better handling of dependencies

Now bridges and interfaces survive a shutdown of the daemons.
Only port numbers are re-exchanged as needed on daemon restart.

This also fixes the 'make -j shutdown' bug.

Now, L/n1-n2 should mean that a link between node n1 and n2 has
been established in the db *and* it is also up (i.e., the daemons are
running).

TODO:
- there is some redundancy between start/stop and {del-,}{db,switch}s

diff --git a/planetlab/exp-tool/Makefile b/planetlab/exp-tool/Makefile
index 8f5f29c..d58e9f1 100644 (file)
--- a/planetlab/exp-tool/Makefile
+++ b/planetlab/exp-tool/Makefile
@@ -155,18 +155,19 @@ cache/db.%:
        @echo "Starting db server on $(call display,$*) - logs in $(call log,$@)"
        @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs start-db &> $(call log,$@) && touch $@
 
-cache/switch.%: cache/db.%
+cache/switch.%: | cache/db.%
        @echo "Starting vswitchd on $(call display,$*) - logs in $(call log,$@)"
        @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs start-switch &> $(call log,$@) && touch $@
 
-cache/bridge.%: cache/switch.%
+cache/bridge.%: | cache/db.%
        @echo "Creating bridge on $(call display,$*) - logs in $(call log,$@)"
-       @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs create-bridge $(BRIDGE) $(IP_$*) > $@ 2> $(call log,$@) || { rm $@; exit 1; }
-       @echo Created bridge with tap $$(cat $@) on $(call display,$*)
+       @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs create-bridge $(BRIDGE) $(IP_$*) &> $(call log,$@) \
+        && touch $@
 
 cache/controller.%: cache/bridge.%
        @echo "Setting controller $(call default,CONTROLLER,$*) on $(call display,$*) - logs in $(call log,$@)"
-       @$(SSH) $(HOST_$*) $(SUDO) ovs-vsctl set-controller $(BRIDGE) $(call default,CONTROLLER,$*) &> $(call log,$@) || { rm $@; exit 1; }
+       @$(SSH) $(HOST_$*) $(SUDO) ovs-vsctl set-controller $(BRIDGE) $(call default,CONTROLLER,$*) &> $(call log,$@) \
+        && touch $@
 
 # xxx this probably needs a more thorough cleanup in cache/
 cache/stop.%: del-bridge.%
@@ -175,42 +176,46 @@ cache/stop.%: del-bridge.%
 
 ### link-oriented targets
 # L/<nodeid>-<node_id>
-L/%: cache/link.%@1 cache/link.%@2
+L/%: cache/endpoint.%@1 cache/endpoint.%@2
        @touch $@
-       @echo "Created link $*"
+       @echo "Link $* is up"
 
 U/%: del-iface.%@1 del-iface.%@2
        @rm -f L/$*
        @echo "Deleted link $*"
 
-del-bridge.%: cache/db.%
+del-bridge.%: | cache/db.%
        @echo "Deleting bridge on $(call display,$*)"
        @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs del-bridge $(BRIDGE);
        @rm -f cache/bridge.$* \
              cache/iface.$*$(SEP)*@1 cache/iface.*$(SEP)$*@2 \
-             cache/link.$*$(SEP)*@?  cache/link.*$(SEP)$*@?  \
+             cache/port.$*$(SEP)*@1  cache/port.*$(SEP)$*@2  \
+             cache/endpoint.$*$(SEP)*@?  cache/endpoint.*$(SEP)$*@?  \
              L/$*$(SEP)*             L/*$(SEP)$*
 
-del-switch.%: del-bridge.%
+del-switch.%:
        @echo "Shutting down switch on $(call display,$*)"
        @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs stop-switch
-       @rm -f cache/switch.$* 
+       @rm -f cache/switch.$* \
+              cache/port.$*$(SEP)*@1  cache/port.*$(SEP)$*@2  \
+              cache/endpoint.$*$(SEP)*@2  cache/endpoint.*$(SEP)$*@1  \
+              L/$*$(SEP)*            L/*$(SEP)$*
 
 del-db.%:
        @echo "Shutting down db on $(call display,$*)"
        @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs stop-db
        @rm -f cache/db.$*
 
-del-controller-%:
+del-controller-%: | cache/db.%
        @echo "Removing controller for $(call display,$*)"
        @$(SSH) $(HOST_$*) $(SUDO) sliver-ovs del-controller $(BRIDGE)
        @rm -f cache/controller.$*
 
-del-links: $(addprefix U/,$(notdir $(filter-out %.log,$(wildcard L/*))))
+del-links: $(addprefix U/,$(notdir $(wildcard L/*)))
 
-del-switchs: $(addprefix del-,$(notdir $(filter-out %.log,$(wildcard cache/switch.*))))
+del-switchs: $(addprefix del-,$(notdir $(wildcard cache/switch.*)))
 
-del-dbs: $(addprefix del-,$(notdir $(filter-out %.log,$(wildcard cache/db.*))))
+del-dbs: $(addprefix del-,$(notdir $(wildcard cache/db.*)))
 
 shutdown: del-switchs del-dbs
 
@@ -253,11 +258,11 @@ cache/rsnap.links.%:
 
