From 0930da785e62c180cac4e3777a1a7df18565ce39 Mon Sep 17 00:00:00 2001
From: pav <pav@FreeBSD.org>
Date: Thu, 21 May 2009 16:17:15 +0000
Subject: - Update to reflect a reality on pointyhat

---
 Tools/portbuild/scripts/build             |  59 +-
 Tools/portbuild/scripts/buildproxy-client |  14 +-
 Tools/portbuild/scripts/buildscript       |   4 +
 Tools/portbuild/scripts/dopackages        |  75 +--
 Tools/portbuild/scripts/dosetupnode       |  17 +-
 Tools/portbuild/scripts/makeduds          |  11 +-
 Tools/portbuild/scripts/makerestr         |   3 +-
 Tools/portbuild/scripts/makeworld         |   2 +
 Tools/portbuild/scripts/packagebuild      | 936 ++++++++++++++++--------------
 Tools/portbuild/scripts/pdispatch         | 264 ++++-----
 Tools/portbuild/scripts/pollmachine       | 148 +++--
 Tools/portbuild/scripts/ptimeout.c        |   2 +-
 Tools/portbuild/scripts/reportload        |   4 +-
 Tools/portbuild/scripts/straslivy.py      |   2 +-
 Tools/portbuild/scripts/zbackup           |   2 +-
 Tools/portbuild/scripts/zclient           | 124 ++++
 Tools/portbuild/scripts/zsync             |  38 ++
 17 files changed, 989 insertions(+), 716 deletions(-)
 create mode 100755 Tools/portbuild/scripts/zclient
 create mode 100755 Tools/portbuild/scripts/zsync

(limited to 'Tools')

diff --git a/Tools/portbuild/scripts/build b/Tools/portbuild/scripts/build
index bfa94e004d78..ed2c33971eff 100755
--- a/Tools/portbuild/scripts/build
+++ b/Tools/portbuild/scripts/build
@@ -94,7 +94,7 @@ do_clone() {
 
     if [ -d ${newbuilddir} ]; then
 	if [ ! -f ${pbab}/builds/previous/.keep ]; then
-	    build destroy ${arch} ${branch} previous
+	    /var/portbuild/scripts/build destroy ${arch} ${branch} previous
 	fi
         rm -f ${pbab}/builds/previous
 	mv ${pbab}/builds/latest ${pbab}/builds/previous
@@ -111,13 +111,22 @@ do_portsupdate() {
     buildid=$3
     builddir=$4
     shift 4
+    if [ $# -gt 0 ]; then
+	arg=$1
+	shift
+    fi
 
     portsfs=a/portbuild/${arch}/${buildid}/ports
 
+    destroy_fs a/portbuild/${arch} ${buildid} /ports || exit 1
+
+    if [ "${arg}" = "-umount" ]; then
+	return
+    fi
+
     echo "================================================"
     echo "Reimaging ZFS ports tree on ${builddir}/ports"
     echo "================================================"
-    destroy_fs a/portbuild/${arch} ${buildid} /ports || exit 1
     
     now=$(now)
     zfs snapshot a/snap/ports@${now}
@@ -132,15 +141,24 @@ do_srcupdate() {
     buildid=$3
     builddir=$4
     shift 4
+    if [ $# -gt 0 ]; then
+	arg=$1
+	shift
+    fi
 
     srcfs=a/portbuild/${arch}/${buildid}/src
 
+    destroy_fs a/portbuild/${arch} ${buildid} /src || exit 1
+
+    if [ "${arg}" = "-umount" ]; then
+	return
+    fi
+
     echo "================================================"
     echo "Reimaging ZFS src tree on ${builddir}/src"
     echo "================================================"
     
-    destroy_fs a/portbuild/${arch} ${buildid} /src || exit 1
-    
+
     case ${branch} in
 	8|8-exp)
 	    srcbranch=HEAD
@@ -173,11 +191,8 @@ cleanup_client() {
     test -f ${pb}/${arch}/portbuild.${mach} && . ${pb}/${arch}/portbuild.${mach}
 
     # Kill off builds and clean up chroot
-    ${pb}/scripts/dosetupnode ${arch} ${branch} ${buildid} ${mach} -nocopy -queue
-    
-    if [ "${arg}" = "-full" ]; then
-	${ssh_cmd} ${client_user}@${mach} ${sudo_cmd} rm -rf ${pb}/${arch}/${branch}/builds/${buildid}/.ready ${pb}/${arch}/${branch}/builds/${buildid} /tmp/.setup-${buildid}
-    fi
+    ${pb}/scripts/dosetupnode ${arch} ${branch} ${buildid} ${mach} -nocopy -queue -full
+
     echo "Finished cleaning up ${arch}/${branch} build ID ${buildid} on ${mach}"
 
 }
@@ -211,7 +226,7 @@ do_upload() {
 test_fs() {
     local fs=$1
 
-    zfs list -Ht filesystem | awk '{print $1}' | grep -q "$fs"
+    zfs list -Ht filesystem ${fs} > /dev/null 2>&1
 
 }
 
@@ -286,22 +301,22 @@ destroy_fs() {
 	# We might have snapshots on the target filesystem, e.g. if it
 	# is both the head and tail of its clone tree.  They should be
 	# unreferenced.
-	(zfs list -H -o name | grep "^${fullfs}@" | xargs -n 1 zfs destroy) || return 1
+	# We have to grep because zfs list -H returns an error instead of
+	# a null list if no snapshots exist
+	if ! (zfs list -r -H -o name -t snapshot ${fullfs} | grep "^${fullfs}@" | xargs -n 1 zfs destroy); then
+	    return 1
+	fi
 
 	# The target filesystem should now be unreferenced
-	zfs destroy -f "${fullfs}" || return 1
-
-	# Clean up the initial snapshot(s) that were promoted onto a
-        # cloned filesystem.  It could have been propagated several
-        # times so we don't know where it ended up.  Therefore we
-        # can't match for the ${buildid} part of ${fullfs}.
-	#
-	# XXX might be doing a substring match of subfs but we can't
-	# prepend / because a null subfs will not match
+	if ! zfs destroy -f "${fullfs}"; then
+	    return 1
+	fi
 
 	# Destroy the origin snapshot, which should be unreferenced
 	if [ ! -z "${parent}" ]; then
-	    zfs destroy -f ${parent} || return 1
+	    if ! zfs destroy -f ${parent}; then
+		return 1
+	    fi
 	fi
     fi
 }	
@@ -374,7 +389,7 @@ proxy_user() {
 	eval "do_${cmd} ${arch} ${branch} ${buildid} \"${builddir}\" ${args}"
 	error=$?
     else
-	su ports-${arch} -c "build ${cmd} ${arch} ${branch} ${buildid} \"${builddir}\" ${args}"
+	su ports-${arch} -c "/var/portbuild/scripts/build ${cmd} ${arch} ${branch} ${buildid} \"${builddir}\" ${args}"
 	error=$?
     fi
 
diff --git a/Tools/portbuild/scripts/buildproxy-client b/Tools/portbuild/scripts/buildproxy-client
index e5ebcfed05cf..f8de6fe878e0 100755
--- a/Tools/portbuild/scripts/buildproxy-client
+++ b/Tools/portbuild/scripts/buildproxy-client
@@ -25,8 +25,18 @@ try:
     s.close()
 
     sys.exit(int(code))
-except:
-    raise # XXX debug
+except Exception, e:
+    print "buildproxy-client: exception:"
+    print e
+    try:
+        if code == None:
+            print "buildproxy-client: error: code was None"
+        else:
+            print "buildproxy-client: error: code was '" + code + "'"
+    except Exception, e2:
+        print "buildproxy-client: exception 2:"
+        print e2
+    raise e  # XXX debug
     sys.exit(254)
 
 
diff --git a/Tools/portbuild/scripts/buildscript b/Tools/portbuild/scripts/buildscript
index ec05c6594347..fc32d7d42b02 100755
--- a/Tools/portbuild/scripts/buildscript
+++ b/Tools/portbuild/scripts/buildscript
@@ -126,11 +126,15 @@ restr=$(make -V RESTRICTED)
 # Keep restricted distfiles in a subdirectory for extra protection
 # against leakage
 if [ ! -z "$restr" ]; then
+  echo "DISTDIR=${DISTDIR}"
   export DISTDIR=${DISTDIR}/RESTRICTED
+  echo "DISTDIR=${DISTDIR}"
+  mkdir -p ${DISTDIR}
 fi
 
 if [ $phase = 1 ]; then
 
+  # note: if you change this header, also change processonelog and processlogs2
   cd $dir || exit 1
   echo "building for: $(uname -mr)"
   echo "maintained by: $(make maintainer)"
diff --git a/Tools/portbuild/scripts/dopackages b/Tools/portbuild/scripts/dopackages
index 29e7f3398dfe..e2fc7676f37a 100755
--- a/Tools/portbuild/scripts/dopackages
+++ b/Tools/portbuild/scripts/dopackages
@@ -71,12 +71,14 @@ mailexit () {
 }
 
 srctar() {
-    tar cfCj ${builddir}/src-${buildid}.tbz ${builddir} src/
+    rm -f ${builddir}/src-2*.tbz*
+    tar cfCj ${builddir}/src-${buildid}.tbz ${builddir} src/ 2>/dev/null
     md5 ${builddir}/src-${buildid}.tbz > ${builddir}/src-${buildid}.tbz.md5
 }
 
 portstar() {
-    tar cfCj ${builddir}/ports-${buildid}.tbz ${builddir} ports/
+    rm -f ${builddir}/ports-2*.tbz*
+    tar cfCj ${builddir}/ports-${buildid}.tbz ${builddir} ports/ 2>/dev/null
     md5 ${builddir}/ports-${buildid}.tbz > ${builddir}/ports-${buildid}.tbz.md5
 }
 
@@ -204,24 +206,16 @@ dobuild() {
     arch=$2
     branch=$3
     builddir=$4
-    phase=$5
-    
-    count=0
-    for i in `cat ${pb}/${arch}/mlist`; do
-	. ${pb}/${arch}/portbuild.conf
-	test -f ${pb}/${arch}/portbuild.${i} && . ${pb}/${arch}/portbuild.${i}
-	count=$((${count}+${maxjobs}))
-    done
-    
+        
     echo "================================================"
-    echo "building packages (phase ${phase})"
+    echo "building packages"
     echo "================================================"
     echo "started at $(date)"
-    phasestart=$(date +%s)
-    make -k -j${count} quickports all > ${builddir}/make.${phase} 2>&1 </dev/null
+    start=$(date +%s)
+    /var/portbuild/evil/qmanager/packagebuild ${arch} ${branch} ${buildid} > ${builddir}/make 2>&1 < /dev/null
     echo "ended at $(date)"
-    phaseend=$(date +%s)
-    echo "phase ${phase} took $(date -u -j -r $(($phaseend - $phasestart)) | awk '{print $4}')"
+    end=$(date +%s)
+    echo "Build took $(date -u -j -r $((end - start)) | awk '{print $4}')"
     echo $(echo $(ls -1 ${builddir}/packages/All | wc -l) - 2 | bc) "packages built"
     
     echo $(wc -l ${PORTSDIR}/${INDEXFILE} | awk '{print $1}') "lines in INDEX"
@@ -230,9 +224,9 @@ dobuild() {
     echo $(echo $(du -sk ${builddir}/distfiles | awk '{print $1}') / 1024 | bc) "MB of distfiles"
     
     cd ${builddir}
-    if grep -qE '(ptimeout|pnohang): killing' make.${phase}; then
+    if grep -qE '(ptimeout|pnohang): killing' make; then
 	echo "The following port(s) timed out:"
-	grep -E '(ptimeout|pnohang): killing' make.${phase} | sed -e 's/^.*ptimeout:/ptimeout:/' -e 's/^.*pnohang:/pnohang:/'
+	grep -E '(ptimeout|pnohang): killing' make | sed -e 's/^.*ptimeout:/ptimeout:/' -e 's/^.*pnohang:/pnohang:/'
     fi
     
