Fix CORE-5252 #29
Fix CORE-5252 #29
Conversation
| @@ -3579,6 +3585,15 @@ void jrd_tra::rollbackToSavepoint(thread_db* tdbb, SavNumber number) | |||
| * | |||
| **************************************/ | |||
| { | |||
| #ifdef SAVEPOINT_DEBUG | |||
| printf("Request to rollback savepoint %" QUADFORMAT "d. Stack: ", number); | |||
| for (Savepoint* itr = tra_save_point; itr; itr = itr->getNext()) | |||
There was a problem hiding this comment.
Please, don't use printf for debug purposes. Use common way and define appropriate (for ex SAV_trace) routine
There was a problem hiding this comment.
On 05/26/2016 05:22 PM, Dimitry Sibiryakov wrote:
@@ -3579,6 +3585,15 @@ void jrd_tra::rollbackToSavepoint(thread_db* tdbb, SavNumber number)
*
**************************************/
{
+#ifdef SAVEPOINT_DEBUG
- printf("Request to rollback savepoint %" QUADFORMAT "d. Stack: ", number);
- for (Savepoint* itr = tra_save_point; itr; itr = itr->getNext())
Ok.
And it should go to stderr, not stdout.
Only stderr (2) remains opened on posix when -d switch is used, stdout
(1) can become network socket or something else.
| } | ||
|
|
||
| // Release stored looper savepoints | ||
| Savepoint::destroy(request->req_savepoints); |
There was a problem hiding this comment.
This line causes engine to crash on test bugs.core_0214 as if savepoint was released from non-existing pool, but I cannot understand how it can be: pointers to transaction and pool are valid.
Changing it to Savepoint::mergeStacks(request->req_transaction->tra_save_free, request->req_savepoints); helps, but I'm afraid that it only put problem off, not fixing it.
There was a problem hiding this comment.
On 06/14/2016 01:13 PM, Dimitry Sibiryakov wrote:
return;
- }
- // Release stored looper savepoints
- Savepoint::destroy(request->req_savepoints);
This line causes engine to crash on test bugs.core_0214 as if savepoint was released from non-existing pool, but I cannot understand how it can be: pointers to transaction and pool are valid.
Always good idea to provide stack trace in such cases.
There was a problem hiding this comment.
On 06/14/2016 01:58 PM, Dimitry Sibiryakov wrote:
engine13.dll!Firebird::MemPool::release(void * object, bool flagDecr) Line 2125 C++
at this frame please print *block and *pool
There was a problem hiding this comment.
Now I see that block's pool doesn't match savepoint's transaction pool. But savepoint's transaction match request transaction.
There was a problem hiding this comment.
For this block 'pool' pointer is invalid, 'next' pointer is valid, but code handle it as a pointer to pool, that's a reason for crash. Perhaps, single savepoints are not supposed to be freed at all, they will die at once with transaction pool...
There was a problem hiding this comment.
On 06/14/2016 04:47 PM, Dimitry Sibiryakov wrote:
return;
- }
- // Release stored looper savepoints
- Savepoint::destroy(request->req_savepoints);
For this block 'pool' pointer is invalid, 'next' pointer is valid, but code handle it as a pointer to pool, that's a reason for crash.
That means that code tries to free some block once again.
Perhaps, single savepoints are not supposed to be freed at all, they will die at once with transaction pool...
You are receiving this because you commented.
Reply to this email directly or view it on GitHub:
https://github.com/FirebirdSQL/firebird/pull/29/files/057fc3a9ee0a851a39de07ad6497b28ac48ce931#r66972581
|
Finally I found the reason of crash. |
No description provided.