Re: Parallel Bitmap scans a bit broken

On Wed, Mar 15, 2017 at 8:11 PM, Emre Hasegeli <emre(at)hasegeli(dot)com> wrote:
>> * thread #1: tid = 0x5045a8f, 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414, queue = 'com.apple.main-thread', stop reason = signal SIGUSR1
>> * frame #0: 0x000000010ae44558 postgres`brin_deform_tuple(brdesc=0x00007fea3c86a3a8, tuple=0x00007fea3c891040) + 40 at brin_tuple.c:414 [opt]
>> frame #1: 0x000000010ae4000c postgres`bringetbitmap(scan=0x00007fea3c875c20, tbm=<unavailable>) + 428 at brin.c:398 [opt]
>> frame #2: 0x000000010ae9b451 postgres`index_getbitmap(scan=0x00007fea3c875c20, bitmap=<unavailable>) + 65 at indexam.c:726 [opt]
>> frame #3: 0x000000010b0035a9 postgres`MultiExecBitmapIndexScan(node=<unavailable>) + 233 at nodeBitmapIndexscan.c:91 [opt]
>> frame #4: 0x000000010b002840 postgres`BitmapHeapNext(node=<unavailable>) + 400 at nodeBitmapHeapscan.c:143 [opt]

Further analyzing the call stack, seems like this is not exact call
stack where it crashed. Because, if you notice the code in the
brin_deform_tuple (line 414)

brin_deform_tuple(BrinDesc *brdesc, BrinTuple *tuple)

{
dtup = brin_new_memtuple(brdesc);

if (BrinTupleIsPlaceholder(tuple))
dtup->bt_placeholder = true;
dtup->bt_blkno = tuple->bt_blkno; --> line 414

This can crash at line:414, if either tuple is invalid memory(but I
think it's not because we have already accessed this memory in above
if check) or dtup is invalid (this is also not possible because
brin_new_memtuple has already accessed this).

--
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com