From 2fb09671658af441991118dbb345e5b658d55f1e Mon Sep 17 00:00:00 2001
From: =?utf8?q?Milo=C5=A1=20Mula=C4=8D?= <mulac@civ.zcu.cz>
Date: Tue, 29 Jul 2008 13:15:05 +0000
Subject: [PATCH] fix #29041 - Handle DB reconnect wrt. transactions - no
 reconnect number limit

---
 org.glite.lb.common/interface/context.h |  1 +
 org.glite.lb.server/src/db_supp.c       | 14 ++++++++++++++
 org.glite.lbjp-common.db/src/db.c       | 10 +++++++++-
 3 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/org.glite.lb.common/interface/context.h b/org.glite.lb.common/interface/context.h
index 02fe955..b94a521 100644
--- a/org.glite.lb.common/interface/context.h
+++ b/org.glite.lb.common/interface/context.h
@@ -194,6 +194,7 @@ typedef enum _edg_wll_ErrorCode {
 	EDG_WLL_IL_EVENTS_WAITING, /**< Interlogger still has events pending delivery. */
 	EDG_WLL_ERROR_COMPARE_EVENTS, /**< Two compared events differ. */
 	EDG_WLL_ERROR_DB_TRANS_DEADLOCK,	/**< Deadlock detected during DB operation. */
+	EDG_WLL_ERROR_DB_LOST_CONNECTION,	/**< Lost connection to DB */
 } edg_wll_ErrorCode;
 
 /**
diff --git a/org.glite.lb.server/src/db_supp.c b/org.glite.lb.server/src/db_supp.c
index 050a48b..cf5817e 100644
--- a/org.glite.lb.server/src/db_supp.c
+++ b/org.glite.lb.server/src/db_supp.c
@@ -18,6 +18,7 @@ int edg_wll_SetErrorDB(edg_wll_Context ctx) {
 	if (ctx->dbctx) {
 		code = glite_lbu_DBError(ctx->dbctx, NULL, &ed);
 		if (code == EDEADLOCK) code = EDG_WLL_ERROR_DB_TRANS_DEADLOCK;
+		if (code == ERESTART) code = EDG_WLL_ERROR_DB_LOST_CONNECTION;
 		edg_wll_SetError(ctx, code, ed);
 		free(ed);
 	} else {
@@ -95,6 +96,7 @@ int edg_wll_TransNeedRetry(edg_wll_Context ctx) {
 	char *errd;
 
 	ret = edg_wll_Error(ctx,NULL,NULL);
+
 	if (ret == EDG_WLL_ERROR_DB_TRANS_DEADLOCK) {
 		if (debug)
 			printf("[%d]: DB deadlock detected. Rolling back transaction and retrying... \n",getpid());
@@ -104,6 +106,18 @@ int edg_wll_TransNeedRetry(edg_wll_Context ctx) {
 		edg_wll_Rollback(ctx);
 		edg_wll_ResetError(ctx);
 		return 1;
+	}
+	if (ret == EDG_WLL_ERROR_DB_LOST_CONNECTION) {
+		if (debug)
+			printf("[%d]: Lost connection to DB. "
+				"Rolling back transaction and retrying... \n",getpid());
+		else 
+			syslog(LOG_INFO,"[%d]: Lost connection to DB. "
+				"Rolling back transaction and retrying... \n",getpid());
+
+		edg_wll_Rollback(ctx);
+		edg_wll_ResetError(ctx);
+		return 1;
 	} else if (ret==0) {
 		edg_wll_Commit(ctx); /* errors propagated further */
 		return 0;
diff --git a/org.glite.lbjp-common.db/src/db.c b/org.glite.lbjp-common.db/src/db.c
index 0686f62..033b9f9 100644
--- a/org.glite.lbjp-common.db/src/db.c
+++ b/org.glite.lbjp-common.db/src/db.c
@@ -55,6 +55,7 @@ struct glite_lbu_DBContext_s {
 		int code;
 		char *desc;
 	} err;
+	int in_transaction;	/* this flag is set whenever we are in DB transaction */
 };
 
 
@@ -311,6 +312,7 @@ int glite_lbu_Transaction(glite_lbu_DBContext ctx) {
 	if (USE_TRANS(ctx)) {
 		if (glite_lbu_ExecSQL(ctx, "SET AUTOCOMMIT=0", NULL) < 0) goto err;
 		if (glite_lbu_ExecSQL(ctx, "BEGIN", NULL) < 0) goto err;
+		ctx->in_transaction = 1;
 	}
 err:
 	return STATUS(ctx);
@@ -322,6 +324,7 @@ int glite_lbu_Commit(glite_lbu_DBContext ctx) {
 	if (USE_TRANS(ctx)) {
 		if (glite_lbu_ExecSQL(ctx, "COMMIT", NULL) < 0) goto err;
 		if (glite_lbu_ExecSQL(ctx, "SET AUTOCOMMIT=1", NULL) < 0) goto err;
+		ctx->in_transaction = 0;
 	}
 err:
 	return STATUS(ctx);
@@ -333,6 +336,7 @@ int glite_lbu_Rollback(glite_lbu_DBContext ctx) {
 	if (USE_TRANS(ctx)) { 
 		if (glite_lbu_ExecSQL(ctx, "ROLLBACK", NULL) < 0) goto err;
 		if (glite_lbu_ExecSQL(ctx, "SET AUTOCOMMIT=1", NULL) < 0) goto err;
+		ctx->in_transaction = 0;
 	}
 err:
 	return STATUS(ctx);
@@ -495,7 +499,11 @@ int glite_lbu_ExecSQL(glite_lbu_DBContext ctx, const char *cmd, glite_lbu_Statem
 					break;
 				case CR_SERVER_LOST:
 				case CR_SERVER_GONE_ERROR:
-					if (retry_nr <= 0) 
+					if (ctx->in_transaction) {
+						ERR(ctx, ERESTART, db_handle.mysql_error(ctx->mysql));
+						return -1;
+					}
+					else if (retry_nr <= 0) 
 						do_reconnect = 1;
 					break;
 				case ER_LOCK_DEADLOCK:
-- 
1.8.2.3