_core_event_t");
    TEST_ASSERT_EQUAL_INT(XCE_CORE_EVENT_JOBS_PUSH, translated_ev.type);
    TEST_ASSERT_EQUAL_INT_MESSAGE(
        1, translated_ev.u.jobs_push.njobs,
        "the real translator must carry through the real njobs count -- "
        "this exact field is what a regression in "
        "replay_event_to_core_event()'s own body would corrupt");
    TEST_ASSERT_NOT_NULL(translated_ev.u.jobs_push.jobs[0]);

    /* NOT a pointer-identity check against the original `job` local: this
     * message went through a REAL on-disk capnp encode/decode round trip,
     * so decode_ReplayEvent() (per sim_process_replay_event_message()'s
     * own doc comment, sim_loop.c) deep-copies a brand-new xce_job_t/
     * job_submit_data_t pair out of the arena -- `job` itself was only
     * ever the SOURCE this test serialized from, never aliased back. The
     * real proof is that the decoded copy's own real fields (jobid, and
     * -- the point of this whole test -- mintasks/maxtasks below) match
     * what was serialized. */
    TEST_ASSERT_EQUAL_UINT64(JOBID_SIM_DRIVER_PARITY,
                             translated_ev.u.jobs_push.jobs[0]->jobid);

    /* ── Push the REAL translator's own output through the identical
     * xce_core_push_events()/core_build_job() ingest chain this feature's
     * fix lives in, and assert the resulting job's dispatched standalone
     * primary resreq carries the job's correct real task count -- proving
     * sim_loop.c's OWN, actual translator code (not a hand-written mirror
     * of it) applies this feature's wiring correctly. ─────────────────── */
    rc = xce_core_push_events(&(sim.core), &translated_ev, 1);
    TEST_ASSERT_TRUE_MESSAGE(
        rc >= 0, "pushing the real translator's own output through "
                 "xce_core_push_events() must succeed");

    sjob = sched_get_job(&(sim.core), JOBID_SIM_DRIVER_PARITY);
    TEST_ASSERT_NOT_NULL(sjob);
    TEST_ASSERT_NOT_NULL_MESSAGE(
        sjob->schedobj, "core_build_job() must have built job->schedobj "
                        "for the job carried by the real translator's "
                        "output");
    TEST_ASSERT_EQUAL_INT(XCE_RESREQ_PRIMARY, sjob->schedobj->tag);
    primary = sjob->schedobj->u.primary->resreq;
    TEST_ASSERT_NOT_NULL(primary);

    TEST_ASSERT_EQUAL_INT_MESSAGE(
        4, primary->minTasks,
        "the REAL simulation-driver translator path (sim_process_replay_"
        "event_message() -> replay_event_to_core_event(), sim_loop.c), not "
        "a hand-built mirror of it, must have carried the job's real "
        "minTasks through to the dispatched standalone primary");
    TEST_ASSERT_EQUAL_INT_MESSAGE(
        6, primary->maxTasks,
        "the REAL simulation-driver translator path (sim_process_replay_"
        "event_message() -> replay_event_to_core_event(), sim_loop.c), not "
        "a hand-built mirror of it, must have carried the job's real "
        "maxTasks through to the dispatched standalone primary");

    printf("test_sim_process_replay_event_jobs_push_wires_real_task_count_"
           "into_resreq_primary: PASS\n");
}

int main(void)
{
    UNITY_BEGIN();
    RUN_TEST(
        test_sim_process_replay_event_jobs_push_wires_real_task_count_into_resreq_primary);
    return UNITY_END();
}