Branch HEAD

b=19154
i=rread
i=tappro

an ability to wait for the completion of an MDS-OST synchronization is added

diff --git a/lustre/include/obd.h b/lustre/include/obd.h
index e596a23..10da129 100644 (file)
--- a/lustre/include/obd.h
+++ b/lustre/include/obd.h
@@ -329,8 +329,9 @@ struct filter_obd {
         int                  fo_tot_granted_clients;
 
         obd_size             fo_readcache_max_filesize;
-        int                  fo_read_cache;
-        int                  fo_writethrough_cache;
+        int                  fo_read_cache:1,   /**< enable read-only cache */
+                             fo_writethrough_cache:1,/**< read cache writes */
+                             fo_mds_ost_sync:1; /**< MDS-OST orphan recovery*/
 
         struct obd_import   *fo_mdc_imp;
         struct obd_uuid      fo_mdc_uuid;

diff --git a/lustre/obdclass/llog_cat.c b/lustre/obdclass/llog_cat.c
index edf822b..fdc7248 100644 (file)
--- a/lustre/obdclass/llog_cat.c
+++ b/lustre/obdclass/llog_cat.c
@@ -446,7 +446,7 @@ int llog_cat_process_thread(void *data)
         struct llog_process_cat_args *args = data;
         struct llog_ctxt *ctxt = args->lpca_ctxt;
         struct llog_handle *llh = NULL;
-        void  *cb = args->lpca_cb;
+        llog_cb_t cb = args->lpca_cb;
         struct llog_logid logid;
         int rc;
         ENTRY;
@@ -466,9 +466,10 @@ int llog_cat_process_thread(void *data)
         }
 
         if (cb) {
-                rc = llog_cat_process(llh, (llog_cb_t)cb, NULL, 0, 0);
+                rc = llog_cat_process(llh, cb, NULL, 0, 0);
                 if (rc != LLOG_PROC_BREAK && rc != 0)
                         CERROR("llog_cat_process() failed %d\n", rc);
+                cb(llh, NULL, NULL);
         } else {
                 CWARN("No callback function for recovery\n");
         }

diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c
index 5553323..21c24a7 100644 (file)
--- a/lustre/obdfilter/filter.c
+++ b/lustre/obdfilter/filter.c
@@ -2577,6 +2577,9 @@ static int filter_llog_connect(struct obd_export *exp,
               obd->obd_name, body->lgdc_logid.lgl_oid,
               body->lgdc_logid.lgl_ogr, body->lgdc_logid.lgl_ogen);
 
+        spin_lock_bh(&obd->obd_processing_task_lock);
+        obd->u.filter.fo_mds_ost_sync = 1;
+        spin_unlock_bh(&obd->obd_processing_task_lock);
         rc = llog_connect(ctxt, &body->lgdc_logid,
                           &body->lgdc_gen, NULL);
         llog_ctxt_put(ctxt);

diff --git a/lustre/obdfilter/filter_log.c b/lustre/obdfilter/filter_log.c
index d1f4c18..61f84a3 100644 (file)
--- a/lustre/obdfilter/filter_log.c
+++ b/lustre/obdfilter/filter_log.c
@@ -252,6 +252,13 @@ int filter_recov_log_mds_ost_cb(struct llog_handle *llh,
         if (ctxt->loc_obd->obd_stopping)
                 RETURN(LLOG_PROC_BREAK);
 
+        if (rec == NULL) {
+                spin_lock_bh(&ctxt->loc_obd->obd_processing_task_lock);
+                ctxt->loc_obd->u.filter.fo_mds_ost_sync = 0;
+                spin_unlock_bh(&ctxt->loc_obd->obd_processing_task_lock);
+                RETURN(0);
+        }
+
         if (!(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)) {
                 CERROR("log is not plain\n");
                 RETURN(-EINVAL);

diff --git a/lustre/obdfilter/lproc_obdfilter.c b/lustre/obdfilter/lproc_obdfilter.c
index 7d4da23..bcc46ec 100644 (file)
--- a/lustre/obdfilter/lproc_obdfilter.c
+++ b/lustre/obdfilter/lproc_obdfilter.c
@@ -299,7 +299,9 @@ static int lprocfs_filter_wr_cache(struct file *file, const char *buffer,
         if (rc)
                 return rc;
 
+        spin_lock_bh(&obd->obd_processing_task_lock);
         obd->u.filter.fo_read_cache = val;
+        spin_unlock_bh(&obd->obd_processing_task_lock);
         return count;
 }
 
@@ -324,10 +326,21 @@ static int lprocfs_filter_wr_wcache(struct file *file, const char *buffer,
         if (rc)
                 return rc;
 
+        spin_lock_bh(&obd->obd_processing_task_lock);
         obd->u.filter.fo_writethrough_cache = val;
+        spin_unlock_bh(&obd->obd_processing_task_lock);
         return count;
 }
 
+static int lprocfs_filter_rd_mds_sync(char *page, char **start, off_t off,
+                                      int count, int *eof, void *data)
+{
+        struct obd_device *obd = (struct obd_device *)data;
+        LASSERT(obd != NULL);
+
+        return snprintf(page, count, "%u\n", obd->u.filter.fo_mds_ost_sync);
+}
+
 static struct lprocfs_vars lprocfs_filter_obd_vars[] = {
         { "uuid",         lprocfs_rd_uuid,          0, 0 },
         { "blocksize",    lprocfs_rd_blksize,       0, 0 },
@@ -369,6 +382,7 @@ static struct lprocfs_vars lprocfs_filter_obd_vars[] = {
         { "read_cache_enable", lprocfs_filter_rd_cache, lprocfs_filter_wr_cache, 0},
         { "writethrough_cache_enable", lprocfs_filter_rd_wcache,
                           lprocfs_filter_wr_wcache, 0},
+        { "mds_sync",     lprocfs_filter_rd_mds_sync, 0, 0},
         { 0 }
 };
 

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 74fe997..57c26b8 100755 (executable)
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -477,9 +477,21 @@ test_20b() { # bug 10480
     df -P $DIR || df -P $DIR || true    # reconnect
     wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
 
-    # FIXME just because recovery is done doesn't mean we've finished
-    # orphan cleanup.  Fake it with a sleep for now...
-    sleep 10
+    # just because recovery is done doesn't mean we've finished
+    # orphan cleanup. Wait for llogs to get synchronized.
+    echo waiting for orphan cleanup...
+    while [ true ]; do
+            local -a sync=($(do_facet ost "$LCTL get_param obdfilter.*.mds_sync" | awk -F= ' {print $2}'))
+            local con=1
+            for ((i=0; i<${#sync[@]}; i++)); do
+                    [ ${sync[$i]} -eq 0 ] && continue
+                    # there is a not finished MDS-OST synchronization
+                    con=0
+                    break;
+            done
+            [ ${con} -eq 1 ] && break
+            sleep 1
+    done
     AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
     log "before $BEFOREUSED, after $AFTERUSED"
     [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \