feat(scheduler-core): add picker configuration fields (task 1.3)

Adds xce_scheduler_t.job_pick_plugin (global picker name), and on
xce_sched_queue_t: project_name + job_pick_plugin (per-queue override,
NULL = inherit global). Adds xce_sched_project_t.priority (scaffolding
only, no quota logic). sched_set_default_job_pick_plugin() is added but
not yet wired into sched_init() -- task 2.4 owns that, alongside
replacing the hardcoded "xce_sched_fcfs" literal.

Requirements: 2.1, 2.2, 2.3, 2.4
This commit is contained in:
Rongsong Shen 2026-08-03 05:45:53 +00:00
parent 4d8d72142e
commit 0d63a789a1
6 changed files with 398 additions and 2 deletions

View file

@ -15,7 +15,7 @@
- _Requirements: 1.6, 2.5_
- _Depends: 1.1_
- [ ] 1.3 (P) Add picker configuration fields
- [x] 1.3 (P) Add picker configuration fields
- Add a global picker plugin name field to the scheduler, defaulting to `builtin_fcfs` when unset
- Add a per-queue picker plugin override field, defaulting to NULL (inherit global)
- Add a project-name field to the queue struct, and extend the project struct with a priority field (scaffolding only — no quota fields)
@ -349,3 +349,10 @@
- Observable: the soak run completes with a clean pass across all rounds, matching the zero-regression bar set by the Phase 1 and Phase 2 soak runs
- _Requirements: 4.3_
- _Depends: 12.4, 10.6_
## Implementation Notes
- Task 1.3 added `sched_set_default_job_pick_plugin()` (schedobj.c) but deliberately did NOT call it from `sched_init()` — task 2.4 owns replacing `scheduler.c:43`'s hardcoded `"xce_sched_fcfs"` literal and must also wire this defaulting call (or equivalent logic) into scheduler startup, or `sched->job_pick_plugin` stays NULL through a real boot.
- Background IDE diagnostics ("file not found: rtti.h", "unknown type name" for various xce_* types) have appeared consistently on every scheduler.h/plugin.h/types.h/schedobj.c change so far (tasks 1.1-1.3) and are confirmed stale clangd artifacts each time via independent real `meson compile`/`meson test` runs — not real build failures. Don't let these block a task; verify with the real build instead.
- Meson build dir is `src/xce.build`, not repo-root `xce.build` — build/test commands must run from `src/` (e.g. `cd src && nix develop . --command bash -c "meson test -C xce.build --suite scheduler"`).
- Scheduler testsuite convention: run test binaries directly with `ASAN_OPTIONS=detect_leaks=0` (fixtures aren't torn down, by design — matches existing `meson.build` test registration).

View file

@ -64,7 +64,12 @@ typedef struct {
} xce_sched_user_t;
typedef struct {
char* name;
char* name;
/* Req 2 AC (scheduler-core spec, task 1.3): inter-project ordering
* scaffolding for a future spec (project-resource-ownership). No
* quota/usage/hierarchy fields yet -- see design.md's "Out of Scope".
* Defaults to 0 (make_sched_project, schedobj.c). */
int32_t priority;
} xce_sched_project_t;
typedef struct {
@ -75,6 +80,17 @@ typedef struct {
int32_t accept_interval;
int32_t sched_delay;
char* resreq;
/* Req 2 AC 3 (scheduler-core spec, task 1.3): owning project name,
* establishing the queue->project relationship. Scaffolding only --
* xce_queue_t (the raw config struct) has no project field yet, so
* make_sched_queue (schedobj.c) leaves this NULL; the picker plugin
* may use it internally for project-scoped iteration (v2). */
char* project_name;
/* Req 2 AC 2: per-queue picker-plugin override. NULL means "inherit
* sched->job_pick_plugin" (the global default). Non-NULL means jobs in
* this queue use this picker instead -- how a picker plugin *uses* the
* override is the plugin's concern, not the core loop's (task 2.4+). */
char* job_pick_plugin;
} xce_sched_queue_t;
typedef struct xce_sched_job_ {

