diff options
Diffstat (limited to 'tests/unit/cluster')
-rw-r--r-- | tests/unit/cluster/announced-endpoints.tcl | 42 | ||||
-rw-r--r-- | tests/unit/cluster/cli.tcl | 416 | ||||
-rw-r--r-- | tests/unit/cluster/cluster-response-tls.tcl | 110 | ||||
-rw-r--r-- | tests/unit/cluster/hostnames.tcl | 203 | ||||
-rw-r--r-- | tests/unit/cluster/human-announced-nodename.tcl | 29 | ||||
-rw-r--r-- | tests/unit/cluster/links.tcl | 292 | ||||
-rw-r--r-- | tests/unit/cluster/misc.tcl | 26 | ||||
-rw-r--r-- | tests/unit/cluster/multi-slot-operations.tcl | 109 | ||||
-rw-r--r-- | tests/unit/cluster/scripting.tcl | 70 | ||||
-rw-r--r-- | tests/unit/cluster/slot-ownership.tcl | 61 |
10 files changed, 1358 insertions, 0 deletions
diff --git a/tests/unit/cluster/announced-endpoints.tcl b/tests/unit/cluster/announced-endpoints.tcl new file mode 100644 index 0000000..941a8e0 --- /dev/null +++ b/tests/unit/cluster/announced-endpoints.tcl @@ -0,0 +1,42 @@ +start_cluster 2 2 {tags {external:skip cluster}} { + + test "Test change cluster-announce-port and cluster-announce-tls-port at runtime" { + set baseport [lindex [R 0 config get port] 1] + set count [expr [llength $::servers] +1 ] + set used_port [find_available_port $baseport $count] + + R 0 config set cluster-announce-tls-port $used_port + R 0 config set cluster-announce-port $used_port + + assert_match "*:$used_port@*" [R 0 CLUSTER NODES] + wait_for_condition 50 100 { + [string match "*:$used_port@*" [R 1 CLUSTER NODES]] + } else { + fail "Cluster announced port was not propagated via gossip" + } + + R 0 config set cluster-announce-tls-port 0 + R 0 config set cluster-announce-port 0 + assert_match "*:$baseport@*" [R 0 CLUSTER NODES] + } + + test "Test change cluster-announce-bus-port at runtime" { + set baseport [lindex [R 0 config get port] 1] + set count [expr [llength $::servers] +1 ] + set used_port [find_available_port $baseport $count] + + # Verify config set cluster-announce-bus-port + R 0 config set cluster-announce-bus-port $used_port + assert_match "*@$used_port *" [R 0 CLUSTER NODES] + wait_for_condition 50 100 { + [string match "*@$used_port *" [R 1 CLUSTER NODES]] + } else { + fail "Cluster announced port was not propagated via gossip" + } + + # Verify restore default cluster-announce-port + set base_bus_port [expr $baseport + 10000] + R 0 config set cluster-announce-bus-port 0 + assert_match "*@$base_bus_port *" [R 0 CLUSTER NODES] + } +} diff --git a/tests/unit/cluster/cli.tcl b/tests/unit/cluster/cli.tcl new file mode 100644 index 0000000..76e9721 --- /dev/null +++ b/tests/unit/cluster/cli.tcl @@ -0,0 +1,416 @@ +# Primitive tests on cluster-enabled redis using redis-cli + +source tests/support/cli.tcl + +# make sure the test infra won't use SELECT +set old_singledb $::singledb +set ::singledb 1 + +# cluster creation is complicated with TLS, and the current tests don't really need that coverage +tags {tls:skip external:skip cluster} { + +# start three servers +set base_conf [list cluster-enabled yes cluster-node-timeout 1000] +start_multiple_servers 3 [list overrides $base_conf] { + + set node1 [srv 0 client] + set node2 [srv -1 client] + set node3 [srv -2 client] + set node3_pid [srv -2 pid] + set node3_rd [redis_deferring_client -2] + + test {Create 3 node cluster} { + exec src/redis-cli --cluster-yes --cluster create \ + 127.0.0.1:[srv 0 port] \ + 127.0.0.1:[srv -1 port] \ + 127.0.0.1:[srv -2 port] + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + } + + test "Run blocking command on cluster node3" { + # key9184688 is mapped to slot 10923 (first slot of node 3) + $node3_rd brpop key9184688 0 + $node3_rd flush + + wait_for_condition 50 100 { + [s -2 blocked_clients] eq {1} + } else { + fail "Client not blocked" + } + } + + test "Perform a Resharding" { + exec src/redis-cli --cluster-yes --cluster reshard 127.0.0.1:[srv -2 port] \ + --cluster-to [$node1 cluster myid] \ + --cluster-from [$node3 cluster myid] \ + --cluster-slots 1 + } + + test "Verify command got unblocked after resharding" { + # this (read) will wait for the node3 to realize the new topology + assert_error {*MOVED*} {$node3_rd read} + + # verify there are no blocked clients + assert_equal [s 0 blocked_clients] {0} + assert_equal [s -1 blocked_clients] {0} + assert_equal [s -2 blocked_clients] {0} + } + + test "Wait for cluster to be stable" { + # Cluster check just verifies the config state is self-consistent, + # waiting for cluster_state to be okay is an independent check that all the + # nodes actually believe each other are healthy, prevent cluster down error. + wait_for_condition 1000 50 { + [catch {exec src/redis-cli --cluster check 127.0.0.1:[srv 0 port]}] == 0 && + [catch {exec src/redis-cli --cluster check 127.0.0.1:[srv -1 port]}] == 0 && + [catch {exec src/redis-cli --cluster check 127.0.0.1:[srv -2 port]}] == 0 && + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + } + + set node1_rd [redis_deferring_client 0] + + test "use previous hostip in \"cluster-preferred-endpoint-type unknown-endpoint\" mode" { + + # backup and set cluster-preferred-endpoint-type unknown-endpoint + set endpoint_type_before_set [lindex [split [$node1 CONFIG GET cluster-preferred-endpoint-type] " "] 1] + $node1 CONFIG SET cluster-preferred-endpoint-type unknown-endpoint + + # when redis-cli not in cluster mode, return MOVE with empty host + set slot_for_foo [$node1 CLUSTER KEYSLOT foo] + assert_error "*MOVED $slot_for_foo :*" {$node1 set foo bar} + + # when in cluster mode, redirect using previous hostip + assert_equal "[exec src/redis-cli -h 127.0.0.1 -p [srv 0 port] -c set foo bar]" {OK} + assert_match "[exec src/redis-cli -h 127.0.0.1 -p [srv 0 port] -c get foo]" {bar} + + assert_equal [$node1 CONFIG SET cluster-preferred-endpoint-type "$endpoint_type_before_set"] {OK} + } + + test "Sanity test push cmd after resharding" { + assert_error {*MOVED*} {$node3 lpush key9184688 v1} + + $node1_rd brpop key9184688 0 + $node1_rd flush + + wait_for_condition 50 100 { + [s 0 blocked_clients] eq {1} + } else { + puts "Client not blocked" + puts "read from blocked client: [$node1_rd read]" + fail "Client not blocked" + } + + $node1 lpush key9184688 v2 + assert_equal {key9184688 