Implement bbot agent startup and monitoring

1 files changed, 387 insertions, 182 deletions

diff --git a/buildos b/buildos
index e9c6acd..2de20f9 100755
--- a/buildos
+++ b/buildos
@@ -95,7 +95,7 @@ function restart ()
   sudo systemctl reboot
 }
 
-# Toolchain-related funtions.
+# Process toolchains.
 #
 
 # Return the value of one of the toolchain_* variables for this toolchain.
@@ -106,172 +106,74 @@ function tc_value () # <toolchain-prefix> <variable>
   echo "${!n}"
 }
 
-# Calculate the file checksum using the shaNNNsum utility.
-#
-function tc_checksum () # <toolchain-prefix> <file>
-{
-  "$(tc_value "$1" toolchain_csum)sum" -b "$2" | \
-    sed -n -re 's/^([^ ]+) .+$/\1/p'
-}
-
-# Fetch a file from the sums file into $toolchain_root, verify its checksum,
-# and make a predictable name (without version) symlink.
-#
-function tc_fetch () # <toolchain-prefix> <line>
-{
-  local s p f u l tp tu tr tv
-
-  tp="$1"
+toolchain_names=()
+for tn in "${!toolchains[@]}"; do
+  tp="${toolchains["$tn"]}"
   tu="$(tc_value "$tp" toolchain_url)"
-  tr="$(tc_value "$tp" toolchain_root)"
 
-  s="$(sed -n -re 's/^([^ ]+) .+$/\1/p' <<<"$2")"       # Checksum.
-  p="$(sed -n -re 's/^[^ ]+ \*([^ ]+)$/\1/p' <<<"$2")"  # File path (relative).
-  f="$(sed -n -re 's%^(.+/)?([^/]+)$%\2%p' <<<"$p")"    # File name.
-  u="$(sed -n -re 's%^(.+)/[^/]+$%\1%p' <<<"$tu")/$p"   # File URL.
-
-  if [ -z "$s" -o -z "$p" -o -z "$f" -o -z "$u" ]; then
-    info "invalid sum line '$2'"
-    return 1
-  fi
-
-  # Extract the version.
-  #
-  tv="$(tc_value "$tp" toolchain_ver)"
-
-  if [ -z "$tv" ]; then
-    tv="$(sed -n -re 's/build2-toolchain-(.+)\.tar.*/\1/p' <<<"$f")"
-
-    if [ -z "$tv" ]; then
-      info "unable to extract toolchain version from '$f'"
-      return 1
-    fi
-
-    declare -g "${tp}toolchain_ver=$tv"
-
-    info "toolchain version $tv"
-    echo "$tv" >"$tr/version"
+  if [ -z "$tu" ]; then
+    continue
   fi
 
-  # Derive a predictable name link.
-  #
-  l="$(sed -n -re "s/^(.+)-$tv(.*)$/\1\2/p" <<<"$f")"
-
-  if [ -z "$l" ]; then
-    info "unable to derive predicatable name from '$f' and '$tv'"
-    return 1
-  fi
+  toolchain_names+=("$tn")
 
-  # Fetch the file.
+  # The toolchain "sums" file (a list of SHA sums and relative file names, as
+  # produced by shaNNNsum). The first entry should always be build2-toolchain
+  # tar archive itself (which we use to figure out the version). Blank lines
+  # and lines that start with '#' are ignored.
   #
-  info "fetching $u [$l]"
+  tf="$(sed -n -re 's%^.+/([^/]+)$%\1%p' <<<"$tu")"
 
-  if ! curl -f -L -s -S -o "$tr/$f" "$u"; then
-    info "unable to fetch $u"
-    return 1
-  fi
+  declare "${tp}toolchain_file=$tf"
+  declare "${tp}toolchain_csum=$(sed -n -re 's%^.+\.([^.]+)$%\1%p' <<<"$tf")"
+  declare "${tp}toolchain_root=/build/tftp/toolchain/$tn"
+  declare "${tp}toolchain_ver="
 
-  # Verify the checksum.
+  # If buildos.toolchain_trust was not specified, set it to "no" so that
+  # we don't prompt if the repository happens to be signed.
   #
-  info "verifying checksum for $f"
-
-  local cs
-  cs="$(tc_checksum "$tp" "$tr/$f")"
-
-  if [ "$cs" != "$s" ]; then
-    info "checksum mismatch for $u"
-    info "  expected:   $s"
-    info "  calculated: $cs"
-    return 1
+  if [ -z "$(tc_value "$tp" toolchain_trust)" ]; then
+    declare "${tp}toolchain_trust=no"
   fi
+done
 
