db_cluster: if a db query fails, mark connection inactive

- default inactive time is 300sec
- value can be chanded via inactive_interval parameter

modules_k/db_cluster/db_cluster_mod.c

@@ -36,6 +36,8 @@ int db_cluster_bind_api(db_func_t *dbb);
 int dbcl_con_param(modparam_t type, void *val);
 int dbcl_cls_param(modparam_t type, void *val);
 
+int dbcl_inactive_interval = 300;
+
 /*! \brief
  * DB Cluster module interface
  */
@@ -50,6 +52,7 @@ static cmd_export_t cmds[] = {
 static param_export_t params[] = {
 	{"connection",  STR_PARAM|USE_FUNC_PARAM, (void*)dbcl_con_param},
 	{"cluster",     STR_PARAM|USE_FUNC_PARAM, (void*)dbcl_cls_param},
+	{"inactive_interval",     INT_PARAM,    &dbcl_inactive_interval},
 	{0, 0, 0}
 };
 

modules_k/db_cluster/dbcl_api.c

@@ -47,7 +47,7 @@
 	db1_con_t  *dbh=NULL;\
 	dbcl_cls_t *cls=NULL;\
 	cls = (dbcl_cls_t*)_h->tail;\
-	ret = 0;\
+	ret = -1;\
 	for(i=DBCL_PRIO_SIZE-1; i>0; i--)\
 	{\
 		switch(cls->rlist[i].mode) {\
@@ -55,8 +55,7 @@
 			case 'S':\
 				for(j=0; j<cls->rlist[i].clen; j++)\
 				{\
-					if(cls->rlist[i].clist[j] != NULL && cls->rlist[i].clist[j]->flags!=0\
-						&& cls->rlist[i].clist[j]->dbh != NULL)\
+					if(dbcl_valid_con(cls->rlist[i].clist[j])==0)\
 					{\
 						LM_DBG("serial operation - cluster [%.*s] (%d/%d)\n",\
 								cls->name.len, cls->name.s, i, j);\
@@ -65,6 +64,11 @@
 						if (ret==0) {\
 							cls->usedcon = cls->rlist[i].clist[j];\
 							return 0;\
+						} else {\
+							LM_DBG("serial operation - failre on cluster"\
+									" [%.*s] (%d/%d)\n",\
+									cls->name.len, cls->name.s, i, j);\
+							dbcl_inactive_con(cls->rlist[i].clist[j]);\
 						}\
 					}\
 				}\
@@ -74,8 +78,7 @@
 				for(k=0; k<cls->rlist[i].clen; k++)\
 				{\
 					j = (process_no + k + cls->rlist[i].crt) % cls->rlist[i].clen;\
-					if(cls->rlist[i].clist[j] != NULL && cls->rlist[i].clist[j]->flags!=0\
-						&& cls->rlist[i].clist[j]->dbh != NULL)\
+					if(dbcl_valid_con(cls->rlist[i].clist[j])==0)\
 					{\
 						LM_DBG("round robin operation - cluster [%.*s] (%d/%d)\n",\
 								cls->name.len, cls->name.s, i, j);\
@@ -86,6 +89,11 @@
 							cls->usedcon = cls->rlist[i].clist[j];\
 							cls->rlist[i].crt = (j+1) % cls->rlist[i].clen;\
 							return 0;\
+						} else {\
+							LM_DBG("round robin operation - failre on cluster"\
+									" [%.*s] (%d/%d)\n",\
+									cls->name.len, cls->name.s, i, j);\
+							dbcl_inactive_con(cls->rlist[i].clist[j]);\
 						}\
 					}\
 				}\
@@ -110,7 +118,7 @@
 	db1_con_t  *dbh=NULL;\
 	dbcl_cls_t *cls=NULL;\
 	cls = (dbcl_cls_t*)_h->tail;\
-	ret = 0;\
+	ret = -1;\
 	rok = 0;\
 	rc = 0;\
 	for(i=DBCL_PRIO_SIZE-1; i>0; i--)\
@@ -120,8 +128,7 @@
 			case 'S':\
 				for(j=0; j<cls->wlist[i].clen; j++)\
 				{\
-					if(cls->wlist[i].clist[j] != NULL && cls->wlist[i].clist[j]->flags!=0\
-						&& cls->wlist[i].clist[j]->dbh != NULL)\
+					if(dbcl_valid_con(cls->wlist[i].clist[j])==0)\
 					{\
 						LM_DBG("serial operation - cluster [%.*s] (%d/%d)\n",\
 								cls->name.len, cls->name.s, i, j);\
@@ -130,6 +137,11 @@
 						if (ret==0) {\
 							cls->usedcon = cls->wlist[i].clist[j];\
 							return 0;\
+						} else {\
+							LM_DBG("serial operation - failure on cluster"\
+									" [%.*s] (%d/%d)\n",\
+									cls->name.len, cls->name.s, i, j);\
+							dbcl_inactive_con(cls->wlist[i].clist[j]);\
 						}\
 					}\
 				}\
@@ -139,8 +151,7 @@
 				for(k=0; k<cls->wlist[i].clen; k++)\
 				{\
 					j = (process_no + k + cls->wlist[i].crt) % cls->wlist[i].clen;\
-					if(cls->wlist[i].clist[j] != NULL && cls->wlist[i].clist[j]->flags!=0\
-						&& cls->wlist[i].clist[j]->dbh != NULL)\
+					if(dbcl_valid_con(cls->wlist[i].clist[j])==0)\
 					{\
 						LM_DBG("round robin operation - cluster [%.*s] (%d/%d)\n",\
 								cls->name.len, cls->name.s, i, j);\
@@ -151,6 +162,11 @@
 							cls->usedcon = cls->wlist[i].clist[j];\
 							cls->wlist[i].crt = (j+1) % cls->wlist[i].clen;\
 							return 0;\
+						} else {\
+							LM_DBG("round robin operation - failure on cluster"\
+									" [%.*s] (%d/%d)\n",\
+									cls->name.len, cls->name.s, i, j);\
+							dbcl_inactive_con(cls->wlist[i].clist[j]);\
 						}\
 					}\
 				}\
@@ -159,8 +175,7 @@
 			case 'P':\
 				for(j=0; j<cls->wlist[i].clen; j++)\
 				{\
-					if(cls->wlist[i].clist[j] != NULL && cls->wlist[i].clist[j]->flags!=0\
-						&& cls->wlist[i].clist[j]->dbh != NULL)\
+					if(dbcl_valid_con(cls->wlist[i].clist[j])==0)\
 					{\
 						LM_DBG("parallel operation - cluster [%.*s] (%d/%d)\n",\
 								cls->name.len, cls->name.s, i, j);\
@@ -169,6 +184,11 @@
 						if(rc==0) {\
 							cls->usedcon = cls->wlist[i].clist[j];\
 							rok = 1;\
+						} else {\
+							LM_DBG("parallel operation - failure on cluster"\
+									" [%.*s] (%d/%d)\n",\
+									cls->name.len, cls->name.s, i, j);\
+							dbcl_inactive_con(cls->wlist[i].clist[j]);\
 						}\
 						ret |= rc;\
 					}\

