path: root/tests/ovsdb-cluster.at

OVS_START_SHELL_HELPERS
# ovsdb_check_cluster N_SERVERS SCHEMA_FUNC OUTPUT TRANSACTION...
ovsdb_check_cluster () {
    local n=$1 schema_func=$2 output=$3
    shift; shift; shift

    $schema_func > schema
    schema=`ovsdb-tool schema-name schema`
    AT_CHECK([ovsdb-tool '-vPATTERN:console:%c|%p|%m' create-cluster s1.db schema unix:s1.raft], [0], [], [stderr])
    AT_CHECK([grep -v 'from ephemeral to persistent' stderr], [1])
    cid=`ovsdb-tool db-cid s1.db`
    for i in `seq 2 $n`; do
        AT_CHECK([ovsdb-tool join-cluster s$i.db $schema unix:s$i.raft unix:s1.raft])
    done

    on_exit 'kill `cat *.pid`'
    for i in `seq $n`; do
        AT_CHECK([ovsdb-server -vraft -vconsole:off -vsyslog:off --detach --no-chdir --log-file=s$i.log --pidfile=s$i.pid --unixctl=s$i --remote=punix:s$i.ovsdb s$i.db])
    done
    for i in `seq $n`; do
        AT_CHECK([ovsdb_client_wait unix:s$i.ovsdb $schema connected])
    done

    for txn
    do
      AT_CHECK([ovsdb-client --timeout=30 -vjsonrpc -vconsole:off -vsyslog:off -vvlog:off --log-file transact unix:s1.ovsdb,unix:s2.ovsdb,unix:s3.ovsdb "$txn"], [0], [stdout])
      cat stdout >> output
    done
    AT_CHECK_UNQUOTED([uuidfilt output], [0], [$output])
    for i in `seq $n`; do
        OVS_APP_EXIT_AND_WAIT_BY_TARGET([`pwd`/s$i], [s$i.pid])
    done

    AT_CHECK([ovsdb-tool check-cluster s*.db])
}
OVS_END_SHELL_HELPERS

# Test a 1-server cluster.
AT_BANNER([OVSDB - clustered transactions (1 server)])
m4_define([OVSDB_CHECK_EXECUTION],
  [AT_SETUP([$1 - cluster of 1])
   AT_KEYWORDS([ovsdb server positive unix cluster cluster1 $5])
   ovsdb_check_cluster 1 "$2" '$4' m4_foreach([txn], [$3], ['txn' ])
   AT_CLEANUP])
EXECUTION_EXAMPLES

# Test a 3-server cluster.
AT_BANNER([OVSDB - clustered transactions (3 servers)])
m4_define([OVSDB_CHECK_EXECUTION],
  [AT_SETUP([$1 - cluster of 3])
   AT_KEYWORDS([ovsdb server positive unix cluster cluster3 $5])
   ovsdb_check_cluster 3 "$2" '$4' m4_foreach([txn], [$3], ['txn' ])
   AT_CLEANUP])
EXECUTION_EXAMPLES

# Test a 5-server cluster.
AT_BANNER([OVSDB - clustered transactions (5 servers)])
m4_define([OVSDB_CHECK_EXECUTION],
  [AT_SETUP([$1 - cluster of 5])
   AT_KEYWORDS([ovsdb server positive unix cluster cluster5 $5])
   ovsdb_check_cluster 5 "$2" '$4' m4_foreach([txn], [$3], ['txn' ])
   AT_CLEANUP])
EXECUTION_EXAMPLES


OVS_START_SHELL_HELPERS
# ovsdb_cluster_failure_test SCHEMA_FUNC OUTPUT TRANSACTION...
ovsdb_cluster_failure_test () {
    # Initial state: s1 is leader, s2 and s3 are followers
    remote_1=$1
    remote_2=$2
    crash_node=$3
    crash_command=$4
    if test "$crash_node" == "1"; then
        new_leader=$5
    fi

    cp $top_srcdir/ovn/ovn-nb.ovsschema schema
    schema=`ovsdb-tool schema-name schema`
    AT_CHECK([ovsdb-tool '-vPATTERN:console:%c|%p|%m' create-cluster s1.db schema unix:s1.raft], [0], [], [dnl
ovsdb|WARN|schema: changed 2 columns in 'OVN_Northbound' database from ephemeral to persistent, including 'status' column in 'Connection' table, because clusters do not support ephemeral columns
])

    n=3
    join_cluster() {
        local i=$1
        others=
        for j in `seq 1 $n`; do
            if test $i != $j; then
                others="$others unix:s$j.raft"
            fi
        done
        AT_CHECK([ovsdb-tool join-cluster s$i.db $schema unix:s$i.raft $others])
    }
    start_server() {
        local i=$1
        printf "\ns$i: starting\n"
        AT_CHECK([ovsdb-server -vjsonrpc -vconsole:off -vsyslog:off --detach --no-chdir --log-file=s$i.log --pidfile=s$i.pid --unixctl=s$i --remote=punix:s$i.ovsdb s$i.db])
    }
    connect_server() {
        local i=$1
        printf "\ns$i: waiting to connect to storage\n"
        AT_CHECK([ovsdb_client_wait --log-file=connect$i.log unix:s$i.ovsdb $schema connected])
    }
    cid=`ovsdb-tool db-cid s1.db`
    for i in `seq 2 $n`; do join_cluster $i; done

    on_exit 'kill `cat *.pid`'
    for i in `seq $n`; do start_server $i; done
    for i in `seq $n`; do connect_server $i; done

    export OVN_NB_DB=unix:s$remote_1.ovsdb,unix:s$remote_2.ovsdb

    # To ensure $new_leader node the new leader, we delay election timer for
    # the other follower.
    if test -n "$new_leader"; then
        if test "$new_leader" == "2"; then
            delay_election_node=3
        else
            delay_election_node=2
        fi
        AT_CHECK([ovs-appctl -t "`pwd`"/s$delay_election_node cluster/failure-test delay-election], [0], [ignore])
    fi
    AT_CHECK([ovs-appctl -t "`pwd`"/s$crash_node cluster/failure-test $crash_command], [0], [ignore])
    AT_CHECK([ovn-nbctl -v --timeout=10 --no-leader-only --no-shuffle-remotes create logical_switch name=ls1], [0], [ignore], [ignore])

    # Make sure that the node really crashed.
    AT_CHECK([ls s$crash_node.ovsdb], [2], [ignore], [ignore])
    # XXX: Client will fail if remotes contains unix socket that doesn't exist (killed).
    if test "$remote_1" == "$crash_node"; then
        export OVN_NB_DB=unix:s$remote_2.ovsdb
    fi
    AT_CHECK([ovn-nbctl --no-leader-only ls-list | awk '{ print $2 }'], [0], [(ls1)
])
}
OVS_END_SHELL_HELPERS
AT_BANNER([OVSDB - cluster failure with pending transaction])

AT_SETUP([OVSDB cluster - txn on follower-2, leader crash before sending appendReq, follower-2 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 3 1 crash-before-sending-append-request 2
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, leader crash before sending appendReq, follower-3 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 3 1 crash-before-sending-append-request 3
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, leader crash before sending execRep, follower-2 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 3 1 crash-before-sending-execute-command-reply 2
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, leader crash before sending execRep, follower-3 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 3 1 crash-before-sending-execute-command-reply 3
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, leader crash after sending execRep, follower-2 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 3 1 crash-after-sending-execute-command-reply 2
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, leader crash after sending execRep, follower-3 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 3 1 crash-after-sending-execute-command-reply 3
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on leader, leader crash before sending appendReq, follower-2 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 1 2 1 crash-before-sending-append-request 2
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on leader, leader crash before sending appendReq, follower-3 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 1 2 1 crash-before-sending-append-request 3
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on leader, leader crash after sending appendReq, follower-2 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
# XXX: Detect and skip repeated transaction before enabling this test
AT_CHECK([exit 77])
ovsdb_cluster_failure_test 1 2 1 crash-after-sending-append-request 2
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on leader, leader crash after sending appendReq, follower-3 becomes leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
# XXX: Detect and skip repeated transaction before enabling this test
AT_CHECK([exit 77])
ovsdb_cluster_failure_test 1 2 1 crash-after-sending-append-request 3
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, follower-2 crash before sending execReq, reconnect to follower-3])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 3 2 crash-before-sending-execute-command-request
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, follower-2 crash before sending execReq, reconnect to leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 1 2 crash-before-sending-execute-command-request
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, follower-2 crash after sending execReq, reconnect to follower-3])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
# XXX: Detect and skip repeated transaction before enabling this test
AT_CHECK([exit 77])
ovsdb_cluster_failure_test 2 3 2 crash-after-sending-execute-command-request
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, follower-2 crash after sending execReq, reconnect to leader])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
# XXX: Detect and skip repeated transaction before enabling this test
AT_CHECK([exit 77])
ovsdb_cluster_failure_test 2 1 2 crash-after-sending-execute-command-request
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on leader, follower-2 crash after receiving appendReq for the update])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 1 1 2 crash-after-receiving-append-request-update
AT_CLEANUP