View file

@ -91,6 +91,14 @@ typedef struct {
uint64_t nextid;
xce_conf4sched_t* config;
uint64_t configver;
/* Req 2 AC 1 (scheduler-core spec, task 1.3): global picker plugin
* name. NULL until sched_set_default_job_pick_plugin() (schedobj.c)
* applies the "builtin_fcfs" default, or a future config-driven
* assignment sets it explicitly. This field only carries the *name* --
* resolving it against the plugin registry (plugin_register/key_find)
* and failing startup loudly on an unregistered name (Req 2 AC 5) is
* task 2.4's job. */
char* job_pick_plugin;
var plugins;
void* seqjob_allocator;
void* parjob_allocator;
@ -125,6 +133,10 @@ extern int sched_init(xce_scheduler_t*, xce_client_config_t*,
const xce_cluster_config_t*, xce_request_handler_t);
extern void sched_fini(xce_scheduler_t*);
extern int sched_init_config(xce_scheduler_t* sched, xce_conf4sched_t*);
/* Req 2 AC 1 (scheduler-core spec, task 1.3): applies the "builtin_fcfs"
* default to sched->job_pick_plugin when unset. See the field comment on
* xce_scheduler_t above -- this does NOT resolve/register the name. */
extern void sched_set_default_job_pick_plugin(xce_scheduler_t*);
extern int sched_update(xce_scheduler_t*);
extern void sched_go(xce_scheduler_t*);
extern void sched_begin_session(xce_scheduler_t*);

View file

@ -237,6 +237,14 @@ xce_sched_queue_t* make_sched_queue(void* arg, xce_queue_t* q)
queue->priority = q->priority;
queue->accept_interval = q->accept_interval;
queue->sched_delay = q->sched_delay;
/* Req 2 AC 2/3 (scheduler-core spec, task 1.3): xce_queue_t (`q`, the
* raw config struct) does not carry project_name/job_pick_plugin yet
* -- calloc() above already leaves both NULL (queue inherits the
* global picker; project ownership unset). Explicit here for
* documentation/testability; wiring these from real config is a later
* task. */
queue->project_name = NULL;
queue->job_pick_plugin = NULL;
return queue;
}
@ -363,11 +371,44 @@ xce_sched_project_t* make_sched_project(void* arg, char* name)
project->name = makeString(name);
}
/* Req 2 AC (scheduler-core spec, task 1.3): scaffolding only -- no
* quota/hierarchy inputs exist yet, so priority defaults to 0
* (calloc() above already zeroes it; explicit here for
* documentation/testability). */
project->priority = 0;
sched_add_project(sched, project);
return project;
}
/*
* sched_set_default_job_pick_plugin - Req 2 AC 1 (scheduler-core spec,
* task 1.3). Applies the "builtin_fcfs" default to sched->job_pick_plugin
* when it is unset (NULL). Idempotent: a name already present (whether
* previously defaulted or set by a future config-driven path) is left
* untouched, so this is safe to call defensively.
*
* This function only assigns the *name* -- it never touches the plugin
* registry (sched->plugins) and never calls plugin_register()/key_find().
* Resolving the configured name against a registered picker plugin, and
* failing scheduler startup loudly when the name is not registered
* (Req 2 AC 5), is task 2.4's job. Kept as a separate, standalone function
* (rather than inlined into sched_init()) so the default-value behavior is
* unit-testable without the network/plugin-registry dependencies
* sched_init() pulls in.
*/
void sched_set_default_job_pick_plugin(xce_scheduler_t* sched)
{
if (sched == NULL) {
return;
}
if (sched->job_pick_plugin == NULL) {
sched->job_pick_plugin = makeString("builtin_fcfs");
}
}
static void host_uncommit_killer(void* obj)
{
xvec_t* v = (xvec_t*)obj;

View file

@ -81,6 +81,18 @@ sched_picker_registration_test = executable('sched_picker_registration_test',
dependencies: [scheduler_test_deps],
install: false)
# White-box test for the picker configuration fields added to
# xce_scheduler_t/xce_sched_queue_t/xce_sched_project_t (scheduler-core
# spec, task 1.3). It #includes ../src/schedobj.c directly (same white-box
# pattern as sched_job_update_test) so it exercises the real
# make_sched_queue/make_sched_project/sched_set_default_job_pick_plugin,
# not a re-implementation of any of them.
sched_picker_config_fields_test = executable('sched_picker_config_fields_test',
['test_picker_config_fields.c'],
include_directories:[incdir],
dependencies: [scheduler_test_deps],
install: false)
# Register as meson tests so `meson test --suite scheduler` runs them all.
# detect_leaks=0 matches the suite convention (fixtures are not torn down).
test('scheduler test', scheduler_test, suite: 'scheduler',
@ -97,3 +109,5 @@ test('scheduler picker iface test', sched_picker_iface_test, suite: 'scheduler',
env: ['ASAN_OPTIONS=detect_leaks=0'])
test('scheduler picker registration test', sched_picker_registration_test, suite: 'scheduler',
env: ['ASAN_OPTIONS=detect_leaks=0'])
test('scheduler picker config fields test', sched_picker_config_fields_test, suite: 'scheduler',
env: ['ASAN_OPTIONS=detect_leaks=0'])