v2} [$node1_rd read] + } + + $node3_rd close + + test "Run blocking command again on cluster node1" { + $node1 del key9184688 + # key9184688 is mapped to slot 10923 which has been moved to node1 + $node1_rd brpop key9184688 0 + $node1_rd flush + + wait_for_condition 50 100 { + [s 0 blocked_clients] eq {1} + } else { + fail "Client not blocked" + } + } + + test "Kill a cluster node and wait for fail state" { + # kill node3 in cluster + pause_process $node3_pid + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {fail} && + [CI 1 cluster_state] eq {fail} + } else { + fail "Cluster doesn't fail" + } + } + + test "Verify command got unblocked after cluster failure" { + assert_error {*CLUSTERDOWN*} {$node1_rd read} + + # verify there are no blocked clients + assert_equal [s 0 blocked_clients] {0} + assert_equal [s -1 blocked_clients] {0} + } + + resume_process $node3_pid + $node1_rd close + +} ;# stop servers + +# Test redis-cli -- cluster create, add-node, call. +# Test that functions are propagated on add-node +start_multiple_servers 5 [list overrides $base_conf] { + + set node4_rd [redis_client -3] + set node5_rd [redis_client -4] + + test {Functions are added to new node on redis-cli cluster add-node} { + exec src/redis-cli --cluster-yes --cluster create \ + 127.0.0.1:[srv 0 port] \ + 127.0.0.1:[srv -1 port] \ + 127.0.0.1:[srv -2 port] + + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # upload a function to all the cluster + exec src/redis-cli --cluster-yes --cluster call 127.0.0.1:[srv 0 port] \ + FUNCTION LOAD {#!lua name=TEST + redis.register_function('test', function() return 'hello' end) + } + + # adding node to the cluster + exec src/redis-cli --cluster-yes --cluster add-node \ + 127.0.0.1:[srv -3 port] \ + 127.0.0.1:[srv 0 port] + + wait_for_cluster_size 4 + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} && + [CI 3 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # make sure 'test' function was added to the new node + assert_equal {{library_name TEST engine LUA functions {{name test description {} flags {}}}}} [$node4_rd FUNCTION LIST] + + # add function to node 5 + assert_equal {TEST} [$node5_rd FUNCTION LOAD {#!lua name=TEST + redis.register_function('test', function() return 'hello' end) + }] + + # make sure functions was added to node 5 + assert_equal {{library_name TEST engine LUA functions {{name test description {} flags {}}}}} [$node5_rd FUNCTION LIST] + + # adding node 5 to the cluster should failed because it already contains the 'test' function + catch { + exec src/redis-cli --cluster-yes --cluster add-node \ + 127.0.0.1:[srv -4 port] \ + 127.0.0.1:[srv 0 port] + } e + assert_match {*node already contains functions*} $e + } +} ;# stop servers + +# Test redis-cli --cluster create, add-node. +# Test that one slot can be migrated to and then away from the new node. +test {Migrate the last slot away from a node using redis-cli} { + start_multiple_servers 4 [list overrides $base_conf] { + + # Create a cluster of 3 nodes + exec src/redis-cli --cluster-yes --cluster create \ + 127.0.0.1:[srv 0 port] \ + 127.0.0.1:[srv -1 port] \ + 127.0.0.1:[srv -2 port] + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # Insert some data + assert_equal OK [exec src/redis-cli -c -p [srv 0 port] SET foo bar] + set slot [exec src/redis-cli -c -p [srv 0 port] CLUSTER KEYSLOT foo] + + # Add new node to the cluster + exec src/redis-cli --cluster-yes --cluster add-node \ + 127.0.0.1:[srv -3 port] \ + 127.0.0.1:[srv 0 port] + + # First we wait for new node to be recognized by entire cluster + wait_for_cluster_size 4 + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} && + [CI 3 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + set newnode_r [redis_client -3] + set newnode_id [$newnode_r CLUSTER MYID] + + # Find out which node has the key "foo" by asking the new node for a + # redirect. + catch { $newnode_r get foo } e + assert_match "MOVED $slot *" $e + lassign [split [lindex $e 2] :] owner_host owner_port + set owner_r [redis $owner_host $owner_port 0 $::tls] + set owner_id [$owner_r CLUSTER MYID] + + # Move slot to new node using plain Redis commands + assert_equal OK [$newnode_r CLUSTER SETSLOT $slot IMPORTING $owner_id] + assert_equal OK [$owner_r CLUSTER SETSLOT $slot MIGRATING $newnode_id] + assert_equal {foo} [$owner_r CLUSTER GETKEYSINSLOT $slot 10] + assert_equal OK [$owner_r MIGRATE 127.0.0.1 [srv -3 port] "" 0 5000 KEYS foo] + assert_equal OK [$newnode_r CLUSTER SETSLOT $slot NODE $newnode_id] + assert_equal OK [$owner_r CLUSTER SETSLOT $slot NODE $newnode_id] + + # Using --cluster check make sure we won't get `Not all slots are covered by nodes`. + # Wait for the cluster to become stable make sure the cluster is up during MIGRATE. + wait_for_condition 1000 50 { + [catch {exec src/redis-cli --cluster check 127.0.0.1:[srv 0 port]}] == 0 && + [catch {exec src/redis-cli --cluster check 127.0.0.1:[srv -1 port]}] == 0 && + [catch {exec src/redis-cli --cluster check 127.0.0.1:[srv -2 port]}] == 0 && + [catch {exec src/redis-cli --cluster check 127.0.0.1:[srv -3 port]}] == 0 && + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} && + [CI 3 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # Move the only slot back to original node using redis-cli + exec src/redis-cli --cluster reshard 127.0.0.1:[srv -3 port] \ + --cluster-from $newnode_id \ + --cluster-to $owner_id \ + --cluster-slots 1 \ + --cluster-yes + + # The empty node will become a replica of the new owner before the + # `MOVED` check, so let's wait for the cluster to become stable. + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} && + [CI 3 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # Check that the key foo has been migrated back to the original owner. + catch { $newnode_r get foo } e + assert_equal "MOVED $slot $owner_host:$owner_port" $e + + # Check that the empty node has turned itself into a replica of the new + # owner and that the new owner knows that. + wait_for_condition 1000 50 { + [string match "*slave*" [$owner_r CLUSTER REPLICAS $owner_id]] + } else { + fail "Empty node didn't turn itself into a replica." + } + } +} + +# Test redis-cli --cluster create, add-node with