feat(scheduler-allocation): add resreq fingerprinting for the cache

Adds rusage_hash to xce_primary_resreq_t (computed once by
parse_resreq_primary() via a new hash_rusages() helper, hashing
rusage content rather than raw struct bytes since resname is a
non-interned strdup'd pointer), the xce_resreq_fingerprint_t value
type, and resreq_fp on xce_primary_schedobj_t, populated once at real
job-ingest time in core_build_primary_schedobj().

Task 1.2 (scheduler-allocation spec). Requirement 7.1.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XYdtrEWWZfKRRGED7raEYg
This commit is contained in:
Rongsong Shen 2026-09-04 15:00:12 +00:00
parent d5d2c1caf5
commit 5a6706761d
No known key found for this signature in database
GPG key ID: 5BCDAAE0123E9FE2
9 changed files with 550 additions and 4 deletions

View file

@ -304,6 +304,90 @@ cleanup_parser:
return rc;
}
/* Req 7.1 (scheduler-allocation spec): stable resource-usage fingerprint
* over a primary resreq's parsed `rusages` list.
*
* Style: FNV-1a 64-bit, consistent with this spec's cache-key hash
* (requirements.md "New types (Req 7)", cache_key_hash_ -- FNV-1a 32-bit
* over xce_cache_key_t, a later task's own scope). No existing hash-update
* helper for this purpose exists anywhere in this codebase; this is new,
* self-contained, file-local code.
*
* Design choice -- ORDER-DEPENDENT: each rusage_value_t is folded into the
* hash in xlist storage order (the order its `rusage[...]` clause was
* parsed in), rather than sorted/combined commutatively. This is simpler
* (no auxiliary sort, no need for a commutative combine step) and is
* sufficient for Req 7.1's actual requirement: parsing the *same* resreq
* string twice always produces the same xlist order, so the fingerprint
* is stable for identical input, which is all the unsatisfiable-resreq
* cache needs (a false miss on a reordered-but-equivalent resreq costs one
* redundant, still-correct allocation attempt; it can never cause a false
* hit).
*
* Field selection -- deliberately per-field, NOT a raw memcmp/byte-hash of
* `sizeof(rusage_value_t)`: `rusage_value_t.res` is a `char*` (owned,
* strdup'd per parse -- see kill_rusage_value()'s plain `FREEUP(val->res)`,
* not an xstring-interned/refcounted pointer). Hashing the raw struct
* bytes would hash the POINTER VALUE (heap address), which differs across
* separate parses of identical content and would make the "fingerprint"
* unstable by construction. Each field is hashed individually: the
* resource name's string CONTENT (not its pointer), then the raw bytes of
* `value`/`per_flag`/`duration`/`decay_flag` (safe to hash as raw bytes --
* none of those are pointers). */
static uint64_t hash_rusages(xlist_t* rusages)
{
uint64_t h = 14695981039346656037ULL; /* FNV-1a 64-bit offset basis */
int i;
if (rusages == NULL) {
return h;
}
for (i = 0; i < rusages->count; i++) {
rusage_value_t* v = (rusage_value_t*)xlist_get(rusages, i);
const unsigned char* p;
size_t n;
size_t j;
if (v == NULL) {
continue;
}
/* Resource name: hash the string content, not the pointer. */
p = (v->res != NULL) ? (const unsigned char*)v->res
: (const unsigned char*)"";
n = (v->res != NULL) ? strlen(v->res) : 0;
for (j = 0; j < n; j++) {
h = (h ^ (uint64_t)p[j]) * 1099511628211ULL;
}
/* Field separator -- keeps a name's trailing bytes from aliasing
* with the numeric fields that follow (e.g. "ab"+1 vs "a"+"b1"). */
h = (h ^ 0xFFu) * 1099511628211ULL;
p = (const unsigned char*)&v->value;
for (j = 0; j < sizeof(v->value); j++) {
h = (h ^ (uint64_t)p[j]) * 1099511628211ULL;
}
p = (const unsigned char*)&v->per_flag;
for (j = 0; j < sizeof(v->per_flag); j++) {
h = (h ^ (uint64_t)p[j]) * 1099511628211ULL;
}
p = (const unsigned char*)&v->duration;
for (j = 0; j < sizeof(v->duration); j++) {
h = (h ^ (uint64_t)p[j]) * 1099511628211ULL;
}
p = (const unsigned char*)&v->decay_flag;
for (j = 0; j < sizeof(v->decay_flag); j++) {
h = (h ^ (uint64_t)p[j]) * 1099511628211ULL;
}
}
return h;
}
int parse_resreq_primary(xce_primary_resreq_t* primary, char* resreqString,
var resmgr, char** errmsg)
{
@ -347,6 +431,10 @@ int parse_resreq_primary(xce_primary_resreq_t* primary, char* resreqString,
primary->minTasks = 1;
primary->maxTasks = 1;
/* Req 7.1: compute once, at parse time -- stable for the life of this
* resreq, reused by every later cache lookup instead of recomputing. */
primary->rusage_hash = hash_rusages(&(primary->rusages));
return rc;
}