AT_SETUP([OVSDB cluster - txn on follower-2, follower-3 crash after receiving appendReq for the update])
AT_KEYWORDS([ovsdb server negative unix cluster pending-txn])
ovsdb_cluster_failure_test 2 2 3 crash-after-receiving-append-request-update
AT_CLEANUP


AT_BANNER([OVSDB - cluster tests])

# Torture test.
OVS_START_SHELL_HELPERS
ovsdb_torture_test () {
    local n=$1                  # Number of cluster members
    local victim=$2             # Cluster member to kill or remove
    local variant=$3            # 'kill' and restart or 'remove' and add
    cp $top_srcdir/ovn/ovn-sb.ovsschema schema
    schema=`ovsdb-tool schema-name schema`
    AT_CHECK([ovsdb-tool '-vPATTERN:console:%c|%p|%m' create-cluster s1.db schema unix:s1.raft], [0], [], [dnl
ovsdb|WARN|schema: changed 2 columns in 'OVN_Southbound' database from ephemeral to persistent, including 'status' column in 'Connection' table, because clusters do not support ephemeral columns
])

    join_cluster() {
        local i=$1
        others=
        for j in `seq 1 $n`; do
            if test $i != $j; then
                others="$others unix:s$j.raft"
            fi
        done
        AT_CHECK([ovsdb-tool join-cluster s$i.db $schema unix:s$i.raft $others])
    }

    start_server() {
        local i=$1
        printf "\ns$i: starting\n"
        AT_CHECK([ovsdb-server -vjsonrpc -vconsole:off -vsyslog:off --detach --no-chdir --log-file=s$i.log --pidfile=s$i.pid --unixctl=s$i --remote=punix:s$i.ovsdb s$i.db])
    }
    stop_server() {
        local i=$1
        printf "\ns$i: stopping\n"
        OVS_APP_EXIT_AND_WAIT_BY_TARGET([`pwd`/s$i], [s$i.pid])
    }
    connect_server() {
        local i=$1
        printf "\ns$i: waiting to connect to storage\n"
        AT_CHECK([ovsdb_client_wait --log-file=connect$i.log unix:s$i.ovsdb $schema connected])
    }
    remove_server() {
        local i=$1
        printf "\ns$i: removing from cluster\n"
        AT_CHECK([ovs-appctl --timeout=30 -t "`pwd`"/s$i cluster/leave OVN_Southbound])
        printf "\ns$i: waiting for removal to complete\n"
        AT_CHECK([ovsdb_client_wait --log-file=remove$i.log unix:s$i.ovsdb $schema removed])
        stop_server $i
    }
    add_server() {
        local i=$1
        rm s$i.db
        join_cluster $i
        start_server $i
        connect_server $i
    }

    cid=`ovsdb-tool db-cid s1.db`
    for i in `seq 2 $n`; do join_cluster $i; done

    on_exit 'kill `cat *.pid`'
    for i in `seq $n`; do start_server $i; done
    for i in `seq $n`; do connect_server $i; done

    OVN_SB_DB=unix:s1.ovsdb
    for i in `seq 2 $n`; do
        OVN_SB_DB=$OVN_SB_DB,unix:s$i.ovsdb
    done
    export OVN_SB_DB

    n1=10 n2=5 n3=50
    echo "starting $n1*$n2 ovn-sbctl processes..."
    for i in $(seq 0 $(expr $n1 - 1) ); do
        (for j in $(seq $n2); do
             : > $i-$j.running
             txn="add SB_Global . external_ids $i-$j=$i-$j"
             for k in $(seq $n3); do
                 txn="$txn -- add SB_Global . external_ids $i-$j-$k=$i-$j-$k"
             done
             run_as "ovn-sbctl($i-$j)" ovn-sbctl "-vPATTERN:console:ovn-sbctl($i-$j)|%D{%H:%M:%S}|%05N|%c|%p|%m" --log-file=$i-$j.log -vfile -vsyslog:off -vtimeval:off --timeout=120 --no-leader-only $txn
             status=$?
             if test $status != 0; then
                 echo "$i-$j exited with status $status" > $i-$j:$status
             fi
             rm $i-$j.running
         done
         : > $i.done)&
    done
    echo "...done"