cluster-port. +# Create five nodes, three with custom cluster_port and two with default values. +start_server [list overrides [list cluster-enabled yes cluster-node-timeout 1 cluster-port [find_available_port $::baseport $::portcount]]] { +start_server [list overrides [list cluster-enabled yes cluster-node-timeout 1]] { +start_server [list overrides [list cluster-enabled yes cluster-node-timeout 1 cluster-port [find_available_port $::baseport $::portcount]]] { +start_server [list overrides [list cluster-enabled yes cluster-node-timeout 1]] { +start_server [list overrides [list cluster-enabled yes cluster-node-timeout 1 cluster-port [find_available_port $::baseport $::portcount]]] { + + # The first three are used to test --cluster create. + # The last two are used to test --cluster add-node + set node1_rd [redis_client 0] + set node2_rd [redis_client -1] + set node3_rd [redis_client -2] + set node4_rd [redis_client -3] + set node5_rd [redis_client -4] + + test {redis-cli --cluster create with cluster-port} { + exec src/redis-cli --cluster-yes --cluster create \ + 127.0.0.1:[srv 0 port] \ + 127.0.0.1:[srv -1 port] \ + 127.0.0.1:[srv -2 port] + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # Make sure each node can meet other nodes + assert_equal 3 [CI 0 cluster_known_nodes] + assert_equal 3 [CI 1 cluster_known_nodes] + assert_equal 3 [CI 2 cluster_known_nodes] + } + + test {redis-cli --cluster add-node with cluster-port} { + # Adding node to the cluster (without cluster-port) + exec src/redis-cli --cluster-yes --cluster add-node \ + 127.0.0.1:[srv -3 port] \ + 127.0.0.1:[srv 0 port] + + wait_for_cluster_size 4 + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} && + [CI 3 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # Adding node to the cluster (with cluster-port) + exec src/redis-cli --cluster-yes --cluster add-node \ + 127.0.0.1:[srv -4 port] \ + 127.0.0.1:[srv 0 port] + + wait_for_cluster_size 5 + + wait_for_condition 1000 50 { + [CI 0 cluster_state] eq {ok} && + [CI 1 cluster_state] eq {ok} && + [CI 2 cluster_state] eq {ok} && + [CI 3 cluster_state] eq {ok} && + [CI 4 cluster_state] eq {ok} + } else { + fail "Cluster doesn't stabilize" + } + + # Make sure each node can meet other nodes + assert_equal 5 [CI 0 cluster_known_nodes] + assert_equal 5 [CI 1 cluster_known_nodes] + assert_equal 5 [CI 2 cluster_known_nodes] + assert_equal 5 [CI 3 cluster_known_nodes] + assert_equal 5 [CI 4 cluster_known_nodes] + } +# stop 5 servers +} +} +} +} +} + +} ;# tags + +set ::singledb $old_singledb diff --git a/tests/unit/cluster/cluster-response-tls.tcl b/tests/unit/cluster/cluster-response-tls.tcl new file mode 100644 index 0000000..a099fa7 --- /dev/null +++ b/tests/unit/cluster/cluster-response-tls.tcl @@ -0,0 +1,110 @@ +source tests/support/cluster.tcl + +proc get_port_from_moved_error {e} { + set ip_port [lindex [split $e " "] 2] + return [lindex [split $ip_port ":"] 1] +} + +proc get_pport_by_port {port} { + foreach srv $::servers { + set srv_port [dict get $srv port] + if {$port == $srv_port} { + return [dict get $srv pport] + } + } + return 0 +} + +proc get_port_from_node_info {line} { + set fields [split $line " "] + set addr [lindex $fields 1] + set ip_port [lindex [split $addr "@"] 0] + return [lindex [split $ip_port ":"] 1] +} + +proc cluster_response_tls {tls_cluster} { + + test "CLUSTER SLOTS with different connection type -- tls-cluster $tls_cluster" { + set slots1 [R 0 cluster slots] + set pport [srv 0 pport] + set cluster_client [redis_cluster 127.0.0.1:$pport 0] + set slots2 [$cluster_client cluster slots] + $cluster_client close + # Compare the ports in the first row + assert_no_match [lindex $slots1 0 2 1] [lindex $slots2 0 2 1] + } + + test "CLUSTER NODES return port according to connection type -- tls-cluster $tls_cluster" { + set nodes [R 0 cluster nodes] + set port1 [get_port_from_node_info [lindex [split $nodes "\r\n"] 0]] + set pport [srv 0 pport] + set cluster_client [redis_cluster 127.0.0.1:$pport 0] + set nodes [$cluster_client cluster nodes] + set port2 [get_port_from_node_info [lindex [split $nodes "\r\n"] 0]] + $cluster_client close + assert_not_equal $port1 $port2 + } + + set cluster [redis_cluster 127.0.0.1:[srv 0 port]] + set cluster_pport [redis_cluster 127.0.0.1:[srv 0 pport] 0] + $cluster refresh_nodes_map + + test "Set many keys in the cluster -- tls-cluster $tls_cluster" { + for {set i 0} {$i < 5000} {incr i} { + $cluster set $i $i + assert { [$cluster get $i] eq $i } + } + } + + test "Test cluster responses during migration of slot x -- tls-cluster $tls_cluster" { + set slot 10 + array set nodefrom [$cluster masternode_for_slot $slot] + array set nodeto [$cluster masternode_notfor_slot $slot] + $nodeto(link) cluster setslot $slot importing $nodefrom(id) + $nodefrom(link) cluster setslot $slot migrating $nodeto(id) + + # Get a key from that slot + set key [$nodefrom(link) cluster GETKEYSINSLOT $slot "1"] + # MOVED REPLY + catch {$nodeto(link) set $key "newVal"} e_moved1 + assert_match "*MOVED*" $e_moved1 + # ASK REPLY + catch {$nodefrom(link) set "abc{$key}" "newVal"} e_ask1 + assert_match "*ASK*" $e_ask1 + + # UNSTABLE REPLY + assert_error "*TRYAGAIN*" {$nodefrom(link) mset "a{$key}" "newVal" $key "newVal2"} + + # Connecting using another protocol + array set nodefrom_pport [$cluster_pport masternode_for_slot $slot] + array set nodeto_pport [$cluster_pport masternode_notfor_slot $slot] + + # MOVED REPLY + catch {$nodeto_pport(link) set $key "newVal"} e_moved2 + assert_match "*MOVED*" $e_moved2 + # ASK REPLY + catch {$nodefrom_pport(link) set "abc{$key}" "newVal"} e_ask2 + assert_match "*ASK*" $e_ask2 + # Compare MOVED error's port + set port1 [get_port_from_moved_error $e_moved1] + set port2 [get_port_from_moved_error $e_moved2] + assert_not_equal $port1 $port2 + assert_equal $port1 $nodefrom(port) + assert_equal $port2 [get_pport_by_port $nodefrom(port)] + # Compare ASK error's port + set port1 [get_port_from_moved_error $e_ask1] + set port2 [get_port_from_moved_error $e_ask2] + assert_not_equal $port1 $port2 + assert_equal $port1 $nodeto(port) + assert_equal $port2 [get_pport_by_port $nodeto(port)] + } +} + +if {$::tls} { + start_cluster 3 3 {tags {external:skip cluster tls} overrides {tls-cluster yes tls-replication yes}} { + cluster_response_tls yes + } + start_cluster 3 3 {tags {external:skip cluster tls} overrides {tls-cluster no tls-replication no}} { + cluster_response_tls no + } +} diff --git a/tests/unit/cluster/hostnames.tcl b/tests/unit/cluster/hostnames.tcl new file mode 100644 index 0000000..f318240 --- /dev/null +++ b/tests/unit/cluster/hostnames.tcl @@ -0,0 +1,203 @@ +proc get_slot_field {slot_output shard_id node_id attrib_id} { + return [lindex [lindex [lindex $slot_output $shard_id] $node_id] $attrib_id] +} + +# Start a cluster with 3 masters and 4 replicas. +# These tests rely on specific node ordering, so make sure no node fails over. +start_cluster 3 4 {tags {external:skip cluster} overrides {cluster-replica-no-failover yes}} { +test "Set cluster hostnames and verify they are propagated" { + for {set j 0} {$j < [llength $::servers]} {incr j} { + R $j config set cluster-announce-hostname "host-$j.com" + } + + wait_for_condition 50 100 { + [are_hostnames_propagated "host-*.com"] eq 1 + } else { + fail "cluster hostnames were not propagated" + } + + # Now that everything is propagated, assert everyone agrees + wait_for_cluster_propagation +} + +test "Update hostnames and make sure they are all eventually propagated" { + for {set j 0} {$j < [llength $::servers]} {incr j} { + R $j config set cluster-announce-hostname "host-updated-$j.com" + } + + wait_for_condition 50 100 { + [are_hostnames_propagated "host-updated-*.com"] eq 1 + } else { + fail "cluster hostnames were not propagated" + } + + # Now that everything is propagated, assert everyone agrees + wait_for_cluster_propagation +} + +test "Remove hostnames and make sure they are all eventually propagated" { + for {set j 0} {$j < [llength $::servers]} {incr j} { + R $j config set cluster-announce-hostname "" + } + + wait_for_condition 50 100 { + [are_hostnames_propagated ""] eq 1 + } else { + fail "cluster hostnames were not propagated" + } + + # Now that everything is propagated, assert everyone agrees + wait_for_cluster_propagation +} + +test "Verify cluster-preferred-endpoint-type behavior for redirects and info" { + R 0 config set cluster-announce-hostname "me.com" + R 1 config set cluster-announce-hostname "" + R 2 config set cluster-announce-hostname "them.com" + + wait_for_cluster_propagation + + # Verify default behavior + set slot_result [R 0 cluster slots] + assert_equal "" [lindex [get_slot_field $slot_result 0 2 0] 1] + assert_equal "" [lindex [get_slot_field $slot_result 2 2 0] 1] + assert_equal "hostname" [lindex [get_slot_field $slot_result 0 2 3] 0] + assert_equal "me.com" [lindex [get_slot_field $slot_result 0 2 3] 1] + assert_equal "hostname" [lindex [get_slot_field $slot_result 2 2 3] 0] + assert_equal "them.com" [lindex [get_slot_field $slot_result 2 2 3] 1] + + # Redirect will use the IP address + catch {R 0 set foo foo} redir_err + assert_match "MOVED * 127.0.0.1:*" $redir_err + + # Verify prefer hostname behavior + R 0 config set cluster-preferred-endpoint-type hostname + + set slot_result [R 0 cluster slots] + assert_equal "me.com" [get_slot_field $slot_result 0 2 0] + assert_equal "them.com" [get_slot_field $slot_result 2 2 0] + + # Redirect should use hostname + catch {R 0 set foo foo} redir_err + assert_match "MOVED * them.com:*" $redir_err + + # Redirect to an unknown hostname returns ? + catch {R 0 set barfoo bar} redir_err + assert_match "MOVED * ?:*" $redir_err + + # Verify unknown hostname behavior + R 0 config set cluster-preferred-endpoint-type unknown-endpoint + + # Verify default behavior + set slot_result [R 0 cluster slots] + assert_equal "ip" [lindex [get_slot_field $slot_result 0 2 3] 0] + assert_equal "127.0.0.1" [lindex [get_slot_field $slot_result 0 2 3] 1] + assert_equal "ip" [lindex [get_slot_field $slot_result 2 2 3] 0] + assert_equal "127.0.0.1" [lindex [get_slot_field $slot_result 2 2 3] 1] + assert_equal "ip" [lindex [get_slot_field $slot_result 1 2 3] 0] + assert_equal "127.0.0.1" [lindex [get_slot_field $slot_result 1 2 3] 1] + # Not required by the protocol, but IP comes before hostname + assert_equal "hostname" [lindex [get_slot_field $slot_result 0 2 3] 2] + assert_equal "me.com" [lindex [get_slot_field $slot_result 0 2 3] 3] + assert_equal "hostname" [lindex [get_slot_field $slot_result 2 2 3] 2] + assert_equal "them.com" [lindex [get_slot_field $slot_result 2 2 3] 3] + + # This node doesn't have a hostname + assert_equal 2 [llength [get_slot_field $slot_result 1 2 3]] + + # Redirect should use empty string + catch {R 0 set foo foo} redir_err + assert_match "MOVED * :*" $redir_err + + R 0 config set cluster-preferred-endpoint-type ip +} + +test "Verify the nodes configured with prefer hostname only show hostname for new nodes" { + # Have everyone forget node 6 and isolate it from the cluster. + isolate_node 6 + + # Set hostnames for the masters, now that the node is isolated + R 0 config set cluster-announce-hostname "shard-1.com" + R 1 config set cluster-announce-hostname "shard-2.com" + R 2 config set cluster-announce-hostname "shard-3.com" + + # Prevent Node 0 and Node 6 from properly meeting, + # they'll hang in the handshake phase. This allows us to + # test the case where we "know" about it but haven't + # successfully retrieved information about it yet. + R 0 DEBUG DROP-CLUSTER-PACKET-FILTER 0 + R 6 DEBUG DROP-CLUSTER-PACKET-FILTER 0 + + # Have a replica meet the isolated node + R 3 cluster meet 127.0.0.1 [srv -6 port] + + # Wait for the isolated node to learn about the rest of the cluster, + # which correspond to a single entry in cluster nodes. Note this + # doesn't mean the isolated node has successfully contacted each + # node. + wait_for_condition 50 100 { + [llength [split [R 6 CLUSTER NODES] "\n"]] eq [expr [llength $::servers] + 1] + } else { + fail "Isolated node didn't learn about the rest of the cluster *" + } + + # Now, we wait until the two nodes that aren't filtering packets + # to accept our isolated nodes connections. At this point they will + # start showing up in cluster slots. + wait_for_condition 50 100 { + [llength [R 6 CLUSTER SLOTS]] eq 2 + } else { + fail "Node did not learn about the 2 shards it can talk to" + } + set slot_result [R 6 CLUSTER SLOTS] + assert_equal [lindex [get_slot_field $slot_result 0 2 3] 1] "shard-2.com" + assert_equal [lindex [get_slot_field $slot_result 1 2 3] 1] "shard-3.com" + + # Also make sure we know about the isolated master, we + # just can't reach it. + set master_id [R 0 CLUSTER MYID] + assert_match "*$master_id*" [R 6 CLUSTER NODES] + + # Stop dropping cluster packets, and make sure everything + # stabilizes + R 0 DEBUG DROP-CLUSTER-PACKET-FILTER -1 + R 6 DEBUG DROP-CLUSTER-PACKET-FILTER -1 + + # This operation sometimes spikes to around 5 seconds to resolve the state, + # so it has a higher timeout. + wait_for_condition 50 500 { + [llength [R 6 CLUSTER SLOTS]] eq 3 + } else { + fail "Node did not learn about the 2 shards it can talk to" + } + set slot_result [R 6 CLUSTER SLOTS] + assert_equal [lindex [get_slot_field $slot_result 0 2 3] 1] "shard-1.com" + assert_equal [lindex [get_slot_field $slot_result 1 2 3] 1] "shard-2.com" + assert_equal [lindex [get_slot_field $slot_result 2 2 3] 1] "shard-3.com" +} + +test "Test restart will keep hostname information" { + # Set a new hostname, reboot and make sure it sticks + R 0 config set cluster-announce-hostname "restart-1.com" + + # Store the hostname in the config + R 0 config rewrite + + restart_server 0 true false + set slot_result [R 0 CLUSTER SLOTS] + assert_equal [lindex [get_slot_field $slot_result 0 2 3] 1] "restart-1.com" + + # As a sanity check, make sure everyone eventually agrees + wait_for_cluster_propagation +} + +test "Test hostname validation" { + catch {R 0 config set cluster-announce-hostname [string repeat x 256]} err + assert_match "*Hostnames must be less than 256 characters*" $err + catch {R 0 config set cluster-announce-hostname "?.com"} err + assert_match "*Hostnames may only contain alphanumeric characters, hyphens or dots*" $err + + # Note this isn't a valid hostname, but it passes our internal validation + R 0 config set cluster-announce-hostname "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789-." +} +} diff --git a/tests/unit/cluster/human-announced-nodename.tcl b/tests/unit/cluster/human-announced-nodename.tcl new file mode 100644 index 0000000..a595ca6 --- /dev/null +++ b/tests/unit/cluster/human-announced-nodename.tcl @@ -0,0 +1,29 @@ +# Check if cluster's view of human announced nodename is reported in logs +start_cluster 3 0 {tags {external:skip cluster}} { + test "Set cluster human announced nodename and let it propagate" { + for {set j 0} {$j < [llength $::servers]} {incr j} { + R $j config set cluster-announce-hostname "host-$j.com" + R $j config set cluster-announce-human-nodename "nodename-$j" + } + + # We wait for everyone to agree on the hostnames. Since they are gossiped + # the same way as nodenames, it implies everyone knows the nodenames too. + wait_for_condition 50 100 { + [are_hostnames_propagated "host-*.com"] eq 1 + } else { + fail "cluster hostnames were not propagated" + } + } + + test "Human nodenames are visible in log messages" { + # Pause instance 0, so everyone thinks it is dead + pause_process [srv 0 pid] + + # We're going to use a message we will know will be sent, node unreachable, + # since it includes the other node gossiping. + wait_for_log_messages -1 {"*Node * (nodename-2) reported node * (nodename-0) as not reachable*"} 0 20 500 + wait_for_log_messages -2 {"*Node * (nodename-1) reported node * (nodename-0) as not reachable*"} 0 20 500 + + resume_process [srv 0 pid] + } +} diff --git a/tests/unit/cluster/links.tcl b/tests/unit/cluster/links.tcl new file mode 100644 index 0000000..a202c37 --- /dev/null +++ b/tests/unit/cluster/links.tcl @@ -0,0 +1,292 @@ +proc get_links_with_peer {this_instance_id peer_nodename} { + set links [R $this_instance_id cluster links] + set links_with_peer {} + foreach l $links { + if {[dict get $l node] eq $peer_nodename} { + lappend links_with_peer $l + } + } + return $links_with_peer +} + +# Return the entry in CLUSTER LINKS output by instance identified by `this_instance_id` that +# corresponds to the link established toward a peer identified by `peer_nodename` +proc get_link_to_peer {this_instance_id peer_nodename} { + set links_with_peer [get_links_with_peer $this_instance_id $peer_nodename] + foreach l $links_with_peer { + if {[dict get $l direction] eq "to"} { + return $l + } + } + return {} +} + +# Return the entry in CLUSTER LINKS output by instance identified by `this_instance_id` that +# corresponds to the link accepted from a peer identified by `peer_nodename` +proc get_link_from_peer {this_instance_id peer_nodename} { + set links_with_peer [get_links_with_peer $this_instance_id $peer_nodename] + foreach l $links_with_peer { + if {[dict get $l direction] eq "from"} { + return $l + } + } + return {} +} + +# Reset cluster links to their original state +proc reset_links {id} { + set limit [lindex [R $id CONFIG get cluster-link-sendbuf-limit] 1] + + # Set a 1 byte limit and wait for cluster cron to run + # (executes every 100ms) and terminate links + R $id CONFIG SET cluster-link-sendbuf-limit 1 + after 150 + + # Reset limit + R $id CONFIG SET cluster-link-sendbuf-limit $limit + + # Wait until the cluster links come back up for each node + wait_for_condition 50 100 { + [number_of_links $id] == [expr [number_of_peers $id] * 2] + } else { + fail "Cluster links did not come back up" + } +} + +proc number_of_peers {id} { + expr [llength $::servers] - 1 +} + +proc number_of_links {id} { + llength [R $id cluster links] +} + +proc publish_messages {server num_msgs msg_size} { + for {set i 0} {$i < $num_msgs} {incr i} { + $server PUBLISH channel [string repeat "x" $msg_size] + } +} + +start_cluster 1 2 {tags {external:skip cluster}} { + set primary_id 0 + set replica1_id 1 + + set primary [Rn $primary_id] + set replica1 [Rn $replica1_id] + + test "Broadcast message across a cluster shard while a cluster link is down" { + set replica1_node_id [$replica1 CLUSTER MYID] + + set channelname ch3 + + # subscribe on replica1 + set subscribeclient1 [redis_deferring_client -1] + $subscribeclient1 deferred 1 + $subscribeclient1 SSUBSCRIBE $channelname + $subscribeclient1 