 }
@@ -259,7 +253,7 @@ restart=0
 cont=0
 finish=0
 nofinish=0
-dodistfiles=1
+nodistfiles=0
 fetch_orig=0
 trybroken=0
 incremental=0
@@ -300,7 +294,7 @@ while [ $# -gt 0 ]; do
 	    noplistcheck=1
 	    ;;
 	x-nodistfiles)
-	    dodistfiles=0
+	    nodistfiles=1
 	    ;;
 	x-fetch-original)
 	    fetch_orig=1
@@ -349,9 +343,8 @@ fi
 
 echo | mail -s "$(basename $0) started for ${arch}-${branch} ${buildid} at $(date)" ${mailto}
 
-if [ "$dodistfiles" = 1 ]; then
-    # XXX flip default to always collect
-    export WANT_DISTFILES=1
+if [ "$nodistfiles" = 1 ]; then
+    export NO_DISTFILES=1
 fi
 
 if [ "$noplistcheck" = 1 ]; then
@@ -384,6 +377,11 @@ df -k | grep ${buildid}
 # Set up our environment variables
 buildenv ${pb} ${arch} ${branch} ${builddir}
 
+# bomb out if build clone failed
+if [ ! -d ${builddir} ]; then
+    mailexit 1
+fi
+
 if [ "${keep}" -eq 1 ]; then
     touch ${builddir}/.keep
 fi
@@ -498,8 +496,6 @@ if [ "$skipstart" = 0 ]; then
 	cdromlist ${pb} ${arch} ${branch} ${builddir} &
     fi
     
-    ${pb}/scripts/makeparallel ${arch} ${branch} ${buildid} &
-    
     cd ${builddir}
     mv distfiles/ .distfiles~
     rm -rf .distfiles~ &
@@ -518,7 +514,7 @@ if [ "$skipstart" = 0 ]; then
     
     # Create new log directories for archival
     rm -rf ${newerrors}
-    mkdir -p ${newerrors}/old-errors
+    mkdir -p ${newerrors}
     ln -sf ${newerrors} ${builddir}/errors
     rm -rf ${newlogs}
     mkdir -p ${newlogs}
@@ -533,6 +529,10 @@ if [ "$skipstart" = 0 ]; then
     fi
     cp -p ${builddir}/duds ${newerrors}/duds
     cp -p ${builddir}/duds ${newlogs}/duds
+    if [ -f "${builddir}/duds.verbose" ]; then
+	cp -p ${builddir}/duds.verbose ${newerrors}/duds.verbose
+	cp -p ${builddir}/duds.verbose ${newlogs}/duds.verbose
+    fi
     cp -p ${builddir}/ports/${INDEXFILE} ${newerrors}/INDEX
     cp -p ${builddir}/ports/${INDEXFILE} ${newlogs}/INDEX
     
@@ -594,24 +594,7 @@ if [ "$nobuild" = 0 ]; then
 	cp duds.orig duds
     fi
     
-    cd ${builddir}/packages/All
-    ln -sf ../../Makefile .
-    
-    dobuild ${pb} ${arch} ${branch} ${builddir} 1
-    
-    ls -asFlrt ${builddir}/packages/All > ${builddir}/logs/ls-lrt
-    
-    cd ${builddir}/errors/
-    find . -name '*.log' -depth 1 | cpio -dumpl ${builddir}/errors/old-errors
-
-    # Clean up the clients
-    ${pb}/scripts/build cleanup ${arch} ${branch} ${buildid}
-
-    wait
-    echo "setting up of nodes ended at $(date)"
-    
-    cd ${builddir}/packages/All
-    dobuild ${pb} ${arch} ${branch} ${builddir} 2
+    dobuild ${pb} ${arch} ${branch} ${builddir}
     
 fi
 
@@ -666,7 +649,7 @@ if [ "$nofinish" = 0 ]; then
     ${pb}/scripts/chopindex ${builddir}/ports/${INDEXFILE} ${builddir}/packages > ${builddir}/packages/INDEX
 
     ls -asFlrt ${builddir}/packages/All > ${builddir}/logs/ls-lrt
-    cp -p ${builddir}/make.[12] ${builddir}/logs
+    cp -p ${builddir}/make ${builddir}/logs
 
     echo "================================================"
     echo "copying distfiles"
@@ -712,4 +695,4 @@ echo "all done at $(date)"
 echo "entire process took $(date -u -j -r $(($endtime - $starttime)) | awk '{print $4}')"
 echo "================================================"
 
-mailexit 0
\ No newline at end of file
+mailexit 0
diff --git a/Tools/portbuild/scripts/dosetupnode b/Tools/portbuild/scripts/dosetupnode
index 33a6044a627f..243ce2ae3664 100755
--- a/Tools/portbuild/scripts/dosetupnode
+++ b/Tools/portbuild/scripts/dosetupnode
@@ -5,8 +5,6 @@
 
 # -norsync|-nocopy : Don't copy files, just clean up builds
 #
-# -queue : update queue once we finish setting up
-#
 # -force : force file copying/extraction even if it appears it is
 #          up-to-date
 #
@@ -99,9 +97,16 @@ setup() {
     ${client_setup} post-copy ${args} || (echo "post-copy for ${node} failed"; return 1)
 
     if [ "${queue}" -eq 1 ]; then
-	lockf -k ${pb}/${arch}/queue/.lock \
-	    ${pb}/scripts/pollmachine ${arch}/${node} -queue
+	jobs=$(python /var/portbuild/evil/qmanager/qclient jobs | grep "${node}" | grep "${arch}/${branch}/${buildid} package" | awk '{print $1}' | tail +1)
+	for j in ${jobs}; do
+	    python /var/portbuild/evil/qmanager/qclient release $j
+	done
+    fi
+
+    if [ "${full}" -eq 1 ]; then
+	${ssh_cmd} ${client_user}@${node} ${sudo_cmd} rm -rf ${pb}/${arch}/${branch}/builds/${buildid}/.ready ${pb}/${arch}/${branch}/builds/${buildid} /tmp/.setup-${buildid}
     fi
+
     echo "setting up of $node ended at $(date)"
 }
 
@@ -109,6 +114,7 @@ pbab=${pb}/${arch}/${branch}
 
 norsync=0
 queue=0
+full=0
 
 while [ $# -ge 1 ]; do
     case $1 in
@@ -122,6 +128,9 @@ while [ $# -ge 1 ]; do
 	-force)
 	    force=-force
 	    ;;
+	-full)
+	    full=1
+	    ;;
     esac
     shift
 done
diff --git a/Tools/portbuild/scripts/makeduds b/Tools/portbuild/scripts/makeduds
index a087ebf69a6c..809f73821d07 100755
--- a/Tools/portbuild/scripts/makeduds
+++ b/Tools/portbuild/scripts/makeduds
@@ -40,9 +40,14 @@ export PKG_DBDIR=/nonexistentpkg
 export PORT_DBDIR=/nonexistentport
 
 cd ${PORTSDIR}
-make -j${DUDSJOBS} ignorelist ECHO_MSG=true > ${duds} || exit 1
-sort ${duds} > ${duds}.tmp
-mv -f ${duds}.tmp ${duds}
+make -j${DUDSJOBS} ignorelist-verbose ECHO_MSG=true > ${duds}.verbose 2> /dev/null || exit 1
+sort ${duds}.verbose > ${duds}.verbose.tmp
+mv -f ${duds}.verbose.tmp ${duds}.verbose
+cut -f 1 -d \| ${duds}.verbose > ${duds}
 
 cp ${duds} ${duds}.orig
 grep -Ff ${duds}.orig ${index} | cut -f 1 -d \| > ${duds}.full
+
+cat ${duds} ${duds}.full | sort | uniq -u | sed -e "s@\$@|IGNORE: dependent port@" > ${duds}.full.verbose.tmp
+cat ${duds}.verbose ${duds}.full.verbose.tmp | sort > ${duds}.full.verbose
+rm ${duds}.full.verbose.tmp
diff --git a/Tools/portbuild/scripts/makerestr b/Tools/portbuild/scripts/makerestr
index 4fe8ec944703..9da8db96edce 100755
--- a/Tools/portbuild/scripts/makerestr
+++ b/Tools/portbuild/scripts/makerestr
@@ -22,7 +22,8 @@ builddir=${pb}/${arch}/${branch}/builds/${buildid}
 
 buildenv ${pb} ${arch} ${branch} ${builddir}
 
-duds=${builddir}/duds
+# XXX MCL 20080908 test removal of this
+#duds=${builddir}/duds
 
 unset DISPLAY
 
diff --git a/Tools/portbuild/scripts/makeworld b/Tools/portbuild/scripts/makeworld
index 953f9d02a421..91845fff1e3c 100755
--- a/Tools/portbuild/scripts/makeworld
+++ b/Tools/portbuild/scripts/makeworld
@@ -23,6 +23,8 @@ builddir=${pb}/${arch}/${branch}/builds/${buildid}
 # confuses cross-builds
 
 export TARGET_ARCH=${arch}
+# Workaround needed for zfs - 20090321 erwin
+export NO_FSCHG=1
 