-  # Make the link.
-  #
-  ln -s "$f" "$tr/$l"
-}
-
-# Bootstrap the toolchain.
+# Divide CPUs and memory (in kB) among the toolchains.
 #
-function tc_bootstrap () # <toolchain-name>
-{
-  local tn="$1"
-  local tp="${toolchains["$tn"]}"
-  local tr="$(tc_value "$tp" toolchain_root)"
-  local tf="$(tc_value "$tp" toolchain_file)"
-
-  # Fetch files according to the sums file. Skip empty lines and those that
-  # start with '#'.
-  #
-  local l ls=()
-
-  readarray -t ls < <(sed -e '/^\s*#/d;/^\s*$/d' "$tr/$tf")
-
-  for l in "${ls[@]}"; do
-    if ! tc_fetch "$tp" "$l"; then
-      return 1 # Diagnostics has already been issued.
-    fi
-  done
-
-  local tv="$(tc_value "$tp" toolchain_ver)"   # Should be set by tc_fetch().
-  local tt="$(tc_value "$tp" toolchain_trust)"
-
-  # Bootstrap in /tmp/toolchains/$tn/, install to /build/toolchains/$tn/.
-  #
-  local wd="/tmp/toolchains/$tn"
-  local id="/build/toolchains/$tn"
-
-  mkdir -p "$wd"
-  mkdir -p "$id"
-
-  local r=1
-
-  cd "$wd"
-  while true; do # The "breakout loop".
-
-    # Extract the toolchain.
-    #
-    if ! tar -xf "$tr/build2-toolchain.tar.xz"; then
-      info "unable to extract $tr/build2-toolchain.tar.xz"
-      break
-    fi
-
-    cd "build2-toolchain-$tv"
-
-    # Bootstrap, stage, and install using the provided build.sh script.
-    #
-    if ! ./build.sh --install-dir "$id" --trust "$tt" g++; then
-      info "failed to build $(pwd)"
-      break
-    fi
+# Reserve 4G of RAM for ourselves (rootfs, tmpfs).
+#
+mem_total="$(sed -n -re 's/^MemTotal: *([0-9]+) *kB$/\1/p' </proc/meminfo)"
+cpu_total="$(lscpu | sed -n -re 's/^CPU\(s\): *([0-9]+)$/\1/p')"
 
-    r=0
-    break
-  done
-  cd "$owd"
+mem_slice=$(("$mem_total" - 4 * 1024 * 1024))
+cpu_slice="$cpu_total"
 
-  # Clean up.
-  #
-  rm -r "$wd"
+if [ "${#toolchain_names[@]}" -gt 1 ]; then
+  mem_slice=$(("$mem_slice" / "${#toolchain_names[@]}"))
+  cpu_slice=$(("$cpu_slice" / "${#toolchain_names[@]}"))
 
-  return "$r"
-}
+  if [ "$cpu_slice" -eq 0 ]; then
+    cpu_slice=1
+  fi
+fi
 
 # Print monitor configuration as email body.
 #
 function print ()
 {
+  echo "cpu_total: $cpu_total"
+  echo "cpu_slice: $cpu_slice"
+
+  echo "mem_total: $mem_total kB"
+  echo "mem_slice: $mem_slice kB"
+  echo
+
   echo "buildid:       $buildid"
   echo "buildid_url:   $buildid_url"
   echo
 
-  for tn in "${!toolchains[@]}"; do
+  for tn in "${toolchain_names[@]}"; do
     tp="${toolchains["$tn"]}"
     tu="$(tc_value "$tp" toolchain_url)"
-
-    if [ -z "$tu" ]; then
-      continue
-    fi
-
     tt="$(tc_value "$tp" toolchain_trust)"
 
     echo "$tn.toolchain_url:   $tu"
@@ -286,38 +188,7 @@ if [ -z "$buildid_url" ]; then
   info "no buildos.buildid_url specified, not monitoring for new os builds"
 fi
 
