Commit 83a55670 authored by Marko Mäkelä's avatar Marko Mäkelä

MDEV-24188 fixup: Simplify the wait loop

Starting with commit 7cffb5f6 (MDEV-23399)
the function buf_flush_page() will first acquire block->lock and only
after that invoke set_io_fix(). Before that, it was possible to reach
a livelock between buf_page_create() and buf_flush_page().

buf_page_create(): Directly try acquiring the exclusive page latch
without checking whether the page is io-fixed or buffer-fixed.
(As a matter of fact, the have_x_latch() check is not strictly necessary,
because we still support recursive X-latches.)
In case of a latch conflict, wait while allowing buf_page_write_complete()
to acquire buf_pool.mutex and release the block->lock.

An attempt to wait for exclusive block->lock while holding buf_pool.mutex
would lead to a hang in the tests parts.part_supported_sql_func_innodb
and stress.ddl_innodb, due to a deadlock between buf_page_write_complete()
and buf_page_create().

Similarly, in case of an I/O fixed compressed-only
ROW_FORMAT=COMPRESSED page, we will sleep before retrying.

In both cases, we will sleep for 1ms or until a flush batch is completed.
parent 694926a4
...@@ -3755,9 +3755,9 @@ buf_page_create(fil_space_t *space, uint32_t offset, ...@@ -3755,9 +3755,9 @@ buf_page_create(fil_space_t *space, uint32_t offset,
free_block->initialise(page_id, zip_size, 1); free_block->initialise(page_id, zip_size, 1);
const ulint fold= page_id.fold(); const ulint fold= page_id.fold();
loop:
mysql_mutex_lock(&buf_pool.mutex); mysql_mutex_lock(&buf_pool.mutex);
loop:
buf_block_t *block= reinterpret_cast<buf_block_t*> buf_block_t *block= reinterpret_cast<buf_block_t*>
(buf_pool.page_hash_get_low(page_id, fold)); (buf_pool.page_hash_get_low(page_id, fold));
...@@ -3775,17 +3775,15 @@ buf_page_create(fil_space_t *space, uint32_t offset, ...@@ -3775,17 +3775,15 @@ buf_page_create(fil_space_t *space, uint32_t offset,
if (!mtr->have_x_latch(*block)) if (!mtr->have_x_latch(*block))
{ {
buf_block_buf_fix_inc(block, __FILE__, __LINE__); buf_block_buf_fix_inc(block, __FILE__, __LINE__);
while (!rw_lock_x_lock_nowait(&block->lock))
{ {
while (block->page.io_fix() != BUF_IO_NONE || /* Wait for buf_page_write_complete() to release block->lock.
block->page.buf_fix_count() != 1) We must not hold buf_pool.mutex while waiting. */
{
timespec abstime; timespec abstime;
set_timespec_nsec(abstime, 1000000); set_timespec_nsec(abstime, 1000000);
mysql_cond_timedwait(&buf_pool.done_flush_list, &buf_pool.mutex, mysql_cond_timedwait(&buf_pool.done_flush_list, &buf_pool.mutex,
&abstime); &abstime);
} }
}
rw_lock_x_lock(&block->lock);
mtr_memo_push(mtr, block, MTR_MEMO_PAGE_X_FIX); mtr_memo_push(mtr, block, MTR_MEMO_PAGE_X_FIX);
} }
else else
...@@ -3805,7 +3803,11 @@ buf_page_create(fil_space_t *space, uint32_t offset, ...@@ -3805,7 +3803,11 @@ buf_page_create(fil_space_t *space, uint32_t offset,
if (block->page.io_fix() != BUF_IO_NONE) if (block->page.io_fix() != BUF_IO_NONE)
{ {
hash_lock->write_unlock(); hash_lock->write_unlock();
mysql_mutex_unlock(&buf_pool.mutex); /* Wait for buf_page_write_complete() to release the I/O fix. */
timespec abstime;
set_timespec_nsec(abstime, 1000000);
mysql_cond_timedwait(&buf_pool.done_flush_list, &buf_pool.mutex,
&abstime);
goto loop; goto loop;
} }
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment