PG异常处理

在PG中经常可以看到ereport错误日志，本以为是直接打印日志输出，实际更加复杂。另外PG用C实现了TRY-CATCH语义，这点也十分神奇。

setjmp宏

要探究PG中的异常跳转，首先得熟悉setjmp宏，这里推荐菜鸟教程的一篇短文：

• https://www.runoob.com/cprogramming/c-macro-setjmp.html
• https://www.runoob.com/cprogramming/c-function-longjmp.html

#include <setjmp.h>

int setjmp(jmp_buf env);
void longjmp(jmp_buf env, int val);

主要思想为：通过jmp_buf类型的全局变量保存程序切换的上下文，setjmp将jmp_buf初始化同时创建保存点，第一次调用返回0。然后再通过longjmp返回第一次调用setjmp处，setjmp的返回值为longjmp的第二个整型参数。

ereport 日志输出

在PG中，我们经常可以看到以下的错误处理代码：

if (!OidIsValid(get_role_oid(user_name_reset, false))) {
	ereport(ERROR, (errcode(ERRCODE_UNDEFINED_OBJECT), errmsg("role \"%s\" does not exist", user_name_reset)));
}

ereport接口定义如下：

/* ----------
 * New-style error reporting API: to be used in this way:
 *		ereport(ERROR,
 *				(errcode(ERRCODE_UNDEFINED_CURSOR),
 *				 errmsg("portal \"%s\" not found", stmt->portalname),
 *				 ... other errxxx() fields as needed ...));
 *
 * The error level is required, and so is a primary error message (errmsg
 * or errmsg_internal).  All else is optional.	errcode() defaults to
 * ERRCODE_INTERNAL_ERROR if elevel is ERROR or more, ERRCODE_WARNING
 * if elevel is WARNING, or ERRCODE_SUCCESSFUL_COMPLETION if elevel is
 * NOTICE or below.
 *
 * ereport_domain() allows a message domain to be specified, for modules that
 * wish to use a different message catalog from the backend's.	To avoid having
 * one copy of the default text domain per .o file, we define it as NULL here
 * and have errstart insert the default text domain.  Modules can either use
 * ereport_domain() directly, or preferably they can override the TEXTDOMAIN
 * macro.
 * ----------
 */
#define ereport_domain(elevel, domain, rest) \
    (errstart(elevel, __FILE__, __LINE__, PG_FUNCNAME_MACRO, domain) ? (errfinish rest) : (void)0)

#define ereport(elevel, rest) ereport_domain(elevel, TEXTDOMAIN, rest)

ereport其实就是调用errstart，然后根据errstar的返回值决定调用errfinish还是什么都不做。

errstart

介绍errstart之前先介绍一个数据结构，用于错误数据信息

#define ERRORDATA_STACK_SIZE  5
static ErrorData errordata[ERRORDATA_STACK_SIZE];

elog.c会有一个错误处理的栈，这是因为我们在错误处理的时候还可能会发生错误。

errstart主要完成：

• 初始化errdata信息
• 将errdata压入errodata栈

errfinish

在关注errfinish之前，我们先关注errfinish的参数。在宏定义中，errfinish用于接收rest参数，而在实际使用中，rest这个参数实际上是由括号包裹的一串调用，分别是errcode和errmsg。

ereport(ERROR, (errcode(ERRCODE_UNDEFINED_OBJECT), errmsg("role \"%s\" does not exist", user_name_reset)));

/*
 * errcode --- add SQLSTATE error code to the current error
 *
 * The code is expected to be represented as per MAKE_SQLSTATE().
 */
int errcode(int sqlerrcode)
{
    ErrorData* edata = &t_thrd.log_cxt.errordata[t_thrd.log_cxt.errordata_stack_depth];

    /* we don't bother incrementing t_thrd.log_cxt.recursion_depth */
    CHECK_STACK_DEPTH();

    edata->sqlerrcode = sqlerrcode;

    return 0; /* return value does not matter */
}
/*
 * errmsg --- add a primary error message text to the current error
 *
 * In addition to the usual %-escapes recognized by printf, "%m" in
 * fmt is replaced by the error message for the caller's value of errno.
 *
 * Note: no newline is needed at the end of the fmt string, since
 * ereport will provide one for the output methods that need it.
 */
int errmsg(const char* fmt, ...)
{
    ErrorData* edata = &t_thrd.log_cxt.errordata[t_thrd.log_cxt.errordata_stack_depth];
    MemoryContext oldcontext;

    t_thrd.log_cxt.recursion_depth++;
    CHECK_STACK_DEPTH();
    oldcontext = MemoryContextSwitchTo(ErrorContext);