modules_k/db_cluster/dbcl_data.c

@@ -32,6 +32,7 @@
 #include "../../dprint.h"
 #include "../../hashes.h"
 #include "../../trim.h"
+#include "../../timer.h"
 #include "../../mem/mem.h"
 #include "../../mem/shm_mem.h"
 
@@ -123,6 +124,33 @@ int dbcl_init_con(str *name, str *url)
 	return 0;
 }
 
+int dbcl_valid_con(dbcl_con_t *sc)
+{
+	if(sc==NULL || sc->flags!=0 || sc->dbh!=NULL)
+		return -1;
+	if(sc->sinfo==NULL)
+		return 0;
+	if(sc->sinfo->state & DBCL_CON_INACTIVE)
+	{
+		if(sc->sinfo->aticks>0 && sc->sinfo->aticks<get_ticks())
+			return -1;
+		sc->sinfo->aticks = 0;
+		sc->sinfo->state &= ~DBCL_CON_INACTIVE;
+	}
+	return 0;
+}
+
+extern int dbcl_inactive_interval;
+
+int dbcl_inactive_con(dbcl_con_t *sc)
+{
+	if(sc==NULL || sc->sinfo==NULL)
+		return -1;
+	sc->sinfo->aticks = get_ticks() + dbcl_inactive_interval;
+	sc->sinfo->state |= DBCL_CON_INACTIVE;
+	return 0;
+}
+
 int dbcl_parse_con_param(char *val)
 {
 	str name;

modules_k/db_cluster/dbcl_data.h

@@ -40,10 +40,12 @@
 #define DBCL_PRIO_SIZE	10
 #define DBCL_CLIST_SIZE	5
 
+#define DBCL_CON_INACTIVE	1
+
 typedef struct dbcl_shared
 {
 	int state;
-	int count;
+	unsigned int aticks;
 } dbcl_shared_t;
 
 typedef struct dbcl_con
@@ -84,6 +86,9 @@ int dbcl_init_connections(dbcl_cls_t *cls);
 int dbcl_close_connections(dbcl_cls_t *cls);
 dbcl_cls_t *dbcl_get_cluster(str *name);
 
+int dbcl_valid_con(dbcl_con_t *sc);
+int dbcl_inactive_con(dbcl_con_t *sc);
+
 int dbcl_parse_con_param(char *val);
 int dbcl_parse_cls_param(char *val);
 #endif /* KM_DBASE_H */