read + + # subscribe on replica2 + set subscribeclient2 [redis_deferring_client -2] + $subscribeclient2 deferred 1 + $subscribeclient2 SSUBSCRIBE $channelname + $subscribeclient2 read + + # Verify number of links with cluster stable state + assert_equal [expr [number_of_peers $primary_id]*2] [number_of_links $primary_id] + + # Disconnect the cluster between primary and replica1 and publish a message. + $primary MULTI + $primary DEBUG CLUSTERLINK KILL TO $replica1_node_id + $primary SPUBLISH $channelname hello + set res [$primary EXEC] + + # Verify no client exists on the primary to receive the published message. + assert_equal $res {OK 0} + + # Wait for all the cluster links are healthy + wait_for_condition 50 100 { + [number_of_peers $primary_id]*2 == [number_of_links $primary_id] + } else { + fail "All peer links couldn't be established" + } + + # Publish a message afterwards. + $primary SPUBLISH $channelname world + + # Verify replica1 has received only (world) / hello is lost. + assert_equal "smessage ch3 world" [$subscribeclient1 read] + + # Verify replica2 has received both messages (hello/world) + assert_equal "smessage ch3 hello" [$subscribeclient2 read] + assert_equal "smessage ch3 world" [$subscribeclient2 read] + } {} {needs:debug} +} + +start_cluster 3 0 {tags {external:skip cluster}} { + test "Each node has two links with each peer" { + for {set id 0} {$id < [llength $::servers]} {incr id} { + # Assert that from point of view of each node, there are two links for + # each peer. It might take a while for cluster to stabilize so wait up + # to 5 seconds. + wait_for_condition 50 100 { + [number_of_peers $id]*2 == [number_of_links $id] + } else { + assert_equal [expr [number_of_peers $id]*2] [number_of_links $id] + } + + set nodes [get_cluster_nodes $id] + set links [R $id cluster links] + + # For each peer there should be exactly one + # link "to" it and one link "from" it. + foreach n $nodes { + if {[cluster_has_flag $n myself]} continue + set peer [dict get $n id] + set to 0 + set from 0 + foreach l $links { + if {[dict get $l node] eq $peer} { + if {[dict get $l direction] eq "to"} { + incr to + } elseif {[dict get $l direction] eq "from"} { + incr from + } + } + } + assert {$to eq 1} + assert {$from eq 1} + } + } + } + + test {Validate cluster links format} { + set lines [R 0 cluster links] + foreach l $lines { + if {$l eq {}} continue + assert_equal [llength $l] 12 + assert_equal 1 [dict exists $l "direction"] + assert_equal 1 [dict exists $l "node"] + assert_equal 1 [dict exists $l "create-time"] + assert_equal 1 [dict exists $l "events"] + assert_equal 1 [dict exists $l "send-buffer-allocated"] + assert_equal 1 [dict exists $l "send-buffer-used"] + } + } + + set primary1_id 0 + set primary2_id 1 + + set primary1 [Rn $primary1_id] + set primary2 [Rn $primary2_id] + + test "Disconnect link when send buffer limit reached" { + # On primary1, set timeout to 1 hour so links won't get disconnected due to timeouts + set oldtimeout [lindex [$primary1 CONFIG get cluster-node-timeout] 1] + $primary1 CONFIG set cluster-node-timeout [expr 60*60*1000] + + # Get primary1's links with primary2 + set primary2_name [dict get [cluster_get_myself $primary2_id] id] + set orig_link_p1_to_p2 [get_link_to_peer $primary1_id $primary2_name] + set orig_link_p1_from_p2 [get_link_from_peer $primary1_id $primary2_name] + + # On primary1, set cluster link send buffer limit to 256KB, which is large enough to not be + # overflowed by regular gossip messages but also small enough that it doesn't take too much + # memory to overflow it. If it is set too high, Redis may get OOM killed by kernel before this + # limit is overflowed in some RAM-limited test environments. + set oldlimit [lindex [$primary1 CONFIG get cluster-link-sendbuf-limit] 1] + $primary1 CONFIG set cluster-link-sendbuf-limit [expr 256*1024] + assert {[CI $primary1_id total_cluster_links_buffer_limit_exceeded] eq 0} + + # To manufacture an ever-growing send buffer from primary1 to primary2, + # make primary2 unresponsive. + set primary2_pid [srv [expr -1*$primary2_id] pid] + pause_process $primary2_pid + + # On primary1, send 128KB Pubsub messages in a loop until the send buffer of the link from + # primary1 to primary2 exceeds buffer limit therefore be dropped. + # For the send buffer to grow, we need to first exhaust TCP send buffer of primary1 and TCP + # receive buffer of primary2 first. The sizes of these two buffers vary by OS, but 100 128KB + # messages should be sufficient. + set i 0 + wait_for_condition 100 0 { + [catch {incr i} e] == 0 && + [catch {$primary1 publish channel [prepare_value [expr 128*1024]]} e] == 0 && + [catch {after 500} e] == 0 && + [CI $primary1_id total_cluster_links_buffer_limit_exceeded] >= 1 + } else { + fail "Cluster link not freed as expected" + } + + # A new link to primary2 should have been recreated + set new_link_p1_to_p2 [get_link_to_peer $primary1_id $primary2_name] + assert {[dict get $new_link_p1_to_p2 create-time] > [dict get $orig_link_p1_to_p2 create-time]} + + # Link from primary2 should not be affected + set same_link_p1_from_p2 [get_link_from_peer $primary1_id $primary2_name] + assert {[dict get $same_link_p1_from_p2 create-time] eq [dict get $orig_link_p1_from_p2 create-time]} + + # Revive primary2 + resume_process $primary2_pid + + # Reset configs on primary1 so config changes don't leak out to other tests + $primary1 CONFIG set cluster-node-timeout $oldtimeout + $primary1 CONFIG set cluster-link-sendbuf-limit $oldlimit + + reset_links $primary1_id + } + + test "Link memory increases with publishes" { + set server_id 0 + set server [Rn $server_id] + set msg_size 10000 + set num_msgs 10 + + # Remove any sendbuf limit + $primary1 CONFIG set cluster-link-sendbuf-limit 0 + + # Publish ~100KB to one of the servers + $server MULTI + $server INFO memory + publish_messages $server $num_msgs $msg_size + $server INFO memory + set res [$server EXEC] + + set link_mem_before_pubs [getInfoProperty $res mem_cluster_links] + + # Remove the first half of the response string which contains the + # first "INFO memory" results and search for the property again + set res [string range $res [expr [string length $res] / 2] end] + set link_mem_after_pubs [getInfoProperty $res mem_cluster_links] + + # We expect the memory to have increased by more than + # the culmulative size of the publish messages + set mem_diff_floor [expr $msg_size * $num_msgs] + set mem_diff [expr $link_mem_after_pubs - $link_mem_before_pubs] + assert {$mem_diff > $mem_diff_floor} + + # Reset links to ensure no leftover data for the next test + reset_links $server_id + } + + test "Link memory resets after publish messages flush" { + set server [Rn 0] + set msg_size 100000 + set num_msgs 10 + + set link_mem_before [status $server mem_cluster_links] + + # Publish ~1MB to one of the servers + $server MULTI + publish_messages $server $num_msgs $msg_size + $server EXEC + + # Wait until the cluster link memory has returned to below the pre-publish value. + # We can't guarantee it returns to the exact same value since gossip messages + # can cause the values to fluctuate. + wait_for_condition 1000 500 { + [status $server mem_cluster_links] <= $link_mem_before + } else { + fail "Cluster link memory did not settle back to expected range" + } + } +} diff --git a/tests/unit/cluster/misc.tcl b/tests/unit/cluster/misc.tcl new file mode 100644 index 0000000..cd66697 --- /dev/null +++ b/tests/unit/cluster/misc.tcl @@ -0,0 +1,26 @@ +start_cluster 2 2 {tags {external:skip cluster}} { + test {Key lazy expires during key migration} { + R 0 DEBUG SET-ACTIVE-EXPIRE 0 + + set key_slot [R 0 CLUSTER KEYSLOT FOO] + R 0 set FOO BAR PX 10 + set src_id [R 0 CLUSTER MYID] + set trg_id [R 1 CLUSTER MYID] + R 0 CLUSTER SETSLOT $key_slot MIGRATING $trg_id + R 1 CLUSTER SETSLOT $key_slot IMPORTING $src_id + after 11 + assert_error {ASK*} {R 0 GET FOO} + R 0 ping + } {PONG} + + test "Coverage: Basic cluster commands" { + assert_equal {OK} [R 0 CLUSTER saveconfig] + + set id [R 0 CLUSTER MYID] + assert_equal {0} [R 0 CLUSTER count-failure-reports $id] + + R 0 flushall + assert_equal {OK} [R 0 CLUSTER flushslots] + } +} + diff --git a/tests/unit/cluster/multi-slot-operations.tcl b/tests/unit/cluster/multi-slot-operations.tcl new file mode 100644 index 0000000..cc7bb7a --- /dev/null +++ b/tests/unit/cluster/multi-slot-operations.tcl @@ -0,0 +1,109 @@ +# This test uses a custom slot allocation for testing +proc cluster_allocate_with_continuous_slots_local {n} { + R 0 cluster ADDSLOTSRANGE 0 3276 + R 1 cluster ADDSLOTSRANGE 3277 6552 + R 2 cluster ADDSLOTSRANGE 6553 9828 + R 3 cluster ADDSLOTSRANGE 9829 13104 + R 4 cluster ADDSLOTSRANGE 13105 16383 +} + +start_cluster 5 0 {tags {external:skip cluster}} { + +set master1 [srv 0 "client"] +set master2 [srv -1 "client"] +set master3 [srv -2 "client"] +set master4 [srv -3 "client"] +set master5 [srv -4 "client"] + +test "Continuous slots distribution" { + assert_match "* 0-3276*" [$master1 CLUSTER NODES] + assert_match "* 3277-6552*" [$master2 CLUSTER NODES] + assert_match "* 6553-9828*" [$master3 CLUSTER NODES] + assert_match "* 9829-13104*" [$master4 CLUSTER NODES] + assert_match "* 13105-16383*" [$master5 CLUSTER NODES] + assert_match "*0 3276*" [$master1 CLUSTER SLOTS] + assert_match "*3277 6552*" [$master2 CLUSTER SLOTS] + assert_match "*6553 9828*" [$master3 CLUSTER SLOTS] + assert_match "*9829 13104*" [$master4 CLUSTER SLOTS] + assert_match "*13105 16383*" [$master5 CLUSTER SLOTS] + + $master1 CLUSTER DELSLOTSRANGE 3001 3050 + assert_match "* 0-3000 3051-3276*" [$master1 CLUSTER NODES] + assert_match "*0 3000*3051 3276*" [$master1 CLUSTER SLOTS] + + $master2 CLUSTER DELSLOTSRANGE 5001 5500 + assert_match "* 3277-5000 5501-6552*" [$master2 CLUSTER NODES] + assert_match "*3277 5000*5501 6552*" [$master2 CLUSTER SLOTS] + + $master3 CLUSTER DELSLOTSRANGE 7001 7100 8001 8500 + assert_match "* 6553-7000 7101-8000 8501-9828*" [$master3 CLUSTER NODES] + assert_match "*6553 7000*7101 8000*8501 9828*" [$master3 CLUSTER SLOTS] + + $master4 CLUSTER DELSLOTSRANGE 11001 12000 12101 12200 + assert_match "* 9829-11000 12001-12100 12201-13104*" [$master4 CLUSTER NODES] + assert_match "*9829 11000*12001 12100*12201 13104*" [$master4 CLUSTER SLOTS] + + $master5 CLUSTER DELSLOTSRANGE 13501 14000 15001 16000 + assert_match "* 13105-13500 14001-15000 16001-16383*" [$master5 CLUSTER NODES] + assert_match "*13105 13500*14001 15000*16001 16383*" [$master5 CLUSTER SLOTS] +} + +test "ADDSLOTS command with several boundary conditions test suite" { + assert_error "ERR Invalid or out of range slot" {R 0 cluster ADDSLOTS 3001 aaa} + assert_error "ERR Invalid or out of range slot" {R 0 cluster ADDSLOTS 3001 -1000} + assert_error "ERR Invalid or out of range slot" {R 0 cluster ADDSLOTS 3001 30003} + + assert_error "ERR Slot 3200 is already busy" {R 0 cluster ADDSLOTS 3200} + assert_error "ERR Slot 8501 is already busy" {R 0 cluster ADDSLOTS 8501} + + assert_error "ERR Slot 3001 specified multiple times" {R 0 cluster ADDSLOTS 3001 3002 3001} +} + +test "ADDSLOTSRANGE command with several boundary conditions test suite" { + # Add multiple slots with incorrect argument number + assert_error "ERR wrong number of arguments for 'cluster|addslotsrange' command" {R 0 cluster ADDSLOTSRANGE 3001 3020 3030} + + # Add multiple slots with invalid input slot + assert_error "ERR Invalid or out of range slot" {R 0 cluster ADDSLOTSRANGE 3001 3020 3030 aaa} + assert_error "ERR Invalid or out of range slot" {R 0 cluster ADDSLOTSRANGE 3001 3020 3030 70000} + assert_error "ERR Invalid or out of range slot" {R 0 cluster ADDSLOTSRANGE 3001 3020 -1000 3030} + + # Add multiple slots when start slot number is greater than the end slot + assert_error "ERR start slot number 3030 is greater than end slot number 3025" {R 0 cluster ADDSLOTSRANGE 3001 3020 3030 3025} + + # Add multiple slots with busy slot + assert_error "ERR Slot 3200 is already busy" {R 0 cluster ADDSLOTSRANGE 3001 3020 3200 3250} + + # Add multiple slots with assigned multiple times + assert_error "ERR Slot 3001 specified multiple times" {R 0 cluster ADDSLOTSRANGE 3001 3020 3001 3020} +} + +test "DELSLOTSRANGE command with several boundary conditions test suite" { + # Delete multiple slots with incorrect argument number + assert_error "ERR wrong number of arguments for 