View file

@ -0,0 +1,306 @@
/*
* test_picker_config_fields.c - white-box tests for the picker
* configuration fields added to xce_scheduler_t, xce_sched_queue_t, and
* xce_sched_project_t (scheduler-core spec, task 1.3, Req 2 AC 1-4).
*
* RED: before task 1.3, xce_scheduler_t had no job_pick_plugin field,
* xce_sched_queue_t had no project_name/job_pick_plugin fields,
* xce_sched_project_t had no priority field, and
* sched_set_default_job_pick_plugin() did not exist -- this file failed to
* compile.
* GREEN: the fields exist with the documented types/defaults, and
* make_sched_queue/make_sched_project (schedobj.c) leave the new fields at
* their documented default values.
*
* This task is scaffolding only -- no picker-name lookup/resolution
* (plugin_register/key_find against the plugin registry) exists yet, that
* is task 2.4's job. The tasks.md "Observable" for this task is:
* "a queue config with no override inherits the global picker name;
* a queue config with an override reports that override, not the
* global default"
* Read literally that would require the lookup logic this task explicitly
* excludes, so it is exercised here at the structural level instead: for
* a NULL-override queue, the field values a resolver would read are
* exactly sched->job_pick_plugin; for an overridden queue, the queue's own
* field is a different, non-NULL string, distinct from the scheduler's
* global default -- proving the two fields are independently readable and
* correctly defaulted, without this test (or production code) performing
* the resolution itself.
*
* Strategy: pull the REAL make_sched_queue/make_sched_project/
* sched_set_default_job_pick_plugin into this TU by including schedobj.c
* (same white-box convention as test_job_update.c /
* test_job_array_expand.c / test_plugin_picker_registration.c). The
* scheduler-internal externs schedobj.c references but which are defined
* in other .c files not linked here are stubbed below -- identical stub
* set to test_job_update.c, since this task added no new external calls
* to schedobj.c (only field assignments plus one new self-contained
* function).
*/
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unity/unity.h>
#include "kvmgr.h"
#include "scheduler.h"
#include "xstring.h"
/* Pull the real implementation under test into this TU. */
#include "../src/schedobj.c"
/* Stubs for scheduler-internal externs schedobj.c references but which are
* defined in other .c files not linked into this test (same set as
* test_job_update.c). Not exercised by the tests below -- present only to
* satisfy the linker for this whole-file #include. */
xce_sched_user_t* sched_get_user(xce_scheduler_t* s, char* n)
{
(void)s; (void)n; return NULL;
}
xce_sched_project_t* sched_get_project(xce_scheduler_t* s, char* n)
{
(void)s; (void)n; return NULL;
}
xce_sched_queue_t* sched_get_queue(xce_scheduler_t* s, char* n)
{
(void)s; (void)n; return NULL;
}
xce_sched_job_t* sched_get_job(xce_scheduler_t* s, uint64_t jobid)
{
(void)s; (void)jobid; return NULL;
}
int plugin_job_update(void* s, xce_sched_job_t* j) { (void)s; (void)j; return 0; }
int plugin_queue_update(void* s, xce_sched_queue_t* q) { (void)s; (void)q; return 0; }
int plugin_job_rusage_update(void* s, xce_sched_job_t* j, xce_job_rusages_t* r)
{
(void)s; (void)j; (void)r; return 0;
}
xce_job_account_t* make_job_account(xce_job_alloc_t* a) { (void)a; return NULL; }
void free_job_account(xce_job_account_t* a) { (void)a; }
int sched_add_user(xce_scheduler_t* s, xce_sched_user_t* u) { (void)s; (void)u; return 0; }
int sched_add_project(xce_scheduler_t* s, xce_sched_project_t* p) { (void)s; (void)p; return 0; }
void sched_worker_release_host_ctx(xce_sched_worker_t* w) { (void)w; }
int plugin_prepare_allocation(void* s, xce_sched_ctx_t* c) { (void)s; (void)c; return 0; }