-tc=
-for tn in "${!toolchains[@]}"; do
-  tp="${toolchains["$tn"]}"
-  tu="$(tc_value "$tp" toolchain_url)"
-
-  if [ -z "$tu" ]; then
-    continue
-  fi
-
-  tc="true"
-
-  # The toolchain "sums" file (a list of SHA sums and relative file names, as
-  # produced by shaNNNsum). The first entry should always be build2-toolchain
-  # tar archive itself (which we use to figure out the version). Blank lines
-  # and lines that start with '#' are ignored.
-  #
-  tf="$(sed -n -re 's%^.+/([^/]+)$%\1%p' <<<"$tu")"
-
-  declare "${tp}toolchain_file=$tf"
-  declare "${tp}toolchain_csum=$(sed -n -re 's%^.+\.([^.]+)$%\1%p' <<<"$tf")"
-  declare "${tp}toolchain_root=/build/tftp/toolchains/$tn"
-  declare "${tp}toolchain_ver="
-
-  # If buildos.toolchain_trust was not specified, set it to "no" so that
-  # we don't prompt if the repository happens to be signed.
-  #
-  if [ -z "$(tc_value "$tp" toolchain_trust)" ]; then
-    declare "${tp}toolchain_trust=no"
-  fi
-done
-
-if [ -z "$tc" ]; then
+if [ "${#toolchain_names[@]}" -eq 0 ]; then
   info "no buildos.toolchain_url specified, not bootstrapping"
 fi
 
@@ -400,7 +271,7 @@ for v in /build/machines/*; do
 	# <name>-<toolchain> (bootstrapped image)
 	#
 	f=
-	for tn in "${!toolchains[@]}"; do
+	for tn in "${toolchain_names[@]}"; do
 	  if [[ "$s" =~ ^"$m"-"$tn"$ ]]; then
 	    f="true"
 	    break
@@ -471,20 +342,303 @@ if [ "${#diag[@]}" -gt 0 ]; then
   fi
 fi
 
