refine strategy to spot ip address, keep on calling guest_ipv4

[build.git] / lbuild-nightly.sh

diff --git a/lbuild-nightly.sh b/lbuild-nightly.sh
index 9f21c36..9991abe 100755 (executable)
--- a/lbuild-nightly.sh
+++ b/lbuild-nightly.sh
@@ -11,10 +11,10 @@ COMMAND=$(basename $0)
 export PATH=$PATH:/bin:/sbin
 
 # default values, tunable with command-line options
-DEFAULT_FCDISTRO=f29
+DEFAULT_FCDISTRO=f39
 DEFAULT_PLDISTRO=lxc
 DEFAULT_PERSONALITY=linux64
-DEFAULT_MAILDEST="build at onelab.eu"
+DEFAULT_MAILDEST="thierry.parmentelat at inria.fr"
 DEFAULT_BUILD_SCM_URL="git://git.onelab.eu/build"
 DEFAULT_BASE="@DATE@--@PLDISTRO@-@FCDISTRO@-@PERSONALITY@"
 
@@ -57,13 +57,38 @@ function logfile () {
 # 1st version was relying on virsh net-dhcp-leases
 # however this was too fragile, would not work for fedora14 containers
 # WARNING: this code is duplicated in lbuild-initvm.sh
+function guest_ipv4_old() {
+    lxc=$1; shift
+
+    mac=$(virsh -c lxc:/// domiflist $lxc | grep -E 'network|bridge' | awk '{print $5;}')
+    [ -z "$mac" ] && { echo 1>&2 guest_ipv4_old cannot find mac; return 1; }
+    ip=$(arp -en | grep "$mac" | awk '{print $1;}')
+    # if not known: run a ping and try again
+    if [ -z $ip ]; then
+           ping -c1 -w1 -W1 $lxc >& /dev/null
+           ping -c1 -w1 -W1 $lxc.pl.sophia.inria.fr >& /dev/null
+           ip=$(arp -en | grep "$mac" | awk '{print $1;}')
+    fi
+    [ -z "$ip" ] && { echo 1>&2 guest_ipv4_old cannot find ip; return 1; }
+    echo $ip
+}
+
 function guest_ipv4() {
     lxc=$1; shift
 
-    mac=$(virsh -c lxc:/// domiflist $lxc | egrep 'network|bridge' | awk '{print $5;}')
-    # sanity check
-    [ -z "$mac" ] && return 0
-    arp -en | grep "$mac" | awk '{print $1;}'
+    # this gives us the libvirt_lxc pid for the container
+    local lxc_pid=$(virsh -c lxc:/// dominfo $lxc | grep '^Id:' | awk '{print $2;}' | sed -e "s|-||g")
+    [[ -z "$lxc_pid" ]] && { echo 1>&2 guest_ipv4 cannot find lxc pid; return 1; }
+    # but we need the systemd (pid=1) instance for the container
+    local systemd_pid=$(pgrep -P $lxc_pid systemd)
+    [[ -z "$systemd_pid" ]] && { echo 1>&2 guest_ipv4 cannot systemd pid; return 1; }
+    # from there we can inspect the network interfaces
+    local domip=$(nsenter -t $systemd_pid -n ip -br addr show eth0 \
+                 | awk '{print $3}' \
+                 | cut -d/ -f1 \
+                 )
+    [ -z "$domip" ] && { echo 1>&2 guest_ipv4 cannot find ip; return 1; }
+    echo $domip
 }
 
 # wrap a quick summary of suspicious stuff
@@ -135,12 +160,20 @@ EOF
 
 ### we might build on a box other than the actual web server
 # utilities for handling the pushed material (rpms, logfiles, ...)
-function webpublish_misses_dir () { ssh root@${WEBHOST}  "bash -c \"test \! -d $1\"" ; }
-function webpublish () { ssh root@${WEBHOST} "$@" ; }
-function webpublish_cp_local_to_remote () { scp $1 root@${WEBHOST}:$2 ; }
-function webpublish_cp_stdin_to_file () { ssh root@${WEBHOST} cat \> $1; }
-function webpublish_append_stdin_to_file () { ssh root@${WEBHOST} cat \>\> $1; }
-# provide remote dir as first argument, so any number of local files can be passed next
+function webpublish_misses_dir () {
+    ssh root@${WEBHOST}  "bash -c \"test \! -d $1\""
+}
+function webpublish () {
+    ssh root@${WEBHOST} "$@"
+}
+function webpublish_cp_stdin_to_file () {
+    ssh root@${WEBHOST} cat \> $1 \; chmod g+r,o+r $1
+}
+function webpublish_append_stdin_to_file () {
+    ssh root@${WEBHOST} cat \>\> $1 \; chmod g+r,o+r $1
+}
+# provide remote dir as first argument,
+# so any number of local files can be passed next
 function webpublish_rsync () {
     local remote="$1"; shift
     rsync --archive --delete $VERBOSE "$@" root@${WEBHOST}:"$remote"
@@ -168,7 +201,7 @@ function failure() {
         WEBLOG=/tmp/lbuild-early-$(date +%Y-%m-%d).log.txt
     fi
     webpublish mkdir -p $WEBBASE ||:
-    webpublish_cp_local_to_remote $LOG $WEBLOG ||:
+    webpublish_rsync $WEBLOG $LOG  ||:
     summary $LOG | webpublish_append_stdin_to_file $WEBLOG ||:
     (echo -n "============================== $COMMAND: failure at " ; date ; \
         webpublish tail --lines=1000 $WEBLOG) | \
@@ -196,7 +229,7 @@ function success () {
         WEBLOG=/tmp/lbuild-early-$(date +%Y-%m-%d).log.txt
     fi
     webpublish mkdir -p $WEBBASE
-    webpublish_cp_local_to_remote $LOG $WEBLOG
+    webpublish_rsync $WEBLOG $LOG
     summary $LOG | webpublish_append_stdin_to_file $WEBLOG
     if [ -n "$DO_TEST" ] ; then
         short_message="PASS"
@@ -325,7 +358,8 @@ function run_log () {
     ssh -n ${testmaster_ssh} rm -rf ${testdir} ${testdir}.git
 
     # check it out in the build
-    run_in_build_guest $BASE make -C /build tests-module ${MAKEVARS[@]}
+    # as well as build that might not be here esp. in a short build
+    run_in_build_guest $BASE make -C /build tests-module build-module ${MAKEVARS[@]}
 
     # push it onto the testmaster - just the 'system' subdir is enough
     rsync --verbose --archive $(rootdir $BASE)/build/MODULES/tests/system/ ${testmaster_ssh}:${BASE}