'cluster|delslotsrange' command" {R 0 cluster DELSLOTSRANGE 1000 2000 2100} + assert_match "* 0-3000 3051-3276*" [$master1 CLUSTER NODES] + assert_match "*0 3000*3051 3276*" [$master1 CLUSTER SLOTS] + + # Delete multiple slots with invalid input slot + assert_error "ERR Invalid or out of range slot" {R 0 cluster DELSLOTSRANGE 1000 2000 2100 aaa} + assert_error "ERR Invalid or out of range slot" {R 0 cluster DELSLOTSRANGE 1000 2000 2100 70000} + assert_error "ERR Invalid or out of range slot" {R 0 cluster DELSLOTSRANGE 1000 2000 -2100 2200} + assert_match "* 0-3000 3051-3276*" [$master1 CLUSTER NODES] + assert_match "*0 3000*3051 3276*" [$master1 CLUSTER SLOTS] + + # Delete multiple slots when start slot number is greater than the end slot + assert_error "ERR start slot number 5800 is greater than end slot number 5750" {R 1 cluster DELSLOTSRANGE 5600 5700 5800 5750} + assert_match "* 3277-5000 5501-6552*" [$master2 CLUSTER NODES] + assert_match "*3277 5000*5501 6552*" [$master2 CLUSTER SLOTS] + + # Delete multiple slots with already unassigned + assert_error "ERR Slot 7001 is already unassigned" {R 2 cluster DELSLOTSRANGE 7001 7100 9000 9200} + assert_match "* 6553-7000 7101-8000 8501-9828*" [$master3 CLUSTER NODES] + assert_match "*6553 7000*7101 8000*8501 9828*" [$master3 CLUSTER SLOTS] + + # Delete multiple slots with assigned multiple times + assert_error "ERR Slot 12500 specified multiple times" {R 3 cluster DELSLOTSRANGE 12500 12600 12500 12600} + assert_match "* 9829-11000 12001-12100 12201-13104*" [$master4 CLUSTER NODES] + assert_match "*9829 11000*12001 12100*12201 13104*" [$master4 CLUSTER SLOTS] +} +} cluster_allocate_with_continuous_slots_local diff --git a/tests/unit/cluster/scripting.tcl b/tests/unit/cluster/scripting.tcl new file mode 100644 index 0000000..1ade36e --- /dev/null +++ b/tests/unit/cluster/scripting.tcl @@ -0,0 +1,70 @@ +start_cluster 1 0 {tags {external:skip cluster}} { + + test {Eval scripts with shebangs and functions default to no cross slots} { + # Test that scripts with shebang block cross slot operations + assert_error "ERR Script attempted to access keys that do not hash to the same slot*" { + r 0 eval {#!lua + redis.call('set', 'foo', 'bar') + redis.call('set', 'bar', 'foo') + return 'OK' + } 0} + + # Test the functions by default block cross slot operations + r 0 function load REPLACE {#!lua name=crossslot + local function test_cross_slot(keys, args) + redis.call('set', 'foo', 'bar') + redis.call('set', 'bar', 'foo') + return 'OK' + end + + redis.register_function('test_cross_slot', test_cross_slot)} + assert_error "ERR Script attempted to access keys that do not hash to the same slot*" {r FCALL test_cross_slot 0} + } + + test {Cross slot commands are allowed by default for eval scripts and with allow-cross-slot-keys flag} { + # Old style lua scripts are allowed to access cross slot operations + r 0 eval "redis.call('set', 'foo', 'bar'); redis.call('set', 'bar', 'foo')" 0 + + # scripts with allow-cross-slot-keys flag are allowed + r 0 eval {#!lua flags=allow-cross-slot-keys + redis.call('set', 'foo', 'bar'); redis.call('set', 'bar', 'foo') + } 0 + + # Functions with allow-cross-slot-keys flag are allowed + r 0 function load REPLACE {#!lua name=crossslot + local function test_cross_slot(keys, args) + redis.call('set', 'foo', 'bar') + redis.call('set', 'bar', 'foo') + return 'OK' + end + + redis.register_function{function_name='test_cross_slot', callback=test_cross_slot, flags={ 'allow-cross-slot-keys' }}} + r FCALL test_cross_slot 0 + } + + test {Cross slot commands are also blocked if they disagree with pre-declared keys} { + assert_error "ERR Script attempted to access keys that do not hash to the same slot*" { + r 0 eval {#!lua + redis.call('set', 'foo', 'bar') + return 'OK' + } 1 bar} + } + + test "Function no-cluster flag" { + R 0 function load {#!lua name=test + redis.register_function{function_name='f1', callback=function() return 'hello' end, flags={'no-cluster'}} + } + catch {R 0 fcall f1 0} e + assert_match {*Can not run script on cluster, 'no-cluster' flag is set*} $e + } + + test "Script no-cluster flag" { + catch { + R 0 eval {#!lua flags=no-cluster + return 1 + } 0 + } e + + assert_match {*Can not run script on cluster, 'no-cluster' flag is set*} $e + } +} diff --git a/tests/unit/cluster/slot-ownership.tcl b/tests/unit/cluster/slot-ownership.tcl new file mode 100644 index 0000000..0f3e3cc --- /dev/null +++ b/tests/unit/cluster/slot-ownership.tcl @@ -0,0 +1,61 @@ +start_cluster 2 2 {tags {external:skip cluster}} { + + test "Verify that slot ownership transfer through gossip propagates deletes to replicas" { + assert {[s -2 role] eq {slave}} + wait_for_condition 1000 50 { + [s -2 master_link_status] eq {up} + } else { + fail "Instance #2 master link status is not up" + } + + assert {[s -3 role] eq {slave}} + wait_for_condition 1000 50 { + [s -3 master_link_status] eq {up} + } else { + fail "Instance #3 master link status is not up" + } + + # Set a single key that will be used to test deletion + set key "FOO" + R 0 SET $key TEST + set key_slot [R 0 cluster keyslot $key] + set slot_keys_num [R 0 cluster countkeysinslot $key_slot] + assert {$slot_keys_num > 0} + + # Wait for replica to have the key + R 2 readonly + wait_for_condition 1000 50 { + [R 2 exists $key] eq "1" + } else { + fail "Test key was not replicated" + } + + assert_equal [R 2 cluster countkeysinslot $key_slot] $slot_keys_num + + # Assert other shards in cluster doesn't have the key + assert_equal [R 1 cluster countkeysinslot $key_slot] "0" + assert_equal [R 3 cluster countkeysinslot $key_slot] "0" + + set nodeid [R 1 cluster myid] + + R 1 cluster bumpepoch + # Move $key_slot to node 1 + assert_equal [R 1 cluster setslot $key_slot node $nodeid] "OK" + + wait_for_cluster_propagation + + # src master will delete keys in the slot + wait_for_condition 50 100 { + [R 0 cluster countkeysinslot $key_slot] eq 0 + } else { + fail "master 'countkeysinslot $key_slot' did not eq 0" + } + + # src replica will delete keys in the slot + wait_for_condition 50 100 { + [R 2 cluster countkeysinslot $key_slot] eq 0 + } else { + fail "replica 'countkeysinslot $key_slot' did not eq 0" + } + } +} |