database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-04-18 21:44:02 +03:00
Code

fix(MCOL-5396): Fix possible infinite loop in plugin--PrimProc communication

Browse Source 
If you manage to shut down PrimProc just before plugin is trying to send
Calpont Select Execution Plan to PrimProc, you now get a nice error
message about PrimProc being down instead of endless logs of failed
reconnection attempts.
This commit is contained in:
Serguey Zefirov 2025-02-28 14:05:16 +03:00 committed by Leonid Fedorov
parent 8e50884928
commit ef592038cb
1 changed files with 51 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
dbcon/mysql/ha_mcs_impl.cpp
View File

@ -19,6 +19,8 @@
#define PREFER_MY_CONFIG_H
#include <my_config.h>
#include <unistd.h>
#include <chrono>
#include <thread>
#include <string>
#include <iostream>
#include <stack>
@ -130,6 +132,7 @@ using namespace funcexp;
#include "ha_mcs_logging.h"
#include "ha_subquery.h"
namespace cal_impl_if
{
extern bool nonConstFunc(Item_func* ifp);
@ -2352,8 +2355,18 @@ int ha_mcs::impl_rnd_init(TABLE* table, const std::vector<COND*>& condStack)
ByteStream msg;
ByteStream emsgBs;
int ntries = 10;
// XXX: MCOL-5396: unable to reach this code.
while (true)
{
string emsg;
if (ntries < 0)
{
emsg = "Lost connection to ExeMgr. Please contact your administrator";
setError(thd, ER_INTERNAL_ERROR, emsg);
return ER_INTERNAL_ERROR;
}
try
{
ByteStream::quadbyte qb = 4;
@ -2371,7 +2384,6 @@ int ha_mcs::impl_rnd_init(TABLE* table, const std::vector<COND*>& condStack)
emsgBs.restart();
msg = hndl->exeMgr->read();
emsgBs = hndl->exeMgr->read();
string emsg;
if (msg.length() == 0 || emsgBs.length() == 0)
{
@ -2437,6 +2449,10 @@ int ha_mcs::impl_rnd_init(TABLE* table, const std::vector<COND*>& condStack)
ti.conn_hndl = hndl;
using namespace chrono_literals;
std::this_thread::sleep_for(100ms);
ntries --;
try
{
hndl->connect();
@ -4239,8 +4255,18 @@ int ha_mcs_impl_group_by_init(mcs_handler_info* handler_info, TABLE* table)
ByteStream msg;
ByteStream emsgBs;
int ntries = 10;
// XXX: MCOL-5396: unable to reach this code.
while (true)
{
string emsg;
if (ntries < 0)
{
emsg = "Lost connection to ExeMgr. Please contact your administrator";
setError(thd, ER_INTERNAL_ERROR, emsg);
return ER_INTERNAL_ERROR;
}
try
{
ByteStream::quadbyte qb = 4;
@ -4258,7 +4284,6 @@ int ha_mcs_impl_group_by_init(mcs_handler_info* handler_info, TABLE* table)
emsgBs.restart();
msg = hndl->exeMgr->read();
emsgBs = hndl->exeMgr->read();
string emsg;
if (msg.length() == 0 || emsgBs.length() == 0)
{
@ -4325,6 +4350,11 @@ int ha_mcs_impl_group_by_init(mcs_handler_info* handler_info, TABLE* table)
ci->cal_conn_hndl = hndl;
ci->cal_conn_hndl_st.pop();
ci->cal_conn_hndl_st.push(ci->cal_conn_hndl);
using namespace std::chrono_literals;
std::this_thread::sleep_for(100ms);
ntries --;
try
{
hndl->connect();
@ -4943,8 +4973,23 @@ int ha_mcs_impl_pushdown_init(mcs_handler_info* handler_info, TABLE* table, bool
ByteStream msg;
ByteStream emsgBs;
int ntries = 10;
// The issue is MCOL-5396.
// The delay below is used to trigger old infinite loop condition and
// prove that mitigation works. While it looks like unused code, it is
// important enough to have it just in case.
// using namespace std::chrono_literals;
// std::this_thread::sleep_for(10000ms); // shut PrimProc down now.
while (true)
{
string emsg;
if (ntries < 0)
{
emsg = "Lost connection to ExeMgr. Please contact your administrator";
setError(thd, ER_INTERNAL_ERROR, emsg);
return ER_INTERNAL_ERROR;
}
try
{
ByteStream::quadbyte qb = 4;
@ -4962,7 +5007,6 @@ int ha_mcs_impl_pushdown_init(mcs_handler_info* handler_info, TABLE* table, bool
emsgBs.restart();
msg = hndl->exeMgr->read();
emsgBs = hndl->exeMgr->read();
string emsg;
if (msg.length() == 0 || emsgBs.length() == 0)
{
@ -5030,6 +5074,10 @@ int ha_mcs_impl_pushdown_init(mcs_handler_info* handler_info, TABLE* table, bool
ci->cal_conn_hndl = hndl;
using namespace std::chrono_literals;
std::this_thread::sleep_for(100ms);
ntries --;
try
{
hndl->connect();