void setUp(void) { return; }
void tearDown(void) { return; }
/* Per-test fixture: a zeroed scheduler, with the xstring intern table
* (needed by makeString(), which every constructor under test calls) and
* the CSE intern table live. Mirrors test_job_update.c's fixture_init --
* called once per test so each test is self-contained. */
static void fixture_init(xce_scheduler_t* sched)
{
memset(sched, 0, sizeof(*sched));
resreq_init(1);
xstring_table_init(1024 * 16);
}
static xce_queue_t make_test_raw_queue(const char* name)
{
xce_queue_t q;
memset(&q, 0, sizeof(q));
q.queue = (char*)name;
return q;
}
/* ---------------------------------------------------------------------- */
/* Req 2 AC 1: xce_scheduler_t.job_pick_plugin */
/* ---------------------------------------------------------------------- */
/* Field exists with the right type (char*) and is NULL ("unset") on a
* freshly zeroed scheduler, before any default is applied. */
static void test_scheduler_job_pick_plugin_field_is_char_star_and_unset_by_default(void)
{
xce_scheduler_t sched;
fixture_init(&sched);
TEST_ASSERT_NULL(sched.job_pick_plugin);
/* Type check: assigning a char* literal must compile without a cast --
* this is a compile-time proof of the declared field type. */
sched.job_pick_plugin = "some_name";
TEST_ASSERT_EQUAL_STRING("some_name", sched.job_pick_plugin);
}
/* Req 2 AC 1's literal wording: "defaulting to builtin_fcfs when unset". */
static void test_sched_set_default_job_pick_plugin_applies_builtin_fcfs_when_unset(void)
{
xce_scheduler_t sched;
fixture_init(&sched);
TEST_ASSERT_NULL(sched.job_pick_plugin);
sched_set_default_job_pick_plugin(&sched);
TEST_ASSERT_NOT_NULL(sched.job_pick_plugin);
TEST_ASSERT_EQUAL_STRING("builtin_fcfs", sched.job_pick_plugin);
}
/* "...when unset" implies idempotence: an already-configured name (however
* it got there) must not be clobbered by the default. */
static void
test_sched_set_default_job_pick_plugin_does_not_clobber_configured_value(void)
{
xce_scheduler_t sched;
fixture_init(&sched);
sched.job_pick_plugin = makeString("priority_picker");
sched_set_default_job_pick_plugin(&sched);
TEST_ASSERT_EQUAL_STRING("priority_picker", sched.job_pick_plugin);
}
static void test_sched_set_default_job_pick_plugin_is_null_safe(void)
{
sched_set_default_job_pick_plugin(NULL); /* must not crash */
TEST_PASS();
}
/* ---------------------------------------------------------------------- */
/* Req 2 AC 2/3: xce_sched_queue_t.job_pick_plugin / .project_name */
/* ---------------------------------------------------------------------- */
/* A queue built from real config (make_sched_queue) with no picker/project
* info in the raw config struct leaves both new fields NULL -- the
* documented "unset"/"inherit" sentinel, not garbage or an empty string. */
static void
test_make_sched_queue_defaults_project_name_and_override_to_null(void)
{
xce_scheduler_t sched;
xce_queue_t q = make_test_raw_queue("normal");
xce_sched_queue_t* queue;
fixture_init(&sched);
queue = make_sched_queue(&sched, &q);
TEST_ASSERT_NOT_NULL(queue);
TEST_ASSERT_EQUAL_STRING("normal", queue->name);
TEST_ASSERT_NULL(queue->job_pick_plugin);
TEST_ASSERT_NULL(queue->project_name);
}
/* tasks.md Observable, half 1: "a queue config with no override inherits
* the global picker name." Structural proof: the queue's own override
* field is NULL, and the value any resolver would fall back to
* (sched->job_pick_plugin) is the defaulted global name. */
static void test_queue_with_no_override_inherits_global_picker_name(void)
{
xce_scheduler_t sched;
xce_queue_t q = make_test_raw_queue("batch");
xce_sched_queue_t* queue;
fixture_init(&sched);
sched_set_default_job_pick_plugin(&sched); /* global = "builtin_fcfs" */
queue = make_sched_queue(&sched, &q);
TEST_ASSERT_NOT_NULL(queue);