 case "$branch" in
     6*)
diff --git a/Tools/portbuild/scripts/packagebuild b/Tools/portbuild/scripts/packagebuild
index 9beeeb151d57..e0b73e0624b5 100755
--- a/Tools/portbuild/scripts/packagebuild
+++ b/Tools/portbuild/scripts/packagebuild
@@ -10,511 +10,571 @@
 # where the entire cluster waits for a deep part of the tree to
 # build on a small number of machines
 #
-# Other advantages are that this system is easily customizable and
-# will let us customize things like the matching policy of jobs to
-# machines.
+# We can dynamically respond to changes in build machine availability,
+# since the queue manager will block jobs that cannot be immediately
+# satisfied and will unblock us when a job slot becomes available.
+#
+# When a package build fails, it is requeued with a lower priority
+# such that it will rebuild again as soon as no "phase 1" packages
+# are available to build.  This prevents the cluster staying idle
+# until the last phase 1 package builds.
+#
+# Other advantages are that this system is easily customizable and in
+# the future will let us customize things like the matching policy of
+# jobs to machines.  For example, we could avoid dispatching multiple
+# openoffice builds to the same system.
 #
 # TODO:
-# * External queue manager
-# * Mark completed packages instead of deleting them
+# * Combine build prep stages?
+#    - initial check for file up-to-date
 # * check mtime for package staleness (cf make)
-# * Check for parent mtimes after finishing child
-
-import os, sys, threading, time, subprocess, fcntl, operator
-#from itertools import ifilter, imap
-from random import choice
-
-def parseindex(indexfile):
-
-    tmp={}
-    pkghash={}
-    for i in file(indexfile):
-        line=i.rstrip().split("|")
-        pkg = line[0]
-        tmp[pkg] = line[1:]
-
-        # XXX hash category names too
-
-        # Trick python into storing package names by reference instead of copying strings and wasting 60MB
-        pkghash[pkg] = pkg
-
-    index=dict.fromkeys(tmp.keys())
-    for pkg in tmp.iterkeys():
-        line = tmp[pkg]
-        data={'name': pkg, 'path':line[0],
-                #'prefix':line[1],
-                #'comment':line[2],
-                #'descr':line[3],
-                #'maintainer':line[4],
-                'categories':line[5], # XXX duplicates strings
-                'bdep':[pkghash[i] for i in line[6].split(None)],
-                'rdep':[pkghash[i] for i in line[7].split(None)],
-                #'www':line[8],
-                'edep':[pkghash[i] for i in line[9].split(None)],
-                'pdep':[pkghash[i] for i in line[10].split(None)],
-                'fdep':[pkghash[i] for i in line[11].split(None)],
-                'height':None}
-        if index[pkg] is None:
-            index[pkg] = data
-        else:
-            index[pkg].update(data)
-        if not index[pkg].has_key('parents'):
-            index[pkg]['parents'] = []
-
-        # XXX iter?
-        deps=set()
-        for j in ['bdep','rdep','edep','fdep','pdep']:
-            deps.update(set(index[pkg][j]))
-        index[pkg]['deps'] = [pkghash[i] for i in deps]
-
-        for j in deps:
-            # This grossness is needed to avoid a second pass through
-            # the index, because we might be about to refer to
-            # packages that have not yet been processed
-            if index[j] is not None:
-                if index[j].has_key('parents'):
-                    index[j]['parents'].append(pkghash[pkg])
-                else:
-                    index[j]['parents'] = [pkghash[pkg]]
-            else:
-                index[j] = {'parents':[pkghash[pkg]]}
+# * option to skip phase 2
 
-    return index
+from qmanagerclient import *
 
-def gettargets(index, targets):
-    """ split command line arguments into list of packages to build.  Returns set or iterable """
-    # XXX make this return the full recursive list and use this later for processing wqueue
+import os, sys, threading, time, subprocess
+#import random
+from itertools import chain
+#import gc
+from stat import *
 
-    plist = set()
-    if len(targets) == 0:
-        targets = ["all"]
-    for i in targets:
-        if i == "all":
-            plist = index.iterkeys()
-            break
-        if i.endswith("-all"):
-            cat = i.rpartition("-")[0]
-            plist.update(j for j in index.iterkeys() if cat in index[j]['categories'])
-        elif i.rstrip(".tbz") in index.iterkeys():
-            plist.update([i.rstrip(".tbz")])
+from Queue import Queue
+from heapq import *
 
-    return plist
+categories = {}
+ports = {}
 
-def heightindex(index, targets):
-    """ Initial population of height tree """
+# When a build fails we requeue it with a lower priority such that it
+# will never preempt a phase 1 build but will build when spare
+# capacity is available.
+PHASE2_BASE_PRIO=1000
 
-    for i in targets:
-        heightdown(index, i)
+# Process success quickly so other jobs are started
+SUCCESS_PRIO = -1000
 
-def heightdown(index, pkgname):
-    """
-    Recursively populate the height tree down from a given package,
-    assuming empty values on entries not yet visited
+# Failure should be a less common event :)
+FAILURE_PRIO = -900
+
+# Port status codes
+PENDING = 1 # Yet to build
+PHASE2 = 2  # Failed once
+
+class PriorityQueue(Queue):
+    """Variant of Queue that retrieves open entries in
+    priority order (lowest first).
+    Entries are typically tuples of the form:  (priority number,
+    data)
+    This class can be found at: Python-2.6a3/Lib/Queue.py
     """
+    maxsize = 0
 
-    pkg=index[pkgname]
-    if pkg['height'] is None:
-        if len(pkg['deps']) > 0:
-            max = 0
-            for i in pkg['deps']:
-                w = heightdown(index, i)
-                if w > max:
-                    max = w
-            pkg['height'] = max + 1
-        else:
-            pkg['height'] = 1
-    return pkg['height']
+    def _init(self, maxsize):
+        self.queue = []
 
-def heightup(index, pkgname):
-    """ Recalculate the height tree going upwards from a package """
+    def _qsize(self, len=len):
+        return len(self.queue)
 
-    if not index.has_key(pkgname):
-        raise KeyError
+    def _put(self, item, heappush=heappush):
+        heappush(self.queue, item)
 
-    parents=set(index[pkgname]['parents'])
+    def _get(self, heappop=heappop):
+        return heappop(self.queue) 
 
-    while len(parents) > 0:
-        # XXX use a deque?
-        pkgname = parents.pop()
-        if not index.has_key(pkgname):
-            # XXX can this happen?
-            continue
-        pkg=index[pkgname]
-        oldheight=pkg['height']
-        if oldheight is None:
-            # Parent is in our build target list
-            continue
-        if len(pkg['deps']) == 0:
-            newheight = 1
-        else:
-            newheight=max(index[j]['height'] for j in pkg['deps']) + 1
-        if newheight > oldheight:
-            print "%s height increasing: %d -> %d", pkg, oldheight, newheight
-            assert(False)
-        if newheight != oldheight:
-            pkg['height'] = newheight
-            parents.update(pkg['parents'])
-
-def deleteup(index, pkgname):
-    if not index.has_key(pkgname):
-        raise KeyError
-
-    parents=set([pkgname])
-
-    children=[]
-    removed=[]
-    while len(parents) > 0:
-        pkgname = parents.pop()
-        if not index.has_key(pkgname):
-            # Parent was already deleted via another path
-            # XXX can happen?
-            print "YYYYYYYYYYYYYYYYYYYYYY %s deleted" % pkgname
-            continue
-        if index[pkgname]['height'] is None:
-            # parent is not in our list of build targets
-            continue
-        pkg=index[pkgname]
-
-        children.extend(pkg['deps'])
-        parents.update(pkg['parents'])
-        removed.append(pkgname)
-        del index[pkgname]
-
-    removed = set(removed)
-    children = set(children)
-#    print "Removed %d packages, touching %d children" % (len(removed), len(children))
-
-    for i in children.difference(removed):
-        par=index[i]['parents']
-        index[i]['parents'] = list(set(par).difference(removed))
-
-# XXX return an iter
-def selectheights(index, level):
-    return [i for i in index.iterkeys() if index[i]['height'] == level]
-
-def rank(index, ready, sortd, max = None):
-    """ rank the list of ready packages according to those listed as
-    dependencies in successive entries of the sorted list """
-
-    input=set(ready)
-    output = []
-    count = 0
-    print "Working on depth ",
-    for i in sortd:
-        deps = set(index[i]['deps'])
-        both = deps.intersection(input)
-        if len(both) > 0:
-            print "%d " % index[i]['height'],
-            input.difference_update(both)
-            output.extend(list(both))
-            if len(input) == 0:
-                break
-        if max:
-            count+=len(both)
-            if count > max:
-                return output
-    print
-    output.extend(list(input))
+class Index(object):
 
-    return output
+    def __init__(self, indexfile):
+        self.indexfile = indexfile
 
-def jobsuccess(index, job):
+    def parse(self, targets = None):
 
-    pkg = index[job]
-    # Build succeeded
-    for i in pkg['parents']:
-        index[i]['deps'].remove(job)
+        print "[MASTER] Read index"
+        f = file(self.indexfile)
+        index = f.readlines()
+        f.close()
+        f = None
+        del f
 
-    # deps/parents tree now partially inconsistent but this is
-    # what we need to avoid counting the height of the entry
-    # we are about to remove (which would make it a NOP)
-    heightup(index, job)
+        lines=[]
+        print "[MASTER] Phase 1"
+        for i in index:
+            (name, path, prefix, comment, descr, maintainer, categories, bdep,
+             rdep, www, edep, pdep, fdep) = i.rstrip().split("|")
 
-    del index[job]
+            if targets is None or name in targets:
+                lines.append((name, bdep, rdep, edep, pdep, fdep))
 
-def jobfailure(index, job):
+                Port(name, path, "", "", "", "",
+                     categories, "")
+        index = None
+        del index
 
-    # Build failed
-    deleteup(index, job)
-    
-class worker(threading.Thread):
+        print "[MASTER] Phase 2"
+        for (name, bdep, rdep, edep, pdep, fdep) in lines:
+            ports[name].setdeps(bdep, rdep, edep, pdep, fdep)
 
-    lock = threading.Lock()
+        lines = None
+        del lines
+        print "[MASTER] Done"
 
-    # List of running threads
-    tlist = []
+def depthindex(targets):
+    """ Initial population of depth tree """
 
-    # List of running jobs
-    running = []
+    for i in targets:
+        i.depth_recursive()
 
-    # Used to signal dispatcher when we finish a job
-    event = threading.Event()
+class Port(object):
 
-    def __init__(self, mach, job, queue, arch, branch):
-        threading.Thread.__init__(self) 
-        self.job = job
-        self.mach = mach
-        self.queue = queue
-        self.arch = arch
-        self.branch = branch
+    def __init__(self, name, path, prefix, comment, descr, maintainer,
+                 cats, www):
 
-    def run(self): 
-        global index
+        __slots__ = ["name", "path", "prefix", "comment", "descr",
+                     "maintainer", "www", "bdep", "rdep", "edep", "pdep",
+                     "fdep", "alldep", "parents",  "depth", "categories"]
 
-        pkg = index[self.job]
+        self.name = name
+        self.path = path
+        self.prefix = prefix
+        self.comment = comment
+        self.descr = descr
+        self.maintainer = maintainer
+        self.www = www
+
+        # Populated later
+        self.bdep = []
+        self.rdep = []
+        self.edep = []
+        self.pdep = []
+        self.fdep = []
+
+        self.alldep = []
+        self.parents = []
+        self.id = None # XXX
+
+        self.status = PENDING
+
+        # Whether the package build has completed and is hanging around
+        # to resolve dependencies for others XXX use status
+        self.done = False
+
+        # Depth is the maximum length of the dependency chain of this port
+        self.depth = None
+
+        self.categories=[]
+        scats = cats.split()
+        if len(scats) != len(set(scats)):
+            print "[MASTER] Warning: port %s includes duplicated categories: %s" % (name, cats)
+
+        for c in set(scats):
+            try:
+                cat = categories[c]
+            except KeyError:
+                cat = Category(c)
 
-        if len(pkg['deps']) != 0:
-            print "Running job with non-empty deps: %s" % pkg
-            assert(False)
+            self.categories.append(cat)
+            cat.add(self)
 
