Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

gh-133136: Limit excess memory held by QSBR#135107

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.

Already on GitHub?Sign in to your account

Closed
Closed
Show file tree
Hide file tree
Changes from1 commit
Commits
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
NextNext commit
gh-133136: Limit excess memory held by QSBR
The free threading build uses QSBR to delay the freeing of dictionarykeys and list arrays when the objects are accessed by multiple threadsin order to allow concurrent reads to proceeed with holding the objectlock. The requests are processed in batches to reduce executionoverhead, but for large memory blocks this can lead to excess memoryusage.Take into account the size of the memory block when deciding when toprocess QSBR requests.
  • Loading branch information
@colesbury
colesbury committedJun 3, 2025
commit7ef2e30c9d0d45472c3942ff9733f1d7dfc184ec
2 changes: 1 addition & 1 deletionInclude/internal/pycore_pymem.h
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -88,7 +88,7 @@ extern wchar_t *_PyMem_DefaultRawWcsdup(const wchar_t *str);
extern int _PyMem_DebugEnabled(void);

// Enqueue a pointer to be freed possibly after some delay.
extern void _PyMem_FreeDelayed(void *ptr);
extern void _PyMem_FreeDelayed(void *ptr, size_t size);

// Enqueue an object to be freed possibly after some delay
#ifdef Py_GIL_DISABLED
Expand Down
4 changes: 4 additions & 0 deletionsInclude/internal/pycore_qsbr.h
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -51,6 +51,10 @@ struct _qsbr_thread_state {
// Used to defer advancing write sequence a fixed number of times
int deferrals;

// Estimate for the amount of memory that is held by this thread since
// the last non-deferred advance.
size_t memory_deferred;

// Is this thread state allocated?
bool allocated;
struct _qsbr_thread_state *freelist_next;
Expand Down
View file
Open in desktop
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,2 @@
Limit excess memory usage in the :term:`free threading` build when a
large dictionary or list is resized and accessed by multiple threads.
2 changes: 1 addition & 1 deletionObjects/codeobject.c
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -3350,7 +3350,7 @@ create_tlbc_lock_held(PyCodeObject *co, Py_ssize_t idx)
}
memcpy(new_tlbc->entries, tlbc->entries, tlbc->size * sizeof(void *));
_Py_atomic_store_ptr_release(&co->co_tlbc, new_tlbc);
_PyMem_FreeDelayed(tlbc);
_PyMem_FreeDelayed(tlbc, tlbc->size * sizeof(void *));
tlbc = new_tlbc;
}
char *bc = PyMem_Calloc(1, _PyCode_NBYTES(co));
Expand Down
4 changes: 2 additions & 2 deletionsObjects/dictobject.c
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -813,7 +813,7 @@ free_keys_object(PyDictKeysObject *keys, bool use_qsbr)
{
#ifdef Py_GIL_DISABLED
if (use_qsbr) {
_PyMem_FreeDelayed(keys);
_PyMem_FreeDelayed(keys, _PyDict_KeysSize(keys));
return;
}
#endif
Expand DownExpand Up@@ -858,7 +858,7 @@ free_values(PyDictValues *values, bool use_qsbr)
assert(values->embedded == 0);
#ifdef Py_GIL_DISABLED
if (use_qsbr) {
_PyMem_FreeDelayed(values);
_PyMem_FreeDelayed(values, values_size_from_count(values->capacity));
return;
}
#endif
Expand Down
3 changes: 2 additions & 1 deletionObjects/listobject.c
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -61,7 +61,8 @@ free_list_items(PyObject** items, bool use_qsbr)
#ifdef Py_GIL_DISABLED
_PyListArray *array = _Py_CONTAINER_OF(items, _PyListArray, ob_item);
if (use_qsbr) {
_PyMem_FreeDelayed(array);
size_t size = sizeof(_PyListArray) + array->allocated * sizeof(PyObject *);
_PyMem_FreeDelayed(array, size);
}
else {
PyMem_Free(array);
Expand Down
39 changes: 32 additions & 7 deletionsObjects/obmalloc.c
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -1141,8 +1141,27 @@
}
}

static int
should_advance_qsbr(_PyThreadStateImpl *tstate, size_t size)

Check warning on line 1145 in Objects/obmalloc.c

View workflow job for this annotation

GitHub Actions/ Ubuntu / build and test (ubuntu-24.04)

‘should_advance_qsbr’ defined but not used [-Wunused-function]

Check warning on line 1145 in Objects/obmalloc.c

View workflow job for this annotation

GitHub Actions/ Ubuntu / build and test (ubuntu-24.04-arm)

