DAOS-15431 vos: clear DTX in CoS cache for the one committed by other…

… - b24

It is possible that the DTX is committed locally via commit-on-share
mechanism during some subsequent modification. But such modification
may get failure on some remote paraticipant. Then the DTX entry will
not be removed from CoS cache under such case because we do not know
whether such DTX has already been committed on remote participant(s)
successfully or not. Keeping it in CoS cache will allow us to re-try
the commit globally sometime later.

Next time, when we re-try to commit such DTX globally, if all remote
participants claim as done (return 0 or DER_NONEXIST or DER_ALREADY)
successfully, then even if local VOS calims as DER_ALREADY, we still
need to remove the DTX entry from CoS cache. Otherwise, we will find
such DTX entry in the CoS cache everytime for batched commit (or DTX
resync) as to block related process.

Signed-off-by: Fan Yong <[email protected]>

src/dtx/dtx_common.c

@@ -1555,6 +1555,7 @@ dtx_end(struct dtx_handle *dth, struct ds_cont_child *cont, int result)
 static void
 dtx_flush_on_close(struct dss_module_info *dmi, struct dtx_batched_cont_args *dbca)
 {
+	struct dss_xstream	*dx = dss_current_xstream();
 	struct ds_cont_child	*cont = dbca->dbca_cont;
 	struct dtx_stat		 stat = { 0 };
 	uint64_t		 total = 0;
@@ -1564,7 +1565,8 @@ dtx_flush_on_close(struct dss_module_info *dmi, struct dtx_batched_cont_args *db
 	dtx_stat(cont, &stat);
 
 	/* dbca->dbca_reg_gen != cont->sc_dtx_batched_gen means someone reopen the container. */
-	while (dbca->dbca_reg_gen == cont->sc_dtx_batched_gen && rc >= 0) {
+	while (!dss_xstream_exiting(dx) &&
+	       dbca->dbca_reg_gen == cont->sc_dtx_batched_gen && rc >= 0) {
 		struct dtx_entry	**dtes = NULL;
 		struct dtx_cos_key	 *dcks = NULL;
 
@@ -2197,10 +2199,11 @@ int
 dtx_obj_sync(struct ds_cont_child *cont, daos_unit_oid_t *oid,
 	     daos_epoch_t epoch)
 {
-	int	cnt;
-	int	rc = 0;
+	struct dss_xstream	*dx = dss_current_xstream();
+	int			 cnt;
+	int			 rc = 0;
 
-	while (dtx_cont_opened(cont)) {
+	while (!dss_xstream_exiting(dx) && (dtx_cont_opened(cont) || oid == NULL)) {
 		struct dtx_entry	**dtes = NULL;
 		struct dtx_cos_key	 *dcks = NULL;
 

src/vos/vos_dtx.c

@@ -871,7 +871,8 @@ vos_dtx_commit_one(struct vos_container *cont, struct dtx_id *dti, daos_epoch_t
 	if (rc != 0)
 		D_FREE(dce);
 
-	if (rm_cos != NULL && (rc == 0 || rc == -DER_NONEXIST))
+	if (rm_cos != NULL &&
+	    (rc == 0 || rc == -DER_NONEXIST || (rc == -DER_ALREADY && dae == NULL)))
 		*rm_cos = true;
 
 	return rc;