From: Aleš Křenek <ljocha@ics.muni.cz>
Date: Fri, 23 Mar 2007 11:55:17 +0000 (+0000)
Subject: - avoid deadlock by keeping only a single semaphore at time
X-Git-Tag: merge_31_head_dst~2
X-Git-Url: http://scientific.zcu.cz/git/?a=commitdiff_plain;h=29c0e326f686e86025472d4ee40cc2043371af71;p=jra1mw.git

- avoid deadlock by keeping only a single semaphore at time
- minor fixes (missing prototypes etc.)
---

diff --git a/org.glite.lb.server/Makefile b/org.glite.lb.server/Makefile
index a0c93fa..33b33e0 100644
--- a/org.glite.lb.server/Makefile
+++ b/org.glite.lb.server/Makefile
@@ -382,7 +382,7 @@ install:
 		mkdir -p ${PREFIX}/include/${globalprefix}/${lbprefix} ; \
 		(cd ${top_srcdir}/interface && install -m 644 ${HDRS} ${PREFIX}/include/${globalprefix}/${lbprefix}) ; \
 		install -m 644 ${STATIC_LIB_BK} ${PREFIX}/lib; \
-		if [ x${LB_STANDALONE} = x ]; then \
+		if [ x${LB_STANDALONE} = x -a x${PLUGIN_LIB} != x ]; then \
 			${INSTALL} -m 755 ${PLUGIN_LIB} ${PREFIX}/lib; \
 		fi; \
 		${INSTALL} -m 644 ${top_srcdir}/interface/srv_perf.h ${PREFIX}/include/${globalprefix}/${lbprefix}; \
diff --git a/org.glite.lb.server/interface/store.h b/org.glite.lb.server/interface/store.h
index c42bef2..17e71ac 100644
--- a/org.glite.lb.server/interface/store.h
+++ b/org.glite.lb.server/interface/store.h
@@ -45,6 +45,8 @@ int db_store(edg_wll_Context,char *, char *);
 int db_parent_store(edg_wll_Context, edg_wll_Event *, intJobStat *);
 int handle_request(edg_wll_Context,char *);
 int create_reply(const edg_wll_Context,char **);
+int trans_db_store(edg_wll_Context,char *,edg_wll_Event *,intJobStat *);
+
 
 int edg_wll_delete_event(edg_wll_Context,const char *, int);
 
diff --git a/org.glite.lb.server/src/db_store.c b/org.glite.lb.server/src/db_store.c
index e749423..4513638 100644
--- a/org.glite.lb.server/src/db_store.c
+++ b/org.glite.lb.server/src/db_store.c
@@ -15,6 +15,7 @@
 #include "lbs_db.h"
 #include "lock.h"
 #include "il_lbproxy.h"
+#include "jobstat.h"
 
 #ifdef LB_PERF
 #include "glite/lb/lb_perftest.h"
@@ -95,7 +96,9 @@ db_store(edg_wll_Context ctx,char *ucs, char *event)
     err = edg_wll_StepIntState(ctx,ev->any.jobId, ev, seq, ctx->isProxy? NULL: &newstat);
   }
 
-  if (edg_wll_UnlockJob(ctx,ev->any.jobId)) goto err;
+  /* XXX: in edg_wll_StepIntState() 
+   * if (edg_wll_UnlockJob(ctx,ev->any.jobId)) goto err;
+   */
   if (err) goto err;
 
   db_actual_store(ctx, event, ev, &newstat);
diff --git a/org.glite.lb.server/src/jobstat.c b/org.glite.lb.server/src/jobstat.c
index 6573889..e03ce05 100644
--- a/org.glite.lb.server/src/jobstat.c
+++ b/org.glite.lb.server/src/jobstat.c
@@ -972,6 +972,8 @@ edg_wll_ErrorCode edg_wll_StepIntStateParent(edg_wll_Context ctx,
 /*
  * update stored state according to the new event
  * (must be called with the job locked)
+
+ * XXX: job is locked on entry, unlocked in this function
  */
 
 edg_wll_ErrorCode edg_wll_StepIntState(edg_wll_Context ctx,
@@ -1000,9 +1002,12 @@ edg_wll_ErrorCode edg_wll_StepIntState(edg_wll_Context ctx,
 		res = processEvent(ijsp, e, seq, be_strict, &errstring);
 		if (res == RET_FATAL || res == RET_INTERNAL) { /* !strict */
 			edg_wll_FreeStatus(&oldstat);
+			edg_wll_UnlockJob(ctx,job); /* XXX: error lost */
 			return edg_wll_SetError(ctx, EINVAL, errstring);
 		}
 		edg_wll_StoreIntState(ctx, ijsp, seq);
+		if (edg_wll_UnlockJob(ctx,job)) goto err;
+
 		edg_wll_UpdateStatistics(ctx,&oldstat,e,&ijsp->pub);
 
 		/* check whether subjob state change does not change parent state */
@@ -1026,6 +1031,12 @@ edg_wll_ErrorCode edg_wll_StepIntState(edg_wll_Context ctx,
 		/* FIXME: we miss state change in the case of seq != 0 
 		 * Does anybody care? */
 
+		/* FIXME: collection parent status is wrong in this case.
+		   However, it should not happen (frequently).
+		   Right approach is computing parent status from scratch.
+		*/
+
+		if (edg_wll_UnlockJob(ctx,job)) goto err;
 		edg_wll_UpdateStatistics(ctx,NULL,e,&jobstat.pub);
 
 		if (ctx->rgma_export) write2rgma_status(&jobstat.pub);
@@ -1036,6 +1047,8 @@ edg_wll_ErrorCode edg_wll_StepIntState(edg_wll_Context ctx,
 		}
 		else destroy_intJobStat(&jobstat);
 	}
+	else edg_wll_UnlockJob(ctx,job);
+err:
 	return edg_wll_Error(ctx, NULL, NULL);
 }
 
diff --git a/org.glite.lb.server/src/jobstat.h b/org.glite.lb.server/src/jobstat.h
index 84f3e32..8514751 100644
--- a/org.glite.lb.server/src/jobstat.h
+++ b/org.glite.lb.server/src/jobstat.h
@@ -78,6 +78,9 @@ edg_wll_ErrorCode edg_wll_StoreIntState(edg_wll_Context, intJobStat *, int);
 edg_wll_ErrorCode edg_wll_StoreIntStateEmbryonic(edg_wll_Context, edg_wlc_JobId, char *icnames, char *values, edg_wll_bufInsert *bi);
 edg_wll_ErrorCode edg_wll_LoadIntState(edg_wll_Context , edg_wlc_JobId , int, intJobStat **);
 
+edg_wll_ErrorCode edg_wll_StepIntStateParent(edg_wll_Context,edg_wlc_JobId,edg_wll_Event *,int,intJobStat *,edg_wll_JobStat *);
+
+
 
 intJobStat* dec_intJobStat(char *, char **);
 char *enc_intJobStat(char *, intJobStat* );