 .SECONDEXPANSION:
 
-del-iface.%: cache/db.$$(call get,%)
+del-iface.%: 
        @echo "Removing interface for link $(call linkpart,$*) from $(call get,$*)"
        @$(SSH) $(HOST_$(call get,$*)) \
                $(SUDO) sliver-ovs del-port L$(call linkpart,$*)
-       @rm -f cache/iface.$* cache/link.$* cache/link.$(call opp,$*)
+       @rm -f cache/iface.$* cache/port.$* cache/endpoint.$* cache/endpoint.$(call opp,$*)
 
 
 ### '%' here is leftid-rightid@{1,2}
@@ -265,25 +270,32 @@ del-iface.%: cache/db.$$(call get,%)
 #linkid=$(call linkpart,%)
 #nodeid=$(call get,%)
 #bridgefile=cache/bridge.$(nodeid)
-cache/iface.%: cache/bridge.$$(call get,%)
+cache/iface.%: cache/bridge.$$(call get,%) | cache/switch.$$(call get,%)
        @echo "Creating interface for link $(call linkpart,$(*F)) on $(call display,$(call get,$(*F))) - logs in $(call log,$@)"
        @$(SSH) $(call solve,$(call get,$(*F))) $(SUDO) sliver-ovs create-port $(BRIDGE) \
-               L$(call linkpart,$(*F)) > $@ 2> $(call log,$@) || { rm $@; exit 1; }
+               L$(call linkpart,$(*F)) &> $(call log,$@) \
+        && touch $@
+
+cache/port.%: cache/iface.% cache/switch.$$(call get,%)
+       @echo "Getting port number for link $(call linkpart,$(*F)) on $(call display,$(call get,$(*F))) - logs in $(call log,$@)"
+       @$(SSH) $(call solve,$(call get,$(*F))) $(SUDO) \
+               ovs-appctl netdev-tunnel/get-port L$(call linkpart,$(*F)) > $@ 2> $(call log,$@) \
+        || { rm $@; exit 1; }
 
 
 # linkid=$(call linkpart,%)
 # nodeid=$(call get,%)
 # iface1=cache/iface.%
 # iface2=cache/iface.$(call opp,%)
-cache/link.%: cache/host.$$(call rget,%) cache/iface.% cache/iface.$$(call opp,%)
+cache/endpoint.%: cache/host.$$(call rget,%) cache/port.$$(call opp,%) cache/iface.%
        @echo "Setting port number of link $(call linkpart,$(*F)) on $(call display,$(call get,$(*F))) - logs in $(call log,$@)"
        @$(SSH) $(call solve,$(call get,$(*F))) $(SUDO) sliver-ovs set-remote-endpoint L$(call linkpart,$(*F)) \
                        $$(cat cache/host.$(call rget,$(*F))) \
-                       $$(cat cache/iface.$(call opp,$(*F))) 2> $(call log,$@) \
+                       $$(cat cache/port.$(call opp,$(*F))) 2> $(call log,$@) \
         && touch $@
 
 ####################
-CLEANTARGETS=$(addprefix del-,$(notdir $(filter-out %.log,$(wildcard cache/bridge.*))))
+CLEANTARGETS=$(addprefix del-,$(notdir $(wildcard cache/bridge.*)))
 clean: $(CLEANTARGETS)
 distclean:
        rm -rf L log cache
@@ -315,7 +327,7 @@ $(foreach id,$(ALL_NODE_IDS), $(eval $(call node_shortcuts,$(id))))
 
 define link_shortcuts
 iface.%: cache/iface.%
-link.%: cache/link.%
+endpoint.%: cache/endpoint.%
 endef
 
 $(foreach id,$(ALL_LINK_IDS), $(eval $(call link_shortcuts,$(id))))

diff --git a/planetlab/scripts/sliver-ovs.in b/planetlab/scripts/sliver-ovs.in
index d58c1d2..0bf2fc3 100755 (executable)
--- a/planetlab/scripts/sliver-ovs.in
+++ b/planetlab/scripts/sliver-ovs.in
@@ -162,14 +162,19 @@ function create_bridge () {
     IP=${IP_PREFIX%/*}
     PREFIX=${IP_PREFIX#*/}
 
-    # ensure ovs-vswitchd is running
-    is_switch_running || { echo "ovs-vswitchd not running" >&2 ; exit 1 ; }
+    W=
+    if ! is_switch_running; then
+       # we can create the bridge even if ovs-vswitchd is not running,
+       # but we need a running ovsdb-server
+       is_db_running || { echo "ovsdb-server not running" >&2; exit 1; }
+       W="--no-wait"
+    fi
+
 
     set -e
-    ovs-vsctl --db=unix:$DB_SOCKET -- --may-exist add-br "$bridge" \
+    ovs-vsctl --db=unix:$DB_SOCKET $W -- --may-exist add-br "$bridge" \
        -- set bridge "$bridge" datapath_type=planetlab \
        -- set interface "$bridge" options:local_ip="$IP" option:local_netmask="$PREFIX"
-    ovs-appctl netdev-pltap/get-tapname "$bridge"
 }
 
 function create_port () {