-        print "Running job %s" % (self.job)
-        while True:
-            retcode = subprocess.call(["/usr/bin/env", "FD=%s" % " ".join(["%s.tbz" % i for i in pkg['fdep']]), "ED=%s" % " ".join(["%s.tbz" % i for i in pkg['edep']]), "PD=%s" % " ".join(["%s.tbz" % i for i in pkg['pdep']]), "BD=%s" % " ".join(["%s.tbz" % i for i in pkg['bdep']]), "RD=%s" % " ".join(["%s.tbz" % i for i in pkg['rdep']]), "/var/portbuild/scripts/pdispatch2", self.mach, self.arch, self.branch, "/var/portbuild/scripts/portbuild", "%s.tbz" % self.job, pkg['path']])
-            self.queue.release(self.mach)
-            if retcode != 254:
-                break
+        ports[name] = self
 
-            # Failed to obtain job slot
-            time.sleep(15)
-            (self.mach, dummy) = self.queue.pick()
-            print "Retrying on %s" % self.mach
+    def remove(self):
+        """ Clean ourselves up but don't touch references in other objects;
+they still need to know about us as dependencies etc """
 
-        print "Finished job %s" % self.job,
+        self.fdep = None
+        self.edep = None
+        self.pdep = None
+        self.bdep = None
+        self.rdep = None
+        self.alldep = None
+        self.parents = None
 
-        if retcode == 0:
-            status = True
-            print
-        else:
-            status = False
-            print " with status %d" % retcode
+        for cat in self.categories:
+            cat.remove(self)
 
-        worker.lock.acquire()
-        worker.running.remove(self.job)
-        worker.tlist.remove(self)
-        if status == True:
-            jobsuccess(index, self.job)
+        ports[self.name] = None
+        del ports[self.name]
+        del self
+        
+    def destroy(self):
+        """ Remove a package and all references to it """
+
+        for pkg in self.alldep:
+            if pkg.parents is not None:
+                # Already removed but not destroyed
+                try:
+                    pkg.parents.remove(self)
+                except ValueError:
+                    continue
+        
+        for pkg in self.parents:
+            try:
+                pkg.fdep.remove(self)
+            except ValueError:
+                pass
+            try:
+                pkg.edep.remove(self)
+            except ValueError:
+                pass
+            try:
+                pkg.pdep.remove(self)
+            except ValueError:
+                pass
+            try:
+                pkg.bdep.remove(self)
+            except ValueError:
+                pass
+            try:
+                pkg.rdep.remove(self)
+            except ValueError:
+                pass
+            pkg.alldep.remove(self)
+
+        sys.exc_clear()
+
+        self.remove()
+
+    def setdeps(self, bdep, rdep, edep, pdep, fdep):
+        self.fdep = [ports[p] for p in fdep.split()]
+        self.edep = [ports[p] for p in edep.split()]
+        self.pdep = [ports[p] for p in pdep.split()]
+        self.bdep = [ports[p] for p in bdep.split()]
+        self.rdep = [ports[p] for p in rdep.split()]
+
+        self.alldep = list(set(chain(self.fdep, self.edep, self.pdep,
+                                     self.bdep, self.rdep)))
+
+        for p in self.alldep:
+            p.parents.append(self)
+
+    def depth_recursive(self):
+
+        """
+        Recursively populate the depth tree up from a given package
+        through dependencies, assuming empty values on entries not yet
+        visited
+        """
+
+        if self.depth is None:
+            if len(self.parents) > 0:
+                max = 0
+                for i in self.parents:
+                    w = i.depth_recursive()
+                    if w > max:
+                        max = w
+                self.depth = max + 1
+            else:
+                self.depth = 1
+                for port in ["openoffice", "kde-3"]:
+                    if self.name.startswith(port):
+                        # Artificial boost to try and get it building earlier
+                        self.depth = 100
+        return self.depth
+
+    def destroy_recursive(self):
+        """ Remove a port and everything that depends on it """
+
+        parents=set([self])
+
+        while len(parents) > 0:
+            pkg = parents.pop()
+            assert pkg.depth is not None
+            parents.update(pkg.parents)
+            pkg.destroy()
+
+    def success(self):
+        """ Build succeeded and possibly uncovered some new leaves """
+
+        parents = self.parents[:]
+        self.done = True
+        self.remove()
+
+        newleafs = [p for p in parents if all(c.done for c in p.alldep)]
+        return newleafs
+
+    def failure(self):
+        """ Build failed """
+
+        self.destroy_recursive()
+
+    def packagename(self, arch, branch, buildid):
+        """ Return the path where a package may be found"""
+
+        return "/var/portbuild/%s/%s/builds/%s/packages/All/%s.tbz" \
+            % (arch, branch, buildid, self.name)
+
+    def is_stale(self, arch, branch, buildid):
+        """ Does a package need to be (re)-built?
+
+        Returns: False: if it exists and has newer mtime than all of
+        its dependencies.
+        True: otherwise
+        """
+
+        my_pkgname = self.packagename(arch, branch, buildid)
+        pkg_exists = os.path.exists(my_pkgname)
+
+        if pkg_exists:
+            my_mtime = os.stat(my_pkgname)[ST_MTIME]
+
+        dep_packages = [pkg.packagename(arch, branch, buildid)
+                        for pkg in self.alldep]
+        deps_exist = all(os.path.exists(pkg) for pkg in dep_packages)
+        return not (pkg_exists and deps_exist and
+                all(os.stat(pkg)[ST_MTIME] <= my_mtime
+                    for pkg in dep_packages))
+
+class Category(object):
+    def __init__(self, name):
+        self.name = name
+        self.ports = {}
+        categories[name] = self
+
+    def add(self, port):
+        self.ports[port] = port
+
+    def remove(self, port):
+        self.ports[port]=None
+        del self.ports[port]
+
+def gettargets(targets):
+    """ split command line arguments into list of packages to build.
+    Returns set or iterable of all ports that will be built including
+    dependencies """
+
+    plist = set()
+    if len(targets) == 0:
+        targets = ["all"]
+    for i in targets:
+        if i == "all":
+            return ports.itervalues()
+
+        if i.endswith("-all"):
+            cat = i.rpartition("-")[0]
+            plist.update(p.name for p in categories[cat].ports)
+        elif i.rstrip(".tbz") in ports:
+            plist.update([ports[i.rstrip(".tbz")].name])
         else:
-            jobfailure(index, self.job)
+            raise KeyError, i
 
-        # Wake up dispatcher in case it was blocked
-        worker.event.set()
-        worker.event.clear()
+    # Compute transitive closure of all dependencies
+    pleft=plist.copy()
+    while len(pleft) > 0:
+        pkg = pleft.pop()
+        new = [p.name for p in ports[pkg].alldep]
+        plist.update(new)
+        pleft.update(new)
 
-        worker.lock.release()
+    for p in set(ports.keys()).difference(plist):
+        ports[p].destroy()
 
-    @staticmethod
-    def dispatch(mach, job, queue, arch, branch):
-        worker.lock.acquire()
-        wrk = worker(mach, job, queue, arch, branch)
-        worker.tlist.append(wrk)
-        worker.lock.release()
-        wrk.start()
+    return [ports[p] for p in plist]
 
-class machqueue(object):
-    path = '';
-    fd = -1;
+class worker(threading.Thread):
 
-    # fcntl locks are per-process, so the fcntl lock acquisition will
-    # succeed if another thread already holds it.  We need the fcntl
-    # lock for external visibility between processes but also need an
-    # internal lock for protecting against out own threads.
-    ilock = threading.Lock()
+    # Protects threads
+    lock = threading.Lock()
 
-    def __init__(self, path):
-        super(machqueue, self).__init__()
-        self.path = path
-        self.fd = os.open("%s.lock" % self.path, os.O_RDWR|os.O_CREAT)
-        
-#        print "Initializing with %s %d" % (self.path, self.fd)
-
-    def lock(self):
-        print "Locking...",
-#        ret = fcntl.lockf(self.fd, fcntl.LOCK_EX)
-        self.ilock.acquire()
-        print "success"
-
-    def unlock(self):
-        print "Unlocking fd"
-        self.ilock.release()
-#        ret = fcntl.lockf(self.fd, fcntl.LOCK_UN)
-
-    def poll(self):
-        """ Return currently available machines """
-
-        mfile = file(self.path + "../mlist", "r")
-        mlist = mfile.readlines()
-        mfile.close()
-        mlist = [i.rstrip() for i in mlist] # Chop \n
-
-        list = os.listdir(self.path)
-        special = []
-        machines = []
-        for i in list:
-            if i.startswith('.'):
-                special.append(i)
-            else:
-                if i in mlist:
-                    machines.append(i)
-                else:
-                    os.unlink(self.path + i)
-
-        print "Found machines %s" % machines
-        return (machines, special)
-
-    def pick(self):
-        """ Choose a random machine from the queue """
-
-        min = 999
-        while min == 999:
-            while True:
-                self.lock()
-                (machines, special) = self.poll()
-                if len(machines):
-                    break
-                else:
-                    self.unlock()
-                    time.sleep(15)
-                    # XXX Use kqueue to monitor for changes
-
-            list = []
-            # XXX Choose as fraction of capacity
-            for i in machines:
-                f = file(self.path + i, "r")
-                out = f.readline().rstrip()
-                try:
-                    load = int(out)
-                except ValueError:
-                    print "Bad value for %s: %s" % (i, out)
-                    load = 999
-                f.close()
-                if load < min:
-                    min = load
-                    list=[]
-                if load == min:
-                    list.append(i)
-            print "(%s, %d)" % (list, load)
-
-            if min == 999:
-                print "Bad queue length for %s" % list
-                self.unlock()
-
-        machine = choice(list)
-        # XXX hook up config files
-        if min == 2:
-            # Queue full
-            os.unlink(self.path + machine)
+    # Running threads, used for collecting status
+    threads = {}
+
+    def __init__(self, mach, job, arch, branch, buildid, queue):
+        threading.Thread.__init__(self) 
+        self.machine = mach
+        self.job = job
+        self.arch = arch
+        self.branch = branch
+        self.buildid = buildid
+        self.queue = queue
+
+        self.setDaemon(True)
+
+    def run(self): 
+        pkg = self.job
+
+        print "[MASTER] Running job %s" % (pkg.name),
+        if pkg.status == PHASE2:
+            print " (phase 2)"
         else:
-            f = file(self.path + machine, "w")
-            f.write("%d\n" % (min + 1))
-            f.flush()
-            f.close()
-
-        self.unlock()
-        return (machine, special)
-
-    def release(self, mach):
-        self.lock()
-        print "Releasing %s" % mach,
-        if os.path.exists(self.path + mach):
-            f = file(self.path + mach, "r+")
-            out = f.readline().rstrip()
+            print
+        try:
+            build = subprocess.Popen(
+                ["/bin/sh", "/var/portbuild/scripts/pdispatch",
+                 self.arch, self.branch, self.buildid, self.machine,
+                 "/var/portbuild/scripts/portbuild", "%s.tbz" % pkg.name,
+                 pkg.path],
+                env={'HOME':"/root",
+                     'PATH':'/sbin:/bin:/usr/sbin:/usr/bin:/usr/games:/usr/local/sbin:/usr/local/bin:/var/portbuild/scripts',
+                     'FD':" ".join(["%s.tbz" % p.name for p in pkg.fdep]),
+                     'ED':" ".join(["%s.tbz" % p.name for p in pkg.edep]),
+                     'PD':" ".join(["%s.tbz" % p.name for p in pkg.pdep]),
+                     'BD':" ".join(["%s.tbz" % p.name for p in pkg.bdep]),
+                     'RD':" ".join(["%s.tbz" % p.name for p in pkg.rdep])},
+                stderr=subprocess.STDOUT, stdout=subprocess.PIPE, bufsize=0)
+        except OSError, e:
+            print >>sys.stderr, "[%s:%s]: Execution failed: %s" % \
+                (pkg.id, pkg.name, e)
+        while True:
             try:
-                load = int(out)
-            except ValueError:
-                print "Queue error on release of %s: %s" % (mach, out)
-                load = 3 #XXX
+                line = build.stdout.readline()
+            except:
+                print "[%s:%s]: Failed reading from build script" % \
+                    (pkg.id, pkg.name)
+                break
+            if line == "":
+                break
+            print "[%s:%s] %s" % (pkg.id, pkg.name, line.rstrip())
+
+        retcode = build.wait()
+        
+#        time.sleep(random.randint(0,60))
+#
+#        r = random.random()
+#        if r < 0.1:
+#            retcode = 1
+#        elif r < 0.15:
+#            retcode = 254
+#        else:
+#            retcode = 0
+
+        conn = QManagerClientConn(stderr = sys.stderr)
+        try:
+            (code, vars) = conn.command("release", {'id':pkg.id})
+        except RequestError, e:
+            print "[MASTER] Error releasing job %s (%s): %s" % (pkg.name, pkg.id, e.value)
+
+        if retcode == 254:
+            # Requeue soft failure at original priority
+            # XXX exponential backoff?
+            time.sleep(60)
+#            print "Requeueing %s" % pkg.id
+            self.queue.put((-pkg.depth, pkg))
+        elif retcode == 253:
+            # setting up a machine, we should immediately retry
+            self.queue.put((-pkg.depth, pkg))
+        elif retcode == 0:
+            self.queue.put((SUCCESS_PRIO, pkg))
         else:
-            f = file(self.path + mach, "w")
-            load = 3 #XXX
+            self.queue.put((FAILURE_PRIO, pkg))
 