    EVALUATE_MESSAGE(message, false, true);

    MemoryContextSwitchTo(oldcontext);
    t_thrd.log_cxt.recursion_depth--;
    return 0; /* return value does not matter */
}

errmsg通过EVALUATE_MESSAGE宏格式化字符串，将错误信息字符串计入到edata中的message中？

（1）error以下等级

然后再看errfinish的函数本体：

• 完成error_context_stack的回调功能，为errfinish增加报错信息
• 完成EmitErrorReport，为errfinish发送错误信息

for (econtext = t_thrd.log_cxt.error_context_stack; econtext != NULL; econtext = econtext->previous)
    (*econtext->callback)(econtext->arg);
EmitErrorReport();

至此，一个简单的日志输出完成，返回调用ereport的代码处继续执行。

（2）error以上等级

error以上等级不再返回，调用 PG_RE_THROW(); 完成长跳转，根据PG _exception_stack记录的保存点进行跳转。PG _exception_stack是一个全局变量，在postgresMain中你会看到sigsetjmp的设置

if (sigsetjmp(local_sigjmp_buf, 1) != 0) {

TRY-CATCAH语义实现

/* ----------
 * API for catching ereport(ERROR) exits.  Use these macros like so:
 *
 *		PG_TRY();
 *		{
 *			... code that might throw ereport(ERROR) ...
 *		}
 *		PG_CATCH();
 *		{
 *			... error recovery code ...
 *		}
 *		PG_END_TRY();
 *
 * (The braces are not actually necessary, but are recommended so that
 * pg_indent will indent the construct nicely.)  The error recovery code
 * can optionally do PG_RE_THROW() to propagate the same error outwards.
 *
 * Note: while the system will correctly propagate any new ereport(ERROR)
 * occurring in the recovery section, there is a small limit on the number
 * of levels this will work for.  It's best to keep the error recovery
 * section simple enough that it can't generate any new errors, at least
 * not before popping the error stack.
 *
 * Note: an ereport(FATAL) will not be caught by this construct; control will
 * exit straight through proc_exit().  Therefore, do NOT put any cleanup
 * of non-process-local resources into the error recovery section, at least
 * not without taking thought for what will happen during ereport(FATAL).
 * The PG_ENSURE_ERROR_CLEANUP macros provided by storage/ipc.h may be
 * helpful in such cases.
 *
 * Note: Don't execute statements such as break, continue, goto, or return in
 * PG_TRY. If you need to use these statements, you must recovery
 * PG_exception_stack first.
 * ----------
 */
#define PG_TRY()                                                                       \
    do {                                                                               \
        sigjmp_buf* save_exception_stack = t_thrd.log_cxt.PG_exception_stack;          \
        ErrorContextCallback* save_context_stack = t_thrd.log_cxt.error_context_stack; \
        sigjmp_buf local_sigjmp_buf;                                                   \
        int tryCounter, *oldTryCounter = NULL;                                         \
        if (sigsetjmp(local_sigjmp_buf, 0) == 0) {                                     \
            t_thrd.log_cxt.PG_exception_stack = &local_sigjmp_buf;                     \
        oldTryCounter = gstrace_tryblock_entry(&tryCounter)

#define PG_CATCH()                                                		\
        }                                                             \
        else                                                          \
        {                                                             \
            t_thrd.log_cxt.PG_exception_stack = save_exception_stack; \
            t_thrd.log_cxt.error_context_stack = save_context_stack;  \
            gstrace_tryblock_exit(true, oldTryCounter)

#define PG_END_TRY()                                          		\
        }                                                         \
        t_thrd.log_cxt.PG_exception_stack = save_exception_stack; \
        t_thrd.log_cxt.error_context_stack = save_context_stack;  \
        gstrace_tryblock_exit(false, oldTryCounter);              \
    }while (0)

在PG_TRY() 中主要设置了一个本地跳转保存点local_sigjmp_buf ，并将全局变量PG_exception_stack 设置为它，那么在发生错误时大于error等级的ereport调用就会返回此处，进而执行else分支（也就是PG_CATCH() 所在部分的代码）。

注意在CATCH中加载保存的状态点。

---

reference：https://blog.csdn.net/jackgo73/article/details/135130840