#!/bin/bash . $(dirname $0)/../../include.rc . $(dirname $0)/../../volume.rc . $(dirname $0)/../../common-utils.rc . $(dirname $0)/../../dht.rc # Test 1 overview: # ---------------- # Test whether lookups are sent after a brick comes up again # # 1. Create a 3 brick pure distribute volume # 2. Fuse mount the volume so the layout is set on the root # 3. Kill one brick and try to create a directory which hashes to that brick. # It should fail with EIO. # 4. Restart the brick that was killed. # 5. Do not remount the volume. Try to create the same directory as in step 3. cleanup TEST glusterd TEST pidof glusterd TEST $CLI volume create $V0 $H0:$B0/$V0-{1..3} TEST $CLI volume start $V0 # We want the lookup to reach DHT TEST $CLI volume set $V0 performance.stat-prefetch off # Mount using FUSE and lookup the mount so a layout is set on the brick root TEST glusterfs --entry-timeout=0 --attribute-timeout=0 -s $H0 --volfile-id $V0 $M0 ls $M0/ TEST mkdir $M0/level1 # Find a dirname that will hash to the brick we are going to kill hashed=$V0-client-1 TEST dht_first_filename_with_hashsubvol "$hashed" $M0 "dir-" roottestdir=$fn_return_val hashed=$V0-client-1 TEST dht_first_filename_with_hashsubvol "$hashed" $M0/level1 "dir-" level1testdir=$fn_return_val TEST kill_brick $V0 $H0 $B0/$V0-2 EXPECT_WITHIN $PROCESS_UP_TIMEOUT "0" brick_up_status $V0 $H0 $B0/$V0-2 TEST $CLI volume status $V0 # Unmount and mount the volume again so dht has an incomplete in memory layout umount -f $M0 TEST glusterfs --entry-timeout=0 --attribute-timeout=0 -s $H0 --volfile-id $V0 $M0 mkdir $M0/$roottestdir TEST [ $? -ne 0 ] mkdir $M0/level1/$level1testdir TEST [ $? -ne 0 ] TEST $CLI volume start $V0 force EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" brick_up_status $V0 $H0 $B0/$V0-2 #$CLI volume status # It takes a while for the client to reconnect to the brick sleep 5 mkdir $M0/$roottestdir TEST [ $? -eq 0 ] mkdir $M0/$level1/level1testdir TEST [ $? -eq 0 ] # Cleanup cleanup