+# Toolchain-related funtions.
+#
+
+# Calculate the file checksum using the shaNNNsum utility.
+#
+function tc_checksum () # <toolchain-prefix> <file>
+{
+  "$(tc_value "$1" toolchain_csum)sum" -b "$2" | \
+    sed -n -re 's/^([^ ]+) .+$/\1/p'
+}
+
+# Fetch a file from the sums file into $toolchain_root, verify its checksum,
+# and make a predictable name (without version) symlink.
+#
+function tc_fetch () # <toolchain-prefix> <line>
+{
+  local s p f u l tp tu tr tv
+
+  tp="$1"
+  tu="$(tc_value "$tp" toolchain_url)"
+  tr="$(tc_value "$tp" toolchain_root)"
+
+  s="$(sed -n -re 's/^([^ ]+) .+$/\1/p' <<<"$2")"       # Checksum.
+  p="$(sed -n -re 's/^[^ ]+ \*([^ ]+)$/\1/p' <<<"$2")"  # File path (relative).
+  f="$(sed -n -re 's%^(.+/)?([^/]+)$%\2%p' <<<"$p")"    # File name.
+  u="$(sed -n -re 's%^(.+)/[^/]+$%\1%p' <<<"$tu")/$p"   # File URL.
+
+  if [ -z "$s" -o -z "$p" -o -z "$f" -o -z "$u" ]; then
+    info "invalid sum line '$2'"
+    return 1
+  fi
+
+  # Extract the version.
+  #
+  tv="$(tc_value "$tp" toolchain_ver)"
+
+  if [ -z "$tv" ]; then
+    tv="$(sed -n -re 's/build2-toolchain-(.+)\.tar.*/\1/p' <<<"$f")"
+
+    if [ -z "$tv" ]; then
+      info "unable to extract toolchain version from '$f'"
+      return 1
+    fi
+
+    declare -g "${tp}toolchain_ver=$tv"
+
+    info "toolchain version $tv"
+    echo "$tv" >"$tr/version"
+  fi
+
+  # Derive a predictable name link.
+  #
+  l="$(sed -n -re "s/^(.+)-$tv(.*)$/\1\2/p" <<<"$f")"
+
+  if [ -z "$l" ]; then
+    info "unable to derive predicatable name from '$f' and '$tv'"
+    return 1
+  fi
+
+  # Fetch the file.
+  #
+  info "fetching $u [$l]"
+
+  if ! curl -f -L -s -S -o "$tr/$f" "$u"; then
+    info "unable to fetch $u"
+    return 1
+  fi
+
+  # Verify the checksum.
+  #
+  info "verifying checksum for $f"
+
+  local cs
+  cs="$(tc_checksum "$tp" "$tr/$f")"
+
+  if [ "$cs" != "$s" ]; then
+    info "checksum mismatch for $u"
+    info "  expected:   $s"
+    info "  calculated: $cs"
+    return 1
+  fi
+
+  # Make the link.
+  #
+  ln -s "$f" "$tr/$l"
+}
+
+# Bootstrap the toolchain.
+#
+function tc_bootstrap () # <toolchain-name>
+{
+  local tn="$1"
+  local tp="${toolchains["$tn"]}"
+  local tr="$(tc_value "$tp" toolchain_root)"
+  local tf="$(tc_value "$tp" toolchain_file)"
+
+  # Fetch files according to the sums file. Skip empty lines and those that
+  # start with '#'.
+  #
+  local l ls=()
+
+  readarray -t ls < <(sed -e '/^\s*#/d;/^\s*$/d' "$tr/$tf")
+
+  for l in "${ls[@]}"; do
+    if ! tc_fetch "$tp" "$l"; then
+      return 1 # Diagnostics has already been issued.
+    fi
+  done
+
+  local tv="$(tc_value "$tp" toolchain_ver)"   # Should be set by tc_fetch().
+  local tt="$(tc_value "$tp" toolchain_trust)"
+
+  # Bootstrap in /tmp/toolchain/$tn/, install to /build/toolchain/$tn/.
+  #
+  local wd="/tmp/toolchain/$tn"
+  local id="/build/toolchain/$tn"
+
+  mkdir -p "$wd"
+  mkdir -p "$id"
+
+  local r=1
+
+  cd "$wd"
+  while true; do # The "breakout loop".
+
+    # Extract the toolchain.
+    #
+    if ! tar -xf "$tr/build2-toolchain.tar.xz"; then
+      info "unable to extract $tr/build2-toolchain.tar.xz"
+      break
+    fi
+
+    cd "build2-toolchain-$tv"
+
+    # Bootstrap, stage, and install using the provided build.sh script.
+    #
+    if ! ./build.sh --install-dir "$id" --trust "$tt" g++; then
+      info "failed to build $(pwd)"
+      break
+    fi
+
+    cd "$wd"
+    rm -r "build2-toolchain-$tv"
+    mv -T build2-toolchain-* build2-toolchain # Strip version.
+
+    r=0
+    break
+  done
+  cd "$owd"
+
+  return "$r"
+}
+
+# Check if we need to build/start or rebuild/restart the bbot agent. Return
+# 0 if nothing to do, 1 for upgrades, 2 for first build, and 3 for failure.
+#
+function bb_check () # <toolchain-name>
+{
+  local tn="$1"
+
+  export PATH="/build/toolchain/$tn/bin:$PATH" # Running in subshell.
+
+  cd "/tmp/toolchain/$tn/build2-toolchain"
+
+  local r=3
+
+  local l_stat b_stat
+  while true; do # The "breakout loop".
+
+    l_stat="$(bpkg status libbbot)"
+    b_stat="$(bpkg status bbot)"
+
+    if ! bpkg fetch -q; then
+      info "failed to fetch package information"
+      break
+    fi
+
+    # See if this is the first time or if we need to upgrade.
+    #
+    if [ "$(cut -d ' ' -f 1 <<<"$b_stat")" = "configured" ]; then
+
+      # We assume that if anything has changed in the status line, then we
+      # have a new version.
+      #
+      if [ "$b_stat" = "$(bpkg status bbot)" -a \
+           "$l_stat" = "$(bpkg status libbbot)" ]; then
+	r=0
+	break
+      fi
+
+      r=1
+      break
+    fi
+
+    r=2
+    break
+  done
+  cd "$owd"
+
+  return "$r"
+}
+
+# Build and start bbot agent using the bpkg configuration created by
+# tc_bootstrap().
+#
+function bb_start () # <toolchain-name>
+{
+  local tn="$1"
+
+  local id="/build/bbot/$tn"
+  mkdir -p "$id"
+
+  # Install/uninstall vars.
+  #
+  local vars=(config.install.root="$id" config.bin.rpath="$id/lib")
+
+  export PATH="/build/toolchain/$tn/bin:$PATH" # Running in subshell.
+
+  cd "/tmp/toolchain/$tn/build2-toolchain"
+
+  local r=1
+
+  local b_word
+  while true; do # The "breakout loop".
+
+    b_word="$(bpkg status bbot | cut -d ' ' -f 1)"
+
+    # If upgrading, stop the service and uninstall.
+    #
+    if [ "$b_word" = "configured" ]; then
+
+      if ! sudo systemctl stop "bbot-agent@$tn"; then
+	info "failed to stop bbot-agent@$tn service, assuming not running"
+      fi
+
+      if ! bpkg uninstall "${vars[@]}" bbot; then
+	info "failed to uninstall bbot agent"
+	break
+      fi
+    fi
+
+    # Build and install the bbot agent.
+    #
+    if ! bpkg build --build-option --jobs --build-option "$cpu_slice" \
+	   --yes libbbot bbot; then
+      info "failed to build bbot agent"
+      break
+    fi
+
+    if ! bpkg install "${vars[@]}" bbot; then
+      info "failed to install bbot agent"
+      break
+    fi
+
+    # Post-process and install systemd .service file. Note that we cannot use
+    # the systemd pattern machinery since each version of bbot can have its
+    # own version of the .service file.
+    #
+    sed -i -re "s/%[iI]/$tn/g" "$id/lib/systemd/system/bbot-agent@.service"
+    sudo ln -sf "$id/lib/systemd/system/bbot-agent@.service" \
+      "/usr/lib/systemd/system/bbot-agent@$tn.service"
+
+    # Start the service.
+    #
+    if ! sudo systemctl start "bbot-agent@$tn"; then
+      info "failed to start bbot-agent@$tn service"
+      break
+    fi
+
+    r=0
+    break
+  done
+  cd "$owd"
+
+  return "$r"
+}
+
+# Array of bootstrapped toolchains.
+#
+# The idea is to collect them until we bootstrap all of them and only then
+# start their bbot agents.
+#
+toolchain_boots=()
+
 # Monitoring loop.
 #
+count=0
 while true; do
 