‘should_advance_qsbr’ defined but not used [-Wunused-function]

Check warning on line 1145 in Objects/obmalloc.c

View workflow job for this annotation

GitHub Actions/ Cross build Linux

‘should_advance_qsbr’ defined but not used [-Wunused-function]

Check warning on line 1145 in Objects/obmalloc.c

View workflow job for this annotation

GitHub Actions/ Hypothesis tests on Ubuntu

‘should_advance_qsbr’ defined but not used [-Wunused-function]

Check warning on line 1145 in Objects/obmalloc.c

View workflow job for this annotation

GitHub Actions/ Address sanitizer (ubuntu-24.04)

‘should_advance_qsbr’ defined but not used [-Wunused-function]

Check warning on line 1145 in Objects/obmalloc.c

View workflow job for this annotation

GitHub Actions/ Ubuntu (bolt) / build and test (ubuntu-24.04)

‘should_advance_qsbr’ defined but not used [-Wunused-function]
{
// If the deferred memory exceeds 1 MiB, we force an advance in the
// shared QSBR sequence number to limit excess memory usage.
static const size_t QSBR_DEFERRED_LIMIT = 1024 * 1024;
if (size > QSBR_DEFERRED_LIMIT) {
tstate->qsbr->memory_deferred = 0;
return 1;
}

tstate->qsbr->memory_deferred += size;
if (tstate->qsbr->memory_deferred > QSBR_DEFERRED_LIMIT) {
tstate->qsbr->memory_deferred = 0;
return 1;
}
return 0;
}
Comment on lines +1146 to +1162
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

Do we need early return here?
It looks like it will be same eventually.

Suggested change
should_advance_qsbr(_PyThreadStateImpl*tstate,size_tsize)
{
// If the deferred memory exceeds 1 MiB, we force an advance in the
// shared QSBR sequence number to limit excess memory usage.
staticconstsize_tQSBR_DEFERRED_LIMIT=1024*1024;
if (size>QSBR_DEFERRED_LIMIT) {
tstate->qsbr->memory_deferred=0;
return1;
}
tstate->qsbr->memory_deferred+=size;
if (tstate->qsbr->memory_deferred>QSBR_DEFERRED_LIMIT) {
tstate->qsbr->memory_deferred=0;
return1;
}
return0;
}
should_advance_qsbr(_PyThreadStateImpl*tstate,size_tsize)
{
// If the deferred memory exceeds 1 MiB, we force an advance in the
// shared QSBR sequence number to limit excess memory usage.
staticconstsize_tQSBR_DEFERRED_LIMIT=1024*1024;
tstate->qsbr->memory_deferred+=size;
if (tstate->qsbr->memory_deferred>QSBR_DEFERRED_LIMIT) {
tstate->qsbr->memory_deferred=0;
return1;
}
return0;
}


static void
free_delayed(uintptr_t ptr)
free_delayed(uintptr_t ptr, size_t size)
{
#ifndef Py_GIL_DISABLED
free_work_item(ptr, NULL, NULL);
Expand DownExpand Up@@ -1200,23 +1219,29 @@
}

assert(buf != NULL && buf->wr_idx < WORK_ITEMS_PER_CHUNK);
uint64_t seq = _Py_qsbr_deferred_advance(tstate->qsbr);
uint64_t seq;
int force_advance = should_advance_qsbr(tstate, size);
if (force_advance) {
seq = _Py_qsbr_advance(tstate->qsbr->shared);
}
else {
seq = _Py_qsbr_deferred_advance(tstate->qsbr);
}
buf->array[buf->wr_idx].ptr = ptr;
buf->array[buf->wr_idx].qsbr_goal = seq;
buf->wr_idx++;

if (buf->wr_idx == WORK_ITEMS_PER_CHUNK) {
if (buf->wr_idx == WORK_ITEMS_PER_CHUNK || force_advance) {
_PyMem_ProcessDelayed((PyThreadState *)tstate);
}
#endif
}

void
_PyMem_FreeDelayed(void *ptr)
_PyMem_FreeDelayed(void *ptr, size_t size)
{
assert(!((uintptr_t)ptr & 0x01));
if (ptr != NULL) {
free_delayed((uintptr_t)ptr);
free_delayed((uintptr_t)ptr, size);
}
}

Expand All@@ -1226,7 +1251,7 @@
{
assert(!((uintptr_t)ptr & 0x01));
if (ptr != NULL) {
free_delayed(((uintptr_t)ptr)|0x01);
free_delayed(((uintptr_t)ptr)|0x01, 64);
}
}
#endif
Expand Down
Loading
[8]ページ先頭
©2009-2025 Movatter.jp