-#        f.truncate(0)
-        f.write("%d\n" % (load - 1))
-        print "...now %d" % (load - 1)
-        f.flush()
-        f.close()
-        self.unlock()
+        # Clean up
+        worker.lock.acquire()
+        worker.threads[self]=None
+        del worker.threads[self]
+        worker.lock.release()
+
+    @staticmethod
+    def dispatch(mach, job, arch, branch, buildid, queue):
+        wrk = worker(mach, job, arch, branch, buildid, queue)
+
+        worker.lock.acquire()
+        worker.threads[wrk] = wrk
+        worker.lock.release()
+
+        wrk.start()
 
-def main(arch, branch, args):
+def main(arch, branch, buildid, args):
     global index
 
-    basedir="/var/portbuild/"+arch+"/"+branch
+    basedir="/var/portbuild/"+arch+"/"+branch+"/builds/"+buildid
     portsdir=basedir+"/ports"
-    indexfile=portsdir+"/INDEX-"+branch
-    indexfile="/var/portbuild/i386/7-exp/ports/INDEX-7"
 
-    qlen = 100
+    indexfile=portsdir+"/INDEX-"+branch[0]
 
-    q = machqueue("/var/portbuild/%s/queue/" % arch)
+    print "[MASTER] parseindex..."
+    index = Index(indexfile)
+    index.parse()
+    print "[MASTER] length = %s" % len(ports)
 
-    print "parseindex..."
-    index=parseindex(indexfile)
-    print "length = %s" % len(index)
+    print "[MASTER] Finding targets..."
+    targets = gettargets(args)
 
-    targets = gettargets(index, args)
+    print "[MASTER] Calculating depth..."
+    depthindex(targets)
 
-    print "heightindex..."
-    heightindex(index, targets)
+    print "[MASTER] Pruning duds..."
+    dudsfile=basedir+"/duds"
+    for line in file(dudsfile):
+        try:
+            dud = ports[line.rstrip()]
+        except KeyError:
+            continue
+        print "[MASTER] Skipping %s (duds)" % dud.name
+        dud.destroy_recursive()
 
-    sortd = sorted(((key, val["height"]) for (key, val) in index.iteritems() if val["height"] is not None), key=operator.itemgetter(1), reverse=True)
-    wqueue = rank(index, selectheights(index, 1), (i[0] for i in sortd), qlen)
+    queue = PriorityQueue()
+    # XXX can do this while parsing index if we prune targets/duds
+    # first
+    for pkg in ports.itervalues():
+        if len(pkg.alldep) == 0:
+            queue.put((-pkg.depth, pkg))
 
-    # Main work loop
-    while len(sortd) > 0:
-        worker.lock.acquire()
-        print "Remaining %s" % len(sortd)
-        while len(wqueue) > 0:
-            job = wqueue.pop(0)
+    # XXX check osversion, pool
+    mdl=["arch = %s" % arch]
 
-            if os.path.exists("/var/portbuild/%s/%s/packages/All/%s.tbz" % (arch, branch, job)):
-                print "Skipping %s since it already exists" % job
-                jobsuccess(index, job)
-            else:
-                worker.running.append(job) # Protect against a queue
-                                           # rebalance adding this
-                                           # back during build
-                worker.lock.release()
-                (machine, specials) = q.pick()
-                worker.dispatch(machine, job, q, arch, branch)
-                worker.lock.acquire()
-
-        if len(wqueue) == 0:
-            if len(sortd) == 0:
-                # All jobs in progress, wait for children to exit
-                break
-            print "Rebalancing queue...",
-            sortd = sorted(((key, val["height"]) for (key, val) in index.iteritems() if val["height"] is not None), key=operator.itemgetter(1), reverse=True)
-            if len(sortd) == 0:
-                break
+    # Main work loop
+    while len(ports) > 0:
+        print "[MASTER] Ports remaining=%s, Queue length=%s" % (len(ports), queue.qsize())
 
-            print sortd[0:3]
-            if sortd[0][0] == 1:
-                # Everything left is depth 1, no need to waste time rebalancing further
-                qlen = len(index)
+        if len(ports) < 10:
+            print "[MASTER] Remaining ports: %s" % ports.keys()
 
-            # Don't add too many deps at once (e.g. after we build a
-            # package like gmake), or we will switch to buildinglots
-            # of shallow packages
-            ready = [i for i in selectheights(index, 1) if i not in worker.running]
-            wqueue = rank(index, ready, (i[0] for i in sortd), qlen)[:2*qlen]
-            print "now %s (%s ready)" % (wqueue, len(ready))
+        (prio, job) = queue.get()
+        if prio == SUCCESS_PRIO:
+            print "[MASTER] Job %s succeeded" % job.name
+            for new in job.success():
+                queue.put((-new.depth, new))
+            continue
+        elif prio == FAILURE_PRIO:
+            if job.status == PHASE2:
+                print "[MASTER] Job %s failed" % job.name
+                job.failure()
+                continue
+            else:
+                # Requeue at low priority
+                print "[MASTER] Job %s failed (requeued for phase 2)" % job.name
+                job.status = PHASE2
+                queue.put((PHASE2_BASE_PRIO-job.depth, job))
+                continue
+        elif job.status == PHASE2:
+            depth = -(prio - PHASE2_BASE_PRIO)
+        else:
+            depth = -prio
+
+        print "[MASTER] Working on job %s, depth %d" % (job.name, depth)
+        if job.is_stale(arch, branch, buildid):
+            conn = QManagerClientConn(stderr = sys.stderr)
+            (code, vars) = conn.command("acquire",
+                                        {"name":job.name,
+                                         "type":"%s/%s/%s package" % \
+                                             (arch, branch, buildid),
+                                         "priority":10, "mdl":mdl})
+
+            if code[0] == "2":
+                machine=vars['machine']
+                job.id=vars['id']
+#                print "Got ID %s" % job.id
+
+                worker.dispatch(machine, job, arch, branch, buildid, queue)
+            else:
+                print "[MASTER] Error acquiring job %s: %s" % (pkg.name, code)
+        else:
+            print "[MASTER] Skipping %s since it already exists" % job.name
+            for new in job.success():
+                queue.put((-new.depth, new))
 
-        worker.lock.release()
+    print "[MASTER] Waiting for threads"
+    threads = worker.threads.copy()
 
-        if len(wqueue) == 0:
-            # Ran out of work, wait for workers to free up some more
-            print "No work to do, sleeping on workers"
-            worker.event.wait()
+    for t in threads:
+        print "[MASTER] Outstanding thread: %s" % t.job.name
 
-    for i in worker.tlist:
-        i.join()
+    for t in threads:
+        print "[MASTER] Waiting for thread %s" % t.job.name
+        t.join()
 
-    print "Finished"
+    print "[MASTER] Finished"
 
 if __name__ == "__main__":
 #    from guppy import hpy; h = hpy()
 
-    main(sys.argv[1], sys.argv[2], sys.argv[3:])
-
-#    index = parseindex("/var/portbuild/i386/7-exp/ports/INDEX-7")
-#    print index['gmake-3.81_2']
+    main(sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4:])
 
diff --git a/Tools/portbuild/scripts/pdispatch b/Tools/portbuild/scripts/pdispatch
index 82a3af9ada01..893b6ade6fa2 100755
--- a/Tools/portbuild/scripts/pdispatch
+++ b/Tools/portbuild/scripts/pdispatch
@@ -1,16 +1,16 @@
 #!/bin/sh
 #
-# pdispatch <arch> <branch> <buildid> <command> <package.tbz> [<args> ...]
+# pdispatch <arch> <branch> <buildid> <host> <command> <package.tbz> [<args> ...]
 #
-# Choose a random machine from ${buildroot}/ulist and dispatch the
-# job to it via the ptimeout script.
+# Dispatch the job to a host via the ptimeout script.
 
 pb=/var/portbuild
 arch=$1
 branch=$2
 buildid=$3
-command=$4
-shift 4
+host=$4
+command=$5
+shift 5
 
 pbab=${pb}/${arch}/${branch}
 
@@ -23,13 +23,19 @@ timeout=360000
 loglength=1000
 hdrlength=6
 
+buildid=$(resolve ${pb} ${arch} ${branch} ${buildid})
+if [ -z "${buildid}" ]; then
+    echo "Invalid build ID ${buildid}"
+    exit 1
+fi
+
 builddir=${pbab}/builds/${buildid}
 buildenv ${pb} ${arch} ${branch} ${builddir}
 
-# ssh -x doesn't work on some machines
+# XXX needed still?
 unset DISPLAY
 
-# Use HPN-SSH for performance
+# Allow override by HPN-SSH for performance
 if [ -z "${ssh_cmd}" ]; then
     ssh_cmd=ssh
 fi
@@ -39,11 +45,6 @@ fi
 
 pkgname=$(basename $1 ${PKGSUFFIX})
 
