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

fix(job-array-single-wal-record): close bug-1443's bulk-insert partial-visibility gap

Browse source 
restore_job_array()'s bulk-insert pass could fail partway through
(key_add() on realloc growth exhaustion), leaving an already-inserted
prefix permanently visible in c->jobmgr but excluded from
out_new_jobs -- so those elements existed and were individually
addressable but never reached HOOK_DETAIL_JOB_NEW, never scheduled or
pushed, not on this attempt or any later successful retry. A literal
violation of Requirements 2.3 and 9.4, tracked as bug-1443.

Adds kvmgr_reserve()/xzarray_reserve(), new purely-additive primitives
in shared kvmgr.c/zarray.c (following the existing kvmgr_capture_keys()
template -- bare extern function, no RTTI vtable slot, caller-holds-
write-lock contract), that pre-grow c->jobmgr's backing store before
the bulk-insert pass begins, closing the dominant N-scaling failure
trigger structurally.

Found a genuine off-by-one in the existing, unmodified
xzarray_ensure_items(): its strict less-than sufficiency check means
reserving exactly array_size and then performing array_size real
stores still triggers one more growth cycle on the last store.
Reserving array_size + 1 is the exact minimal correct compensation,
independently re-derived and confirmed by review with concrete
arithmetic.

For the now-much-narrower residual case (kvtab_add()'s own small
per-entry allocation, or a content collision -- both non-N-scaling,
honestly documented as still theoretically possible): restore_job_array()
now populates out_new_jobs with whatever prefix was genuinely inserted
before a failure, and apply_committed_entry() fires HOOK_DETAIL_JOB_NEW
for that non-empty partial prefix even when the overall return is
negative, while still propagating the negative rc so the LSN correctly
halts and the Live Apply Resource Fence still engages for the
remainder. A genuinely-inserted element is never left un-pushed again.

Task 7.1 of job-array-single-wal-record (Requirements 2.3, 9.4, 9.6).
This commit is contained in:
Rongsong Shen 2026-07-16 15:08:02 +00:00
parent 575f928bbc
commit dcf9f39c22
13 changed files with 910 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

58
src/include/zarray.h
View file

@ -43,4 +43,62 @@ typedef struct {
extern typeinfo_t xzarray_type;
/* job-array-single-wal-record task 7.1 (bug-1443): pre-grows an xzarray_type
* instance's backing value store for `n` additional entries, by calling the
* existing, unmodified xzarray_ensure_items(a, a->count + n) internally --
* zero changes to xzarray_ensure_items or any other existing function in
* this file. This is the primitive kvmgr_reserve() (kvmgr.c/kvmgr.h) uses to
* pre-grow c->jobmgr's backing xzarray_type instance (kvm->data) before a
* bulk-insert pass, closing the DOMINANT, N-scaling failure mode for that
* pass (the large realloc() reachable through vstore ->
* xzarray_ensure_items). It does NOT make every subsequent vstore() call
* unconditionally infallible: this only covers the value-store side --
* xkvmgr_type's separate kvtab key->index table (kv.c) has its own,
* independent, much smaller, non-N-scaling per-entry allocation that this
* function does not and cannot pre-reserve.
*
* Caller MUST hold the same write lock the subsequent bulk vstore()/
* key_add() calls will be made under, for the full duration of this call --
* identical contract to kvmgr_capture_keys() (kvmgr.h).
*
* Returns 1 on success (capacity already sufficient, or successfully
* grown). Returns -EINVAL for a NULL array or negative n. Returns -ENOMEM if
* the underlying growth allocation fails. */
extern int xzarray_reserve(var array, int n);
#ifdef XCE_TESTING
/* Test-only fault injection for xzarray_reserve() (job-array-single-wal-
* record task 7.1 / bug-1443). When engaged, EVERY subsequent
* xzarray_reserve() call -- across any xzarray_type instance, including one
* wrapped inside an xkvmgr_type via kvmgr_reserve() -- reports a simulated
* allocation failure (-ENOMEM) without touching the real array. File-scope,
* not per-instance: mirrors job.c's own
* job_array_apply_test_set_bulk_insert_force_fail_at() convention (this
* feature's own established #ifdef XCE_TESTING test-only-symbol style).
*
* NOTE on why these declarations, unlike job.c's own #ifdef XCE_TESTING
* seams, are guarded here but their DEFINITIONS (zarray.c) are NOT: zarray.c
* is compiled exactly once, as part of the shared `libxce` static_library()
* target (src/lib/meson.build), without -DXCE_TESTING -- unlike job.c, which
* every consumer (including test binaries) recompiles from source with its
* own c_args. Guarding the .c definitions the same way would silently drop
* them from libxce.a in every build, including test builds that need them
* (confirmed empirically). So the underlying symbols are unconditionally
* present in the compiled library; only these DECLARATIONS are gated, which
* keeps them un-prototyped (hence uncallable without a compiler error) from
* any translation unit that does not itself define XCE_TESTING -- i.e. every
* production source file. */
extern void xzarray_reserve_test_set_force_fail(void);
extern void xzarray_reserve_test_clear_force_fail(void);
/* Test-only introspection: returns the CURRENT backing-store capacity (item
* slots, not bytes) of an xzarray_type instance. xzarray_t is fully opaque
* outside zarray.c, so this is the only way a test can observe whether a
* reservation actually grew capacity ahead of need, as opposed to deferring
* growth to the next vstore() call. Returns -EINVAL for a NULL array. See
* the NOTE above -- declaration-gated, not definition-gated, for the same
* shared-library build reason. */
extern int xzarray_test_get_caps(var array);
#endif
#endif