ceph-jj-balancer.conf

@@ -17,14 +17,12 @@ MAX_REMAPS="20"
 #Maximum number of pg movement attempts per osd. (--max-move-attempts)
 MAX_MOVE_ATTEMPTS="50"
 
+#Name of the pool that JJ should balance. The ID of this pool will be fed into (--only-poolid)
+POOL="cephfs_data"
+
 ################################
 #### Shell Script arguments ####
 ################################
 
 #Don't run placementoptimizer.py if the number of remapped pgs exceed this threshold 
 PG_THRESHOLD="1"
-
-#The name of the pool used to search for remapped pgs.
-#If left undefined, remapped pgs in the whole cluster will be considered.
-#POOL="mf1fs_data"
-

placementoptimizer.py

@@ -4668,24 +4668,6 @@ def balance(args, cluster):
                 if osd_from not in source_osds:
                     continue
 
-            ## WX BUGFIX --- Only utilize the osds with the right crushclass
-            #Iterate through valid crushclasses
-            for crushclass_name in pg_mappings.get_enabled_crushclasses():
-                osds_in_crushclass = cluster.get_crushclass_osds(crushclass_name)
-
-                #Skip balancing the OSD if it does not have the right crushclass
-                if osd_from not in osds_in_crushclass:
-                    logging.info(f"-BUGFIX: WE ARE ONLY USING CRUSHCLASS: {crushclass_name}")
-                    logging.info(f"-BUGFIX: OSD: {osd_from} HAS INVALID CRUSHCLASS - SKIPPING")
-                    skip_osd=True
-                    continue
-                else:
-                    skip_osd=False
-
-            #Comment/uncomment this if statement to see effects of bugfix
-            if skip_osd:
-               continue
-
             source_attempts += 1
 
             if source_attempts > args.max_move_attempts:

placementoptimizer_service.sh

@@ -47,9 +47,12 @@ if [ -n "$5" ]
     #Which Pool to use to search for remapped pgs.
     POOL=$5 
     POOL_ID=$(ceph df --format json | jq --arg poolname $POOL '.pools[] | select(.name == $poolname) | .id')
-    echo "Using Pool: $POOL with  pool_id: $POOL_ID"
+  else
+    echo "Must Specify which pool to balance. Exiting"
+    exit 1
 fi
 
+echo "Using Pool: $POOL with pool_id: $POOL_ID"
 echo "Using Device Class: $DEV_CLASS"
 echo "Using PG Threshold: $PG_THRESHOLD"
 echo "Using Max Remaps: $MAX_REMAPS"
@@ -81,7 +84,7 @@ wait_remapped() {
 }
 
 while [ 1 ]; do
-  wait_remapped  
-  timeout 600 ./placementoptimizer.py -v balance --max-pg-moves $MAX_REMAPS  --only-crushclass $DEV_CLASS --max-move-attempts $MAX_MOVE_ATTEMPTS | tee /tmp/balance-upmaps
+  wait_remapped
+  timeout 600 ./placementoptimizer.py -v balance --max-pg-moves $MAX_REMAPS  --only-crushclass $DEV_CLASS --max-move-attempts $MAX_MOVE_ATTEMPTS --only-poolid $POOL_ID| tee /tmp/balance-upmaps
   cat /tmp/balance-upmaps | /bin/bash
 done