TEST_ASSERT_NULL(queue->job_pick_plugin);
TEST_ASSERT_EQUAL_STRING("builtin_fcfs", sched.job_pick_plugin);
}
/* tasks.md Observable, half 2: "a queue config with an override reports
* that override, not the global default." Config-driven wiring of this
* field lands in a later task (xce_queue_t has no job_pick_plugin source
* yet), so the override is applied via direct struct assignment here --
* exactly the "structurally representable" test style this task calls
* for -- and then read back. */
static void
test_queue_with_override_reports_its_own_override_not_global_default(void)
{
xce_scheduler_t sched;
xce_queue_t q = make_test_raw_queue("gpu");
xce_sched_queue_t* queue;
fixture_init(&sched);
sched_set_default_job_pick_plugin(&sched); /* global = "builtin_fcfs" */
queue = make_sched_queue(&sched, &q);
TEST_ASSERT_NOT_NULL(queue);
queue->job_pick_plugin = makeString("priority_picker");
TEST_ASSERT_NOT_NULL(queue->job_pick_plugin);
TEST_ASSERT_EQUAL_STRING("priority_picker", queue->job_pick_plugin);
/* The override is reported, NOT the global default. */
TEST_ASSERT_TRUE(strcmp(queue->job_pick_plugin, sched.job_pick_plugin) != 0);
TEST_ASSERT_EQUAL_STRING("builtin_fcfs", sched.job_pick_plugin);
}
/* project_name is independently settable/readable (structural field check
* -- config-source wiring is a later task, same rationale as above). */
static void test_queue_project_name_is_independently_settable(void)
{
xce_scheduler_t sched;
xce_queue_t q = make_test_raw_queue("gpu");
xce_sched_queue_t* queue;
fixture_init(&sched);
queue = make_sched_queue(&sched, &q);
TEST_ASSERT_NOT_NULL(queue);
queue->project_name = makeString("acct-42");
TEST_ASSERT_EQUAL_STRING("acct-42", queue->project_name);
/* Setting project_name must not disturb the independent picker-override
* field (still inheriting -- NULL). */
TEST_ASSERT_NULL(queue->job_pick_plugin);
}
/* ---------------------------------------------------------------------- */
/* Req 2 AC: xce_sched_project_t.priority (scaffolding) */
/* ---------------------------------------------------------------------- */
static void test_make_sched_project_priority_field_defaults_to_zero(void)
{
xce_scheduler_t sched;
xce_sched_project_t* project;
fixture_init(&sched);
project = make_sched_project(&sched, "acct-42");
TEST_ASSERT_NOT_NULL(project);
TEST_ASSERT_EQUAL_STRING("acct-42", project->name);
/* Scaffolding only -- no quota/hierarchy logic exists yet
* (project-resource-ownership spec). */
TEST_ASSERT_EQUAL_INT32(0, project->priority);
}
/* Type check: field accepts a plain int32_t assignment/comparison --
* compile-time proof of the declared type. */
static void test_project_priority_field_is_int32_and_independently_settable(void)
{
xce_scheduler_t sched;
xce_sched_project_t* project;
fixture_init(&sched);
project = make_sched_project(&sched, "acct-7");
project->priority = 7;
TEST_ASSERT_EQUAL_INT32(7, project->priority);
/* Setting priority must not disturb the project's name. */
TEST_ASSERT_EQUAL_STRING("acct-7", project->name);
}
int main(void)
{
UNITY_BEGIN();
RUN_TEST(test_scheduler_job_pick_plugin_field_is_char_star_and_unset_by_default);
RUN_TEST(test_sched_set_default_job_pick_plugin_applies_builtin_fcfs_when_unset);
RUN_TEST(test_sched_set_default_job_pick_plugin_does_not_clobber_configured_value);
RUN_TEST(test_sched_set_default_job_pick_plugin_is_null_safe);
RUN_TEST(test_make_sched_queue_defaults_project_name_and_override_to_null);
RUN_TEST(test_queue_with_no_override_inherits_global_picker_name);
RUN_TEST(test_queue_with_override_reports_its_own_override_not_global_default);
RUN_TEST(test_queue_project_name_is_independently_settable);
RUN_TEST(test_make_sched_project_priority_field_defaults_to_zero);
RUN_TEST(test_project_priority_field_is_int32_and_independently_settable);
return UNITY_END();
}