-if grep -qxF ${pkgname} ${builddir}/duds; then
-    echo "skipping ${pkgname}"
-    exit 1
-fi
-
 if [ -z "${pkgname}" ]; then
     echo "null packagename"
     exit 1
@@ -62,7 +63,7 @@ fi
 if [ "x$NOPLISTCHECK" != "x" ]; then
     flags="${flags} -noplistcheck"
 fi
-if [ "x$WANT_DISTFILES" != "x" ]; then
+if [ "x$NO_DISTFILES" = "x" ]; then
     flags="${flags} -distfiles"
 fi
 if [ "x$FETCH_ORIGINAL" != "x" ]; then
@@ -72,140 +73,123 @@ if [ "x$TRYBROKEN" != "x" ]; then
     flags="${flags} -trybroken"
 fi
 
-while `true`; do
-    host=
-    chroot=
-    while [ -z "${host}" -o -z "${chroot}" ]; do
-	chroot=
-	host=$(lockf -k ${pb}/${arch}/queue/.lock ${pb}/scripts/getmachine ${pb} ${arch} ${branch})
-        # If ulist is empty, then all build machines are busy, so try
-        # again in 15 seconds.
-	if [ -z "${host}" ]; then
-	    sleep 15
-	else
-	    . ${pb}/${arch}/portbuild.conf
-	    test -f ${pb}/${arch}/portbuild.${host} && . ${pb}/${arch}/portbuild.${host}
-	    chrootdata=$(${ssh_cmd} -a -n ${client_user}@${host} ${sudo_cmd} ${pb}/scripts/claim-chroot ${arch} ${branch} ${buildid} ${pkgname} 2>&1)
-	    if [ -z "${chrootdata}" ]; then
-		echo "Failed to claim chroot on ${host}"
-	    fi
-
-	    case "${chrootdata}" in
-		*/var/portbuild/scripts/claim-chroot*)
-		    # Error executing script, assume system is booting
-		    chrootdata="wait boot"
-		    ;;
-	    esac
-
-	    echo "Got ${chrootdata} from ${host}"
-
-	    set -- ${chrootdata}
-	    if [ $# -ge 2 ]; then
-		case $1 in
-		    chroot)
-			chroot=$2
-			;;
-		    setup)
-			echo "Setting up ${arch}/${branch} build ID ${buildid} on ${host}"
-
-			# Run in the background so we can potentially
-			# claim a slot on another machine.  In
-			# practise I think we often end up trying
-			# again on the same machine though.
-
-			# Make sure to close stdin/stderr in the child
-			# or make will hang until the child process
-			# exits
-			# XXX Revert to >&- once this is confirmed as working
-			${pb}/scripts/dosetupnode ${arch} ${branch} ${buildid} ${host} >/tmp/setupnode.$$ 2>&1 &
-			;;
-		    error)
-			echo "Error reported by ${host}: $2"
-			sleep 60
-			;;
-		    wait)
-			echo "Waiting for setup to finish"
-			sleep 60
-			;;
-		esac
-		shift 2
-	    fi
-
-	    if [ -z "${chroot}" ]; then
-		lockf -k ${pb}/${arch}/queue/.lock ${pb}/scripts/releasemachine ${arch} ${host}
-	    fi
-	fi
-    done
-    
-    . ${pb}/${arch}/portbuild.conf
-    test -f ${pb}/${arch}/portbuild.${host} && . ${pb}/${arch}/portbuild.${host}
-    
-    rm -f ${builddir}/logs/${pkgname}.log ${builddir}/logs/${pkgname}.log.bz2
-    rm -f ${builddir}/errors/${pkgname}.log ${builddir}/errors/${pkgname}.log.bz2
-    
-    echo "dispatching: ${ssh_cmd} -a -t -n ${client_user}@${host} ${sudo_cmd} ${command} ${arch} ${branch} ${buildid} ${chroot} ${flags} \"$ED\" \"$PD\" \"$FD\" \"$BD\" \"$RD\" ${args}"
-    ${pb}/scripts/ptimeout.host $timeout ${ssh_cmd} -a -t -n ${client_user}@${host} ${sudo_cmd} ${command} ${arch} ${branch} ${buildid} ${chroot} ${flags} \"$ED\" \"$PD\" \"$FD\" \"$BD\" \"$RD\" ${args} 2>&1
-    error=$?
+chroot=
+. ${pb}/${arch}/portbuild.conf
+test -f ${pb}/${arch}/portbuild.${host} && . ${pb}/${arch}/portbuild.${host}
+chrootdata=$(${ssh_cmd} -a -n ${client_user}@${host} ${sudo_cmd} ${pb}/scripts/claim-chroot ${arch} ${branch} ${buildid} ${pkgname} 2>&1)
+if [ -z "${chrootdata}" ]; then
+    echo "Failed to claim chroot on ${host}"
+    exit 254
+fi
+
+case "${chrootdata}" in
+    */var/portbuild/scripts/claim-chroot*)
+	# Error executing script, assume system is booting
+	chrootdata="wait boot"
+	;;
+esac
+
+#    echo "Got ${chrootdata} from ${host}"
+
+set -- ${chrootdata}
+if [ $# -ge 2 ]; then
+    case $1 in
+	chroot)
+	    chroot=$2
+	    ;;
+	setup)
+	    echo "Setting up ${arch}/${branch} build ID ${buildid} on ${host}"
+
+	    # Run in the background so we can potentially
+	    # claim a slot on another machine.  In
+	    # practise I think we often end up trying
+	    # again on the same machine though.
+		    
+	    # Make sure to close stdin/stderr in the child
+	    # or make will hang until the child process
+	    # exits
+	    ${pb}/scripts/dosetupnode ${arch} ${branch} ${buildid} ${host} > /tmp/setupnode.$$ 2>&1 &
+	    exit 253
+	    ;;
+	error)
+	    echo "Error reported by ${host}: $2"
+	    ;;
+	wait)
+	    echo "Waiting for setup of ${host} to finish"
+	    ;;
+    esac
+    shift 2
+fi
+	
+if [ -z "${chroot}" ]; then
+    exit 254
+fi
+
+. ${pb}/${arch}/portbuild.conf
+test -f ${pb}/${arch}/portbuild.${host} && . ${pb}/${arch}/portbuild.${host}
     
-  #if grep -q " failed unexpectedly on " ${builddir}/logs/${pkgname}.pre.log; then
-  #  cat ${builddir}/logs/${pkgname}.pre.log | mail -s "${pkgname} failed uncleanly on ${arch} ${branch}" ${mailto}
-  #else
-  #  rm ${builddir}/logs/${pkgname}.pre.log
-  #fi
+rm -f ${builddir}/logs/${pkgname}.log ${builddir}/logs/${pkgname}.log.bz2
+rm -f ${builddir}/errors/${pkgname}.log ${builddir}/errors/${pkgname}.log.bz2
     
-    # Pull in the results of the build from the client
+${pb}/scripts/ptimeout.host $timeout ${ssh_cmd} -a -n ${client_user}@${host} ${sudo_cmd} ${command} ${arch} ${branch} ${buildid} ${chroot} ${flags} \"$ED\" \"$PD\" \"$FD\" \"$BD\" \"$RD\" ${args} 2>&1
+error=$?
+        
+# Pull in the results of the build from the client
     
-    ${scp_cmd} ${client_user}@${host}:${chroot}/tmp/${pkgname}.log ${builddir}/logs/${pkgname}.log
-    (${ssh_cmd} -a -n ${client_user}@${host} test -f ${chroot}/tmp/work.tbz ) && ${scp_cmd} ${client_user}@${host}:${chroot}/tmp/work.tbz ${builddir}/wrkdirs/${pkgname}.tbz
+${scp_cmd} ${client_user}@${host}:${chroot}/tmp/${pkgname}.log ${builddir}/logs/${pkgname}.log
+(${ssh_cmd} -a -n ${client_user}@${host} test -f ${chroot}/tmp/work.tbz ) && ${scp_cmd} ${client_user}@${host}:${chroot}/tmp/work.tbz ${builddir}/wrkdirs/${pkgname}.tbz
     
-    # XXX Set dirty flag if any of the scp's fail
+# XXX Set dirty flag if any of the scp's fail
     
-    mkdir -p ${builddir}/distfiles/.pbtmp/${pkgname}
-    ${ssh_cmd} -a -n ${client_user}@${host} tar -C ${chroot}/tmp/distfiles -cf - . | \
-	tar --unlink -C ${builddir}/distfiles/.pbtmp/${pkgname} -xvf -
+mkdir -p ${builddir}/distfiles/.pbtmp/${pkgname}
+${ssh_cmd} -a -n ${client_user}@${host} tar -C ${chroot}/tmp/distfiles -cf - . | \
+    tar --unlink -C ${builddir}/distfiles/.pbtmp/${pkgname} -xvf - && \
     touch ${builddir}/distfiles/.pbtmp/${pkgname}/.done
     