+  count=$(($count + 1))
+
   # Check for toolchain changes. If this is the first run, bootstrap them.
   #
-  for tn in "${!toolchains[@]}"; do
+  for tn in "${toolchain_names[@]}"; do
     tp="${toolchains["$tn"]}"
     tu="$(tc_value "$tp" toolchain_url)"
 
-    if [ -z "$tu" ]; then
-      continue
-    fi
-
     tr="$(tc_value "$tp" toolchain_root)"
     tf="$(tc_value "$tp" toolchain_file)"
     p="$tr/$tf"
@@ -531,20 +685,22 @@ EOF
 	#
 	info "bootstrapping $tn toolchain..."
 
-	tc_bootstrap "$tn" 2>&1 | tee "$tr/bootstrap.log" 1>&2
+	tc_bootstrap "$tn" 2>&1 | tee "$tr/bootstrap-$count.log" 1>&2
 
 	if [ "${PIPESTATUS[0]}" -eq 0 ]; then
 	  v="$(cat $tr/version)"
 	  declare "${tp}toolchain_ver=$v"
 
 	  s="bootstrapped $tn toolchain $v"
+	  toolchain_boots+=("$tn")
 	else
 	  s="failed to bootstrap $tn toolchain, waiting for new version"
+	  toolchain_boots+=("") # Skip.
 	fi
 
 	info "$s"
 	email "$s" <<EOF
-$tn.bootstrap_log: tftp://$hname/toolchains/$tn/bootstrap.log
+$tn.bootstrap_log: tftp://$hname/toolchain/$tn/bootstrap-$count.log
 EOF
       fi
     else
@@ -553,6 +709,55 @@ EOF
     fi
   done
 
+  # If we have boostrapped all the toolchains, (re)build and (re)start their
+  # bbot agents.
+  #
+  if [ "${#toolchain_names[@]}" -eq "${#toolchain_boots[@]}" ]; then
+
+    for tn in "${toolchain_boots[@]}"; do
+
+      # Skip those that failed to bootstrap.
+      #
+      if [ -z "$tn" ]; then
+	continue
+      fi
+
+      s=
+      bb_check "$tn" 2>&1 | tee "$tr/bbot-$count.log" 1>&2
+
+      case "${PIPESTATUS[0]}" in
+	0)
+	  rm -f "$tr/bbot-$count.log"
+	  continue # Nothing to do.
+	  ;;
+	1)
+	  s="re"
+	  ;&
+	2)
+	  info "${s}starting bbot-agent@$tn..."
+
+	  # Append to the same log.
+	  #
+	  bb_start "$tn" 2>&1 | tee -a "$tr/bbot-$count.log" 1>&2
+
+	  if [ "${PIPESTATUS[0]}" -eq 0 ]; then
+	    s="${s}started bbot-agent@$tn"
+	  else
+	    s="failed to ${s}start bbot-agent@$tn, waiting for new version"
+	  fi
+	  ;;
+	*)
+	  s="failed to fetch package information for $tn, will try again"
+	  ;;
+      esac
+
+      info "$s"
+      email "$s" <<EOF
+$tn.start_log: tftp://$hname/toolchain/$tn/bbot-$count.log
+EOF
+    done
+  fi
+
   # Check for OS changes.
   #
   if [ -n "$buildid_url" ]; then
@@ -574,5 +779,5 @@ EOF
   fi
 
   info "monitoring..."
-  sleep 10
+  sleep 20
 done