lei.chen/myxce
1
0
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix(scheduler): allocate every primary of a COMPOUND, not only the last (bug-1863)

Browse source 
Each primary's allocator (seqjob/parjob my_alloc) sets ctx->alloc to a
fresh job allocation covering only its own hosts, so a multi-primary
COMPOUND ("1*{A} + 2*{B}") dispatched only the last primary's hosts --
leaking the earlier allocations while every primary's hosts had been
debited in ctx.uncomitted.

sched_make_compound_alloc_ now folds each primary's allocation into one
job allocation via the new libxce merge_job_alloc(): a host placed by
several primaries becomes one entry with per-resource values added
(matched on resname, instid and reserve, so distinct instances are never
summed); new hosts and shared allocations are appended/merged the same
way. Partial-failure rollback remains bug-1766.

Tests: "1*{..} + 2*{..}" on 2x2 slots now allocates 3 slots across two
distinct hosts; "1*{..} + 1*{..}" yields one host entry with 2 slots
(both kept only the last primary before). Scheduler suite 87/87, lib 1/1.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Rongsong Shen 2026-09-25 13:14:28 +00:00
parent a6a0502d62
commit aea3473823
Signed by untrusted user: rshen
GPG key ID: 80533536B0EB651C
4 changed files with 176 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

28
src/scheduler/src/scheduler.c
View file

@ -280,9 +280,10 @@ static int sched_make_compound_alloc_(xce_sched_core_t* core,
xce_sched_obj_t* obj,
xce_compound_resreq_t* resreq)
{
int i;
xce_sched_ctx_t ctx;
int rc = 0;
int i;
xce_sched_ctx_t ctx;
xce_job_alloc_t* merged = NULL;
int rc = 0;
sched_ctx_init(&ctx, core, obj);
@ -295,10 +296,27 @@ static int sched_make_compound_alloc_(xce_sched_core_t* core,
if (rc <= 0) {
break;
}
/* bug-1863: each primary's allocator sets a fresh ctx.alloc for its
* own hosts only -- fold it into the whole compound's allocation
* (every primary's hosts are already debited in ctx.uncomitted) */
if (ctx.alloc != NULL) {
if (merged == NULL) {
MOVE_PTR(merged, ctx.alloc);
} else {
int mrc = merge_job_alloc(merged, ctx.alloc);
free_JobAlloc_ptr(&(ctx.alloc));
if (mrc < 0) {
rc = mrc;
break;
}
}
}
}
if (ctx.alloc != NULL) {
MOVE_PTR(obj->alloc, ctx.alloc);
if (merged != NULL) {
MOVE_PTR(obj->alloc, merged);
}
if (ctx.reason != NULL) {

52
src/scheduler/testsuite/test_parjob_e2e_multislot.c
View file

@ -797,6 +797,56 @@ static void test_alter_falls_back_to_alternative_that_fits(void)
"the job must be placed by the second alternative (1 slot)");
}
/*
* bug-1863: each primary of a COMPOUND replaced ctx->alloc, so the
* dispatched allocation kept only the LAST primary's hosts while every
* primary's hosts had been debited. The whole compound must be allocated.
*/
static void test_compound_allocation_includes_every_primary(void)
{
xce_scheduler_t* sched = start_sched();
xce_decision_t* decision;
int i, j;
e2e_push_queue_and_workers(sched, "cmp2_queue");
decision = e2e_dispatch_resreq(sched, JOB_ID + 5, "cmp2_queue",
"1*{rusage[mem=1]} + 2*{rusage[mem=1]}");
TEST_ASSERT_NOT_NULL(decision);
TEST_ASSERT_EQUAL_INT(XCE_DECISION_DISPATCH, decision->op);
TEST_ASSERT_EQUAL_FLOAT_MESSAGE(3.0f,
(float)e2e_alloc_total_slots(decision->u.alloc),
"1 + 2 tasks: the allocation must cover both primaries");
/* one entry per host */
for (i = 0; i < decision->u.alloc->nhosts; i++) {
for (j = i + 1; j < decision->u.alloc->nhosts; j++) {
TEST_ASSERT_TRUE_MESSAGE(
strcmp(decision->u.alloc->allocs[i]->host,
decision->u.alloc->allocs[j]->host) != 0,
"a host must appear once in a job allocation");
}
}
}
/* bug-1863: two primaries landing on the same host merge into that host's
* single entry (slots summed), not two entries or only the last one. */
static void test_compound_primaries_on_same_host_merge(void)
{
xce_scheduler_t* sched = start_sched();
xce_decision_t* decision;
e2e_push_queue_and_workers(sched, "cmp3_queue");
decision = e2e_dispatch_resreq(sched, JOB_ID + 6, "cmp3_queue",
"1*{rusage[mem=1]} + 1*{rusage[mem=1]}");
TEST_ASSERT_NOT_NULL(decision);
TEST_ASSERT_EQUAL_INT(XCE_DECISION_DISPATCH, decision->op);
TEST_ASSERT_EQUAL_FLOAT(2.0f,
(float)e2e_alloc_total_slots(decision->u.alloc));
TEST_ASSERT_EQUAL_INT_MESSAGE(1, decision->u.alloc->nhosts,
"both 1-task primaries fit the first 2-slot host -- one merged "
"entry");
}
int main(void)
{
UNITY_BEGIN();
@ -806,5 +856,7 @@ int main(void)
RUN_TEST(test_alter_resreq_job_dispatches_via_real_allocator);
RUN_TEST(test_compound_primary_needing_more_slots_than_exist_pends);
RUN_TEST(test_alter_falls_back_to_alternative_that_fits);
RUN_TEST(test_compound_allocation_includes_every_primary);
RUN_TEST(test_compound_primaries_on_same_host_merge);
return UNITY_END();
}