    echo "waiting for ovn-sbctl processes to exit..."
    # Use file instead of var because code inside "while" runs in a subshell.
    echo 0 > phase
    i=0
    (while :; do echo; sleep 0.1; done) | while read REPLY; do
        printf "t=%2d s:" $i
        done=0
        for j in $(seq 0 $(expr $n1 - 1)); do
            if test -f $j.done; then
                printf " $j"
                done=$(expr $done + 1)
            fi
        done
        printf '\n'
        if test $done = $n1; then
            break
        fi

        case $(cat phase) in # (
        0)
            if test $done -ge $(expr $n1 / 10); then
                if test $variant = kill; then
                    stop_server $victim
                else
                    remove_server $victim
                fi
                echo 1 > phase
                next=$(expr $i + 2)
            fi
            ;; # (
        1)
            if test $i -ge $next; then
                if test $variant = kill; then
                    start_server $victim
                    connect_server $victim
                else
                    add_server $victim
                fi
                echo 2 > phase
            fi
            ;;
        esac

        i=$(expr $i + 1)
    done
    echo "...done"
    AT_CHECK([if test $(cat phase) != 2; then exit 77; fi])

    for i in $(seq 0 $(expr $n1 - 1) ); do
        for j in `seq $n2`; do
            echo "$i-$j=$i-$j"
            for k in `seq $n3`; do
                echo "$i-$j-$k=$i-$j-$k"
            done
        done
    done | sort > expout
    AT_CHECK([ovn-sbctl --timeout=30 --log-file=finalize.log -vtimeval:off -vfile -vsyslog:off --bare get SB_Global . external-ids | tr ',' '\n' | sed 's/[[{}"" ]]//g' | sort], [0], [expout])

    for i in `seq $n`; do
        if test $i != $victim || test $(cat phase) != 1; then
            stop_server $i
        fi
    done

    # We ignore stdout because non-fatal warnings get printed there.
    AT_CHECK([ovsdb-tool check-cluster s*.db], [0], [ignore])
}
OVS_END_SHELL_HELPERS

AT_SETUP([OVSDB 3-server torture test - kill/restart leader])
AT_KEYWORDS([ovsdb server positive unix cluster cluster3])
ovsdb_torture_test 3 1 kill
AT_CLEANUP
AT_SETUP([OVSDB 3-server torture test - kill/restart follower 1])
AT_KEYWORDS([ovsdb server positive unix cluster cluster3])
ovsdb_torture_test 3 2 kill
AT_CLEANUP
AT_SETUP([OVSDB 3-server torture test - kill/restart follower 2])
AT_KEYWORDS([ovsdb server positive unix cluster cluster3])
ovsdb_torture_test 3 3 kill
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - kill/restart leader])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 1 kill
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - kill/restart follower 1])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 2 kill
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - kill/restart follower 2])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 3 kill
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - kill/restart follower 3])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 4 kill
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - kill/restart follower 4])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 5 kill
AT_CLEANUP

AT_SETUP([OVSDB 3-server torture test - remove/re-add leader])
AT_KEYWORDS([ovsdb server positive unix cluster cluster3])
ovsdb_torture_test 3 1 remove
AT_CLEANUP
AT_SETUP([OVSDB 3-server torture test - remove/re-add follower 1])
AT_KEYWORDS([ovsdb server positive unix cluster cluster3])
ovsdb_torture_test 3 2 remove
AT_CLEANUP
AT_SETUP([OVSDB 3-server torture test - remove/re-add follower 2])
AT_KEYWORDS([ovsdb server positive unix cluster cluster3])
ovsdb_torture_test 3 3 remove
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - remove/re-add leader])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 1 remove
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - remove/re-add follower 1])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 2 remove
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - remove/re-add follower 2])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 3 remove
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - remove/re-add follower 3])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 4 remove
AT_CLEANUP
AT_SETUP([OVSDB 5-server torture test - remove/re-add follower 4])
AT_KEYWORDS([ovsdb server positive unix cluster cluster5])
ovsdb_torture_test 5 5 remove
AT_CLEANUP