Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
23.02.4, 23.10.0
-
None
-
2023-12
Description
There is a crash that is caused by FairThreadScheduler trying to send an error over an empty socket. This might happen if the Job that causes an exception in FairThreadScheduler has been received via same node communication messaging queue.
crash trace
Date/time: 2024-01-04 14:11:47 |
Signal: 6 |
|
|
/usr/bin/PrimProc(+0xb74b6)[0x5613d1fb04b6] |
/lib64/libpthread.so.0(+0x12cf0)[0x7fc87b116cf0] |
/lib64/libc.so.6(gsignal+0x10f)[0x7fc879b31aff] |
/lib64/libc.so.6(abort+0x127)[0x7fc879b04ea5] |
/lib64/libc.so.6(+0x21d79)[0x7fc879b04d79] |
/lib64/libc.so.6(+0x47456)[0x7fc879b2a456] |
/lib64/libthreadpool.so(_ZN10threadpool14FairThreadPool12sendErrorMsgEjjN5boost10shared_ptrIN11messageqcpp8IOSocketEEE+0x1ab)[0x7fc87aeb6b7b] |
/lib64/libthreadpool.so(+0x18a6e)[0x7fc87ae9da6e] |
/usr/bin/PrimProc(+0xb8f97)[0x5613d1fb1f97] |
/lib64/libpthread.so.0(+0x81ca)[0x7fc87b10c1ca] |
/lib64/libc.so.6(clone+0x43)[0x7fc879b1ce73] |
crash trace analyzer results:
Line decoded:
|
/lib64/libthreadpool.so(_ZN10threadpool14FairThreadPool12sendErrorMsgEjjN5boost10shared_ptrIN11messageqcpp8IOSocketEEE+0x1ab)[0x7fc87aeb6b7b] |
File:
|
/usr/src/debug/MariaDB-/src_0/storage/columnstore/columnstore/utils/threadpool/fair_threadpool.cpp:311 |
|
|
result:
|
void FairThreadPool::sendErrorMsg(uint32_t id, uint32_t step, primitiveprocessor::SP_UM_IOSOCK sock) |
{
|
ISMPacketHeader ism;
|
PrimitiveHeader ph = {0, 0, 0, 0, 0, 0}; |
|
|
ism.Status = logging::primitiveServerErr;
|
ph.UniqueID = id;
|
ph.StepID = step;
|
messageqcpp::ByteStream msg(sizeof(ISMPacketHeader) + sizeof(PrimitiveHeader)); <--- 311 |
msg.append((uint8_t*)&ism, sizeof(ism));
|
msg.append((uint8_t*)&ph, sizeof(ph));
|
|
|
sock->write(msg);
|
}
|