xenserver: Renice netback process to priority 0 by default.

Under heavy VM network load, we have observed that ovs-vswitchd can be
starved for CPU time, which prevents flows from being set up.  This can
in turn cause connections to XAPI in Dom0 to time out (among other issues).

It is probably not necessary to renice netback all the way to priority 0
as done in this commit.  That is simply the value that we have tested.  QA
has not reported any ill side-effects of this choice of value (yet).  One
reasonable alternative, should any problems be noticed, would be to leave
netback at its default -5 priority and simply boost ovs-vswitchd's priority
to say -6 or -7.

Bug #1656.

diff --git a/xenserver/etc_init.d_vswitch b/xenserver/etc_init.d_vswitch
index 83d84ec2a9b960102dd06d94d6e0d2d0924f1f3d..abd594ecae08605f1c5f177ae35cf2f6d206752f 100755 (executable)
--- a/xenserver/etc_init.d_vswitch
+++ b/xenserver/etc_init.d_vswitch
@@ -34,6 +34,7 @@ VSWITCHD_CONF="${VSWITCHD_CONF:-/etc/ovs-vswitchd.conf}"
 VSWITCHD_PIDFILE="${VSWITCHD_PIDFILE:-/var/run/ovs-vswitchd.pid}"
 VSWITCHD_RUN_DIR="${VSWITCHD_RUN_DIR:-/var/xen/vswitch}"
 VSWITCHD_PRIORITY="${VSWITCHD_PRIORITY:--5}"
+VSWITCHD_NETBACK_PRIORITY="${VSWITCHD_NETBACK_PRIORITY:-0}"
 VSWITCHD_LOGFILE="${VSWITCHD_LOGFILE:-/var/log/ovs-vswitchd.log}"
 VSWITCHD_FILE_LOGLEVEL="${VSWITCHD_FILE_LOGLEVEL:-}"
 VSWITCHD_SYSLOG_LOGLEVEL="${VSWITCHD_SYSLOG_LOGLEVEL:-WARN}"
@@ -115,6 +116,28 @@ function reload_vswitchd {
     fi
 }
 
+function quietly {
+    "$@" > /dev/null
+}
+
+# renice_netback PRIORITY PID [PID...]
+#
+# Renices the netback processes given as each PID to nice level PRIORITY.
+function renice_netback {
+    local pri=$1
+    shift
+
+    if test $# = 0; then
+        warning "Netback not running, nothing to renice"
+        return
+    fi
+
+    for pid
+    do
+        action "Renicing netback (pid $pid) to priority $pri" quietly renice $pri -p $pid
+    done
+}
+
 function start_vswitchd {
     local syslog_opt="-vANY:SYSLOG:${VSWITCHD_SYSLOG_LOGLEVEL}"
     local logfile_file_opt=""
@@ -152,6 +175,11 @@ function start_vswitchd {
     if [ "$ENABLE_FAKE_PROC_NET" = "y" ]; then
         fake_proc_net_opt="--fake-proc-net"
     fi
+    if [ "$VSWITCHD_NETBACK_PRIORITY" != default ]; then
+        renice_netback $VSWITCHD_NETBACK_PRIORITY $(pidof netback)
+    else
+        warning "Netback renicing disabled (see /etc/sysconfig/vswitch)"
+    fi
     if [ "$daemonize" != "y" ]; then
         # Start in background and force a "success" message
         action "Starting ovs-vswitchd ($strace_opt$valgrind_opt)" true

diff --git a/xenserver/etc_sysconfig_vswitch.example b/xenserver/etc_sysconfig_vswitch.example
index 789d61ab13950347d62e6caf3bad9625ec26634f..ea8efe16a773943c67e10e67c4443baf5479a7e5 100644 (file)
--- a/xenserver/etc_sysconfig_vswitch.example
+++ b/xenserver/etc_sysconfig_vswitch.example
@@ -43,6 +43,13 @@
 #     processes.
 # VSWITCHD_PRIORITY=-5
 
+# VSWITCHD_NETBACK_PRIORITY: "nice" priority at which to run netback.
+#     It is a good idea to run ovs-vswitchd at a higher priority than
+#     netback to ensure that it gets enough CPU time to set up flows
+#     (bug #1656).  Set this to 'default' to avoid changing netback's
+#     priority.
+# VSWITCHD_NETBACK_PRIORITY=0
+
 # VSWITCHD_LOGFILE: File to send the FILE_LOGLEVEL log messages to.
 # VSWITCHD_LOGFILE=/var/log/ovs-vswitchd.log