-    if [ "${error}" = 0 ]; then
-	${ssh_cmd} -a -n ${client_user}@${host} tar -C ${chroot}/tmp -cf - packages | \
-	    tar --unlink -C ${builddir} -xvf -
-	test -f ${builddir}/packages/All/${pkgname}${PKGSUFFIX} && \
-	    touch ${builddir}/packages/All/${pkgname}${PKGSUFFIX}
-	rm -f ${builddir}/errors/${pkgname}.log && \
-	    touch ${builddir}/errors/.force
-	lockf -k ${pbab}/failure.lock ${pb}/scripts/buildsuccess ${arch} ${branch} ${buildid} ${pkgname}
-	log=${builddir}/logs/$pkgname.log
-	if grep -q "even though it is marked BROKEN" ${log}; then
-	    echo | mail -s "${pkgname} BROKEN but built on ${arch} ${branch}" ${mailto}
-	fi
-	if grep -q "^list of .*file" ${log}; then
-	    buildlogdir=$(realpath ${builddir}/logs/)
-	    baselogdir=$(basename ${buildlogdir})
-	    (sed -e '/^build started/,$d' $log;echo;echo "For the full build log, see"; echo; echo "  http://${master}/errorlogs/${arch}-errorlogs/${baselogdir}/$(basename $log)";echo;sed -e '1,/^=== Checking filesystem state/d' $log) | mail -s "${pkgname} pkg-plist errors on ${arch} ${branch}" ${mailto}
-	fi
-    else
-	log=${builddir}/errors/${pkgname}.log
-	${scp_cmd} ${client_user}@${host}:${chroot}/tmp/${pkgname}.log ${log} || (echo ${chroot}@${host}; ${ssh_cmd} -a -n ${client_user}@${host} ls -laR ${chroot}/tmp) | mail -s "${pkgname} logfile not found" ${mailto}
-	if ! grep -q "even though it is marked BROKEN" ${log}; then
-	    buildlogdir=$(realpath ${builddir}/logs/)
-	    baselogdir=$(basename ${buildlogdir})
-	    if [ `wc -l ${log} | awk '{print $1}'` -le `expr ${loglength} + ${hdrlength}` ]; then
-		(echo "You can also find this build log at"; echo; echo "  http://${master}/errorlogs/${arch}-errorlogs/${baselogdir}/$(basename $log)";echo;cat ${log}) | mail -s "${pkgname} failed on ${arch} ${branch}" ${mailto}
-	    else
-		(echo "Excerpt from the build log at"; echo; echo "  http://${master}/errorlogs/${arch}-errorlogs/${baselogdir}/$(basename $log)";echo;sed -e '/^build started/,$d' $log;echo;echo "  [... lines trimmed ...]";echo;tail -${loglength} ${log}) | mail -s "${pkgname} failed on ${arch} ${branch}" ${mailto}
-	    fi
+if [ "${error}" = 0 ]; then
+    ${ssh_cmd} -a -n ${client_user}@${host} tar -C ${chroot}/tmp -cf - packages | \
+	tar --unlink -C ${builddir} -xvf -
+
+    # XXX why is this needed? 
+    test -f ${builddir}/packages/All/${pkgname}${PKGSUFFIX} && \
+	touch ${builddir}/packages/All/${pkgname}${PKGSUFFIX}
+
+    if [ -f ${builddir}/errors/${pkgname}.log ]; then
+	rm -f ${builddir}/errors/${pkgname}.log
+	# Force rebuild of html page to remove this package from list
+	touch ${builddir}/errors/.force
+    fi
+    lockf -k ${pbab}/failure.lock ${pb}/scripts/buildsuccess ${arch} ${branch} ${buildid} ${pkgname}
+    log=${builddir}/logs/$pkgname.log
+    if grep -q "even though it is marked BROKEN" ${log}; then
+	echo | mail -s "${pkgname} BROKEN but built on ${arch} ${branch}" ${mailto}
+    fi
+    if grep -q "^list of .*file" ${log}; then
+	buildlogdir=$(realpath ${builddir}/logs/)
+	baselogdir=$(basename ${buildlogdir})
+	(sed -e '/^build started/,$d' $log;echo;echo "For the full build log, see"; echo; echo "  http://${master}/errorlogs/${arch}-errorlogs/${baselogdir}/$(basename $log)";echo;sed -e '1,/^=== Checking filesystem state/d' $log) | mail -s "${pkgname} pkg-plist errors on ${arch} ${branch}" ${mailto}
+    fi
+else
+    log=${builddir}/errors/${pkgname}.log
+    ${scp_cmd} ${client_user}@${host}:${chroot}/tmp/${pkgname}.log ${log} || (echo ${chroot}@${host}; ${ssh_cmd} -a -n ${client_user}@${host} ls -laR ${chroot}/tmp) | mail -s "${pkgname} logfile not found" ${mailto}
+    if ! grep -q "even though it is marked BROKEN" ${log}; then
+	buildlogdir=$(realpath ${builddir}/logs/)
+	baselogdir=$(basename ${buildlogdir})
+	if [ $(wc -l ${log} | awk '{print $1}') -le $((loglength + hdrlength)) ]; then
+	    (echo "You can also find this build log at"; echo; echo "  http://${master}/errorlogs/${arch}-errorlogs/${baselogdir}/$(basename $log)";echo;cat ${log}) | mail -s "${pkgname} failed on ${arch} ${branch}" ${mailto}
+	else
+	    (echo "Excerpt from the build log at"; echo; echo "  http://${master}/errorlogs/${arch}-errorlogs/${baselogdir}/$(basename $log)";echo;sed -e '/^build started/,$d' $log;echo;echo "  [... lines trimmed ...]";echo;tail -${loglength} ${log}) | mail -s "${pkgname} failed on ${arch} ${branch}" ${mailto}
 	fi
-	lockf -k ${pbab}/failure.lock ${pb}/scripts/buildfailure ${arch} ${branch} ${buildid} ${pkgname}
     fi
-    
-    ${ssh_cmd} -a -n ${client_user}@${host} ${sudo_cmd} ${pb}/scripts/clean-chroot ${arch} ${branch} ${buildid} ${chroot} ${clean}
-    
-    lockf -k ${pb}/${arch}/queue/.lock ${pb}/scripts/releasemachine ${arch} ${host}
-    
+    lockf -k ${pbab}/failure.lock ${pb}/scripts/buildfailure ${arch} ${branch} ${buildid} ${pkgname}
+fi
+
+${ssh_cmd} -a -n ${client_user}@${host} ${sudo_cmd} ${pb}/scripts/clean-chroot ${arch} ${branch} ${buildid} ${chroot} ${clean}
+
   # XXX Set a dirty variable earlier and check here
-    if grep -q "^build of .*ended at" ${builddir}/logs/${pkgname}.log; then
-	exit ${error}
-    else
-	echo "Build of ${pkgname} in ${host}:/${chroot} failed uncleanly, rebuilding"
-	sleep 120
-    fi
-done
+if grep -q "^build of .*ended at" ${builddir}/logs/${pkgname}.log; then
+    exit ${error}
+else
+    echo "Build of ${pkgname} in ${host}:/${chroot} failed uncleanly"
+    exit 254
+fi
diff --git a/Tools/portbuild/scripts/pollmachine b/Tools/portbuild/scripts/pollmachine
index c3438041c881..ddc7100a94d1 100755
--- a/Tools/portbuild/scripts/pollmachine
+++ b/Tools/portbuild/scripts/pollmachine
@@ -13,24 +13,20 @@
 #
 # options are:
 #   -daemon : poll repeatedly
-#   -queue  : update queue entries (XXX racy)
 
 #
 # TODO:
 # XXX qmgr notification of new/removed machines
-# XXX log state changes in daemon mode
-# XXX clean up inactive builds
-# XXX test thread shutdown
-# XXX needed an explicit way to request setup?
-# XXX signal handler
+# XXX counter before declaring a machine as dead
+# Declares a machine as online if it reports 0 data from infoseek?
 
 # * Deal with machines change OS/kernel version
 #     - ACL list might change!
 #          - take machine offline, update ACL/arch/etc, reboot, bring online
 
 import sys, threading, socket
-from popen2 import *
 from time import sleep
+import os, subprocess, logging
 
 if len(sys.argv) < 1:
     print "Usage: %s <arch> [<arch> ...]" % sys.argv[0]
@@ -39,14 +35,9 @@ if len(sys.argv) < 1:
 arches=set()
 mlist={}
 polldelay=0
-queue=0
 for i in sys.argv[1:]:
     if i == "-daemon":
-        polldelay = 30
-        continue
-
-    if i == "-queue":
-        queue = 1
+        polldelay = 180
         continue
 
     if "/" in i:
@@ -82,9 +73,6 @@ class MachinePoll(threading.Thread):
     host = None
     port = 414
 
-    # Should we update queue entry?
-    queue = None
-
     timeout = None	# How often to poll
     shutdown = False # Exit at next poll wakeup
 
@@ -94,17 +82,21 @@ class MachinePoll(threading.Thread):
     # Dictionary of variables reported by the client
     vars = None
 
-    def __init__(self, mach, arch, timeout, host, port, queue):
+    def __init__(self, mach, arch, timeout, host, port):
         super(MachinePoll, self).__init__()
         self.mach = mach
         self.arch = arch
         self.timeout = timeout
         self.host = host
         self.port = port
-        self.queue = queue
+
+        # How many times the connection timed out since last success
+        self.timeouts = 0 
 
         self.vars = {}
     
+        self.setDaemon(True)
+
     def run(self):
         while True:
             if self.shutdown:
@@ -124,63 +116,104 @@ class MachinePoll(threading.Thread):
         lines = []
         try:
             s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+            s.settimeout(60)
             s.connect((self.host, self.port))
-            f = s.makefile()
-        
-            lines = f.readlines()
+
+            data = ""
+            while len(data) < 65536:
+                chunk = s.recv(8192)
+                if not chunk:
+                    break
+                data += chunk
+
             nowonline = True
+            self.timeouts = 0
+            lines = data.split("\n")
+        except socket.timeout:
+            if self.online:
+                logging.info("[%s] Connection timeout" % self.mach)
+            self.timeouts += 1
+            if self.timeouts < 3:
+                nowonline = self.online
         except:
             pass
-        finally:
+	finally:
             try:
                 s.close()
             except:
                 pass
 
         if nowonline != self.online:
-            print "State change: %s %s -> %s" % (self.mach, self.online, nowonline)
+            logging.info("[%s] Now %s" % (self.mach, "online" if nowonline else "OFFLINE"))
             self.online = nowonline
+	    if self.online:
+		self.timeouts = 0
             # XXX inform qmgr of state change
 
-        if self.online and not lines:
+        if self.online and not lines and not self.timeouts:
             # reportload script is missing
             dosetup=1
         else:
             dosetup=0
 
         for line in lines:
+            if line == "":
+                continue
             line=line.rstrip()
             part=line.partition('=')
             if part[1] != '=' or not part[0]:
 #                if "No such file or directory" in line:
 #                    # Client may require setting up post-boot
 #                    dosetup=1
-                print "Bad input from %s: %s" % (self.mach, line)
+                logging.info("[%s] Bad input: %s" % (self.mach, line))
                 # Assume client needs setting up
                 dosetup=1
-
             try:
                 old = self.vars[part[0]]
             except KeyError:
                 old = ""
             if old != part[2]:
                 self.vars[part[0]] = part[2]
-#                print "%s@%s: \"%s\" -> \"%s\"" % (part[0], self.mach, old, part[2])
+#                logging.info("%s@%s: \"%s\" -> \"%s\"" % (part[0], self.mach, old, part[2]))
                 # XXX update qmgr
 
+        try:
+            envs = self.vars['buildenvs']
+            for e in envs.split():
+                (arch, branch, buildid) = e.split("/")
+                f = "/var/portbuild/%s/%s/builds/%s/.active" % \
+                    (arch, branch, buildid)
+                if os.path.exists(f):
+                    continue
+                # Clean up a stale buildenv
+                logging.info("[%s] Cleaning up stale build: %s" % (self.mach, e))
+                (err, out) = self.setup(branch, buildid, "-nocopy -full")
+                if err:
+                    logging.info("[%s] Error from cleanup" % (self.mach))
+                for l in out.split("\n"):
+                    if l == "":
+                        continue
+                    logging.info("[%s] %s" % (self.mach, l))
+
+        except KeyError:
+            pass
+
         if dosetup:
-            print "Setting up %s" % (self.mach)
-            (err, out) = self.setup()
+            logging.info("[%s] Setting up machine" % (self.mach))
+            (err, out) = self.setup("-", "-")
             if err:
-                print "Error from setup of %s:" % (self.mach)
-                print out
-            print "Setup of %s complete" % (self.mach)
-            return
+                logging.info("[%s] Error from setup" % (self.mach))
+                for l in out.split("\n"):
+                    if l == "":
+                        continue
+                    logging.info("[%s] %s" % (self.mach, l))
+            logging.info("[%s] Setup complete" % (self.mach))
                             
         # Validate that arch has not changed (e.g. i386 -> amd64)
         try:
             if self.arch != self.vars['arch']:
-                print "Machine %s reporting unexpected arch: %s -> %s" % (self.mach, self.arch, self.vars['arch'])
+                logging.info("[%s] Unexpected arch: %s -> %s" % \
+                    (self.mach, self.arch, self.vars['arch']))
         except KeyError:
             pass
 
@@ -195,24 +228,27 @@ class MachinePoll(threading.Thread):
             pass
         f.close()
 
-        if self.queue:
-            try:
-                f = file("%s/%s/queue/%s" % (pb, self.arch, self.mach), "w")
-            except:
-                return
-
-            try:
-                f.write("%s\n" % self.vars['jobs'])
-            except:
-                pass
-            f.close()
-
-    def setup(self):
-        child = Popen4("su ports-%s -c \"/var/portbuild/scripts/dosetupnode %s - - %s\"" % (self.arch, self.arch, self.mach), 0)
+    def setup(self, branch, buildid, args = ""):
+        cmd = "su ports-%s -c \"/var/portbuild/scripts/dosetupnode %s %s %s %s %s\""\
+            % (self.arch, self.arch, branch, buildid, self.mach, args)
+        child = subprocess.Popen(cmd, shell=True, stderr = subprocess.STDOUT,
+                                 stdout = subprocess.PIPE)
         err = child.wait()
-        out = "".join(child.fromchild.readlines())
+        out = "".join(child.stdout.readlines())
         return (err, out)
 
+logging.basicConfig(level=logging.INFO,
+                    format='[%(asctime)s] %(message)s',
+                    datefmt='%d %b %Y %H:%M:%S',
+                    filename='/var/log/pollmachine.log', filemode='w') 
+
+log_console = logging.StreamHandler()
+log_console.setLevel(logging.INFO)
+formatter = logging.Formatter('[%(asctime)s] %(message)s',
+                              datefmt = '%d %b %Y %H:%M:%S')
+log_console.setFormatter(formatter) 
+logging.getLogger('').addHandler(log_console) 
+
 while True:
     for arch in arches:
         try:
@@ -233,28 +269,30 @@ while True:
         machines[arch]=now
 
         for mach in gone:
-            print "Removing machine %s" % mach
+            logging.info("Removing machine %s/%s" % (arch, mach))
             # XXX disable from qmgr
             pollthreads[mach].shutdown=True
             del pollthreads[mach]
 
         for mach in new:
-            print "Adding machine %s" % mach
+            logging.info("Adding machine %s/%s" % (arch, mach))
             # XXX set up qmgr
 
             pc="%s/%s/portbuild.conf" % (pb, arch)
             pch="%s/%s/portbuild.%s" % (pb, arch, mach)
-            config = Popen4("test -f %s && . %s; test -f %s && . %s; echo $infoseek_host; echo $infoseek_port" % (pc, pc, pch, pch))
-            host=config.fromchild.readline().rstrip()
+            cmd = "test -f %s && . %s; test -f %s && . %s; echo $infoseek_host; echo $infoseek_port" % (pc, pc, pch, pch)
+            config = subprocess.Popen(cmd, shell = True,
+                                      stdout = subprocess.PIPE)
+            host=config.stdout.readline().rstrip()
             if not host:
                 host = mach
-            port=config.fromchild.readline().rstrip()
+            port=config.stdout.readline().rstrip()
             try:
                 port = int(port)
             except (TypeError, ValueError):
                 port = 414
 
-            pollthreads[mach] = MachinePoll(mach, arch, polldelay, host, port, queue)
+            pollthreads[mach] = MachinePoll(mach, arch, polldelay, host, port)
             pollthreads[mach].start()
 
     if not polldelay:
diff --git a/Tools/portbuild/scripts/ptimeout.c b/Tools/portbuild/scripts/ptimeout.c
index 33e9ecc4f3a9..915024ce15df 100644
--- a/Tools/portbuild/scripts/ptimeout.c
+++ b/Tools/portbuild/scripts/ptimeout.c
@@ -47,7 +47,7 @@ main(int argc, char *argv[])
 		    /*printf("exited child is %d, status is %d\n", child, status);*/
 		    if (pid1 = child) {
 			/*printf("killing process %d\n", pid2);*/
-			kill(pid2, SIGTERM);
+			kill(pid2, SIGKILL);
 		    } else {
 			/*printf("killing process %d\n", pid1);*/
 			kill(pid1, SIGTERM);
diff --git a/Tools/portbuild/scripts/reportload b/Tools/portbuild/scripts/reportload
index 130348eceb5c..56ec0df5196d 100755
--- a/Tools/portbuild/scripts/reportload
+++ b/Tools/portbuild/scripts/reportload
@@ -16,7 +16,7 @@ fi
 error=
 for i in squid disk; do
   if [ -f ${scratchdir}/.${i} ]; then
-    error=${i} "${error}"
+    error="${i} ${error}"
   fi
 done
 
@@ -36,4 +36,4 @@ done
 echo "buildenvs=${buildenvs}"
 echo -n "load="
 uptime
-echo "error=${error}"
\ No newline at end of file
+echo "error=${error}"
diff --git a/Tools/portbuild/scripts/straslivy.py b/Tools/portbuild/scripts/straslivy.py
index b67b75dd2a8d..8a899399a4f2 100755
--- a/Tools/portbuild/scripts/straslivy.py
+++ b/Tools/portbuild/scripts/straslivy.py
@@ -26,7 +26,7 @@ def getpdispatch():
 		pid = fields[0]
 		arch = fields[3]
 		branch = fields[4]
-		port = fields[7].replace('/usr/ports/','')
+		port = fields[9].replace('/usr/ports/','')
 		line = ' '.join(fields)
 
 		if len(arch) > archwidth:
diff --git a/Tools/portbuild/scripts/zbackup b/Tools/portbuild/scripts/zbackup
index c51b9342b5fe..18f806fddbb6 100755
--- a/Tools/portbuild/scripts/zbackup
+++ b/Tools/portbuild/scripts/zbackup
@@ -16,7 +16,7 @@ backuplist=["a", "a/nfs", "a/src", "a/local", "a/ports", "a/portbuild",
 backupdir="/dumpster/pointyhat/backup"
 
 # How many days between full backups
-fullinterval=14
+fullinterval=3
 
 def validate():
     fslist = zfs.getallfs()
diff --git a/Tools/portbuild/scripts/zclient b/Tools/portbuild/scripts/zclient
new file mode 100755
index 000000000000..636aa265598c
--- /dev/null
+++ b/Tools/portbuild/scripts/zclient
@@ -0,0 +1,124 @@
+#!/usr/bin/env python
+
+# ZFS snapshot client
+
+import socket, os, sys
+
+ZSERVER = ('gohan10.freebsd.org', 8888)
+ZFSLOCAL = '/tmp/.zserver'
+
+def connect():
+    """ Connects to service, returns (socket, islocal) """
+
+    if os.path.exists(ZFSLOCAL):
+        s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
+        try:
+            s.connect(ZFSLOCAL)
+            return (s, True)
+        except:
+            s.close()
+    
+    s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+    try:
+        s.connect(ZSERVER)
+        return (s, False)
+    except:
+        s.close()
+
+    return (None, None)
+
+def send(sock, cmd):
+    """ Send a command, retrieve single line of reply """
+
+    sock.write(cmd)
+    sock.flush()
+
+    res= sock.readline()
+    #print "res = %s" % res
+    return res
+
+def error(msg):
+    print >>sys.stderr, "%s: %s" % (sys.argv[0], msg.rstrip())
+    sys.exit(1)
+
+def do_list(sockfile, islocal, args):
+    res = send(sockfile, "LIST\n")
+    if res[0] == "2":
+        for i in sockfile:
+            print i.rstrip()
+    else:
+        error(res[4:])
+
+def do_get(sockfile, islocal, args):
+    res = send(sockfile, "GET %s %s\n" % (args[0], args[1]))
+    if res[0] == "2":
+        while True:
+            block = sockfile.read(32*1024)
+            if not block:
+                break
+            sys.stdout.write(block)
+    else:
+        error(res[4:])
+
+def do_diff(sockfile, islocal, args):
+    res = send(sockfile, "DIFF %s %s %s\n" % (args[0], args[1], args[2]))
+    if res[0] == "2":
+        while True:
+            block = sockfile.read(32*1024)
+            if not block:
+                break
+            sys.stdout.write(block)
+    else:
+        error(res[4:])
+
+def do_reg(sockfile, islocal, args):
+    if not sock[1]:
+        error("must register on local machine")
+    res = send(sockfile, "REGISTER %s\n" % args[0])
+    if res[0] == "2":
+        print res[4:]
+    else:
+        error(res[4:])
+
+def do_unreg(sockfile, islocal, args):
+    if not sock[1]:
+        error("must register on local machine")
+    res = send(sockfile, "UNREGISTER %s\n" % args[0])
+
+    if res[0] == "2":
+        print res[4:]
+    else:
+        error(res[4:])
+
+def do_help(sockfile, islocal, args):
+    for (i, val) in sorted(cmddict.iteritems()):
+        print "%15s - %s" % (i, val[1])
+
+cmddict = {'list':(do_list, 'List available filesystem/snapshot pairs'),
+           'get':(do_get, 'Get a snapshot'),
+           'diff':(do_diff, 'Get the diffs between two snapshots'),
+           'register':(do_reg, 'Register a new filesystem (privileged)'),
+           'reg':(do_reg, 'Alias for register'),
+           'unregister':(do_unreg, 'Register a new filesystem (privileged)'),
+           'unreg':(do_unreg, 'Alias for register'),
+           'help':(do_help, 'Display this help')}
+
+if __name__ == "__main__":
+
+    try:
+        sock = connect()
+    except:
+        raise
+        sys.exit(1)
+
+    args = sys.argv
+
+    try:
+        cmd = args[1]
+        arg = args[2:]
+#        print "cmd = %s, arg = %s" % (cmd, arg)
+        cmddict[cmd][0](sock[0].makefile(), sock[1], arg)
+    except (KeyError, IndexError):
+        raise
+        error("No such command\n")
+    
diff --git a/Tools/portbuild/scripts/zsync b/Tools/portbuild/scripts/zsync
new file mode 100755
index 000000000000..62f8e2b27e31
--- /dev/null
+++ b/Tools/portbuild/scripts/zsync
@@ -0,0 +1,38 @@
+#!/bin/sh
+
+base=/a/cache
+zbase=a/cache
+
+fs=$1
+
+rsnap=$(zclient list | grep "^$fs " | tail -1 | awk '{print $2}')
+if [ -z "$rsnap" ]; then
+	echo "No such filesystem $fs"
+	exit 1
+fi
+
+lsnap=$(zfs list -Ht snapshot | grep "^$zbase/$fs@" | tail -1 | sed -e "s,^$zbase/$fs@,," | awk '{print $1}')
+if [ -z "$lsnap" ]; then
+	echo "No local snapshot found"
+	dofull=1
+else
+	if [ "$lsnap" = "$rsnap" ]; then
+		exit 0
+	fi
+	# Check for remotve snapshot
+	if ! (zclient list | grep "^$fs $lsnap " > /dev/null); then
+		echo "Local snapshot not found, removing and resyncing"
+		zfs destroy $zbase/$fs@$lsnap
+		dofull=0
+	else
+		dofull=1
+	fi
+fi
+
+if [ "$dofull" = "1" ]; then
+	zfs destroy -r ${zbase}/${fs}
+	zclient get ${fs} ${rsnap} | zcat | zfs receive ${zbase}/${fs}
+else
+	zclient diff ${fs} ${lsnap} ${rsnap} | zcat | zfs receive -F ${zbase}/${fs}
+fi
+	
-- 
cgit