X-Git-Url: https://iankelling.org/git/?a=blobdiff_plain;f=btrfsmaint;h=0204ed5aa78a4e9043e0aae3a624066252a77955;hb=55d37e930d1cd5b3b9800d33dc62f282050899e9;hp=bcad450bad786b99cf372a9f20142f98d923573c;hpb=12cab163424e3a7b0815646d1d4407f9b5839bcb;p=distro-setup

diff --git a/btrfsmaint b/btrfsmaint
index bcad450..0204ed5 100755
--- a/btrfsmaint
+++ b/btrfsmaint
@@ -3,19 +3,33 @@
 
 [[ $EUID == 0 ]] || exec sudo -E "${BASH_SOURCE[0]}" "$@"
 
-source /a/bin/errhandle/err
+set -e; . /usr/local/lib/err; set +e
 
 # inspired from
 # https://github.com/kdave/btrfsmaintenance
 
+if [[ $INVOCATION_ID ]]; then
+  err-cleanup() {
+    exim -odf -i root <<EOF
+From: root@$(hostname -f)
+To: root@$(hostname -f)
+Subject: btrfsmaint automatically exited on command error
+
+journalctl -u btrfsmaint -n 50:
+$(journalctl -u btrfsmaint -n 50)
+EOF
+  }
+fi
 
-# Man page says we could also use a range, i suppose it would be
-# logical to use a pattern like 5..10 10..20,
-# but I don't know if this would help us at all.
 dusage="5 10"
 musage="5"
 
-e() { echo "cron: $*"; "$@"; }
+e() {
+  echo "btrfsmaint: $*"
+  if ! $dryrun; then
+    "$@"
+  fi
+}
 
 check-idle() {
   type -p xprintidle &>/dev/null || return 0
@@ -36,67 +50,124 @@ check-idle() {
   fi
 }
 
-
 usage() {
   cat <<EOF
-Usage: ${0##*/} args
-Do btrfs maintence or stop if xprintidle shows a user
+Usage: ${0##*/} [OPTIONS]
+Do btrfs maintence or stop if we have X and xprintidle shows a user
+
+Normally, no options are needed.
 
-force  Run regardless of user idle status
-check  Only check if an existing maintence should be cancelled due to
-         nonidle user. Also, runs in a loop every 20 seconds for 10
+--check  Only check if an existing maintence should be cancelled due to
+         nonidle user and run in a loop every 20 seconds for 10
          minutes.
 
+--dryrun Just print out what we would do.
+
+--force  Run regardless of user idle status on all disks and do scrub
+         regardless of when it was last run.
+--no-stats  Avoid checking error statistics. Use this to avoid a rare race
+            condition when running --check concurrently with normal run.
+
+
+-h|--help   Show help
+
 Note: Uses util-linux getopt option parsing: spaces between args and
 options, short options can be combined, options before args.
 EOF
   exit $1
 }
 
+##### begin command line parsing ########
+
+# ensure we can handle args with spaces or empty.
+ret=0; getopt -T || ret=$?
+[[ $ret == 4 ]] || { echo "Install util-linux for enhanced getopt" >&2; exit 1; }
 
-force=false
 check=false
-if [[ $1 ]]; then
+dryrun=false
+force=false
+stats=true
+
+temp=$(getopt -l help,check,dryrun,force,no-stats h "$@") || usage 1
+eval set -- "$temp"
+while true; do
   case $1 in
-    check)
-      check=true
-      ;;
-    force)
-      force=true
-      ;;
-    *)
-      echo "$0: error: unexpected arg" >&2
-      usage 1
-      ;;
+    --check) check=true ;;
+    --dryrun) dryrun=true ;;
+    --force) force=true ;;
+    --no-stats) stats=false ;;
+    -h|--help) usage ;;
+    --) shift; break ;;
+    *) echo "$0: unexpected args: $*" >&2 ; usage 1 ;;
   esac
-fi
+  shift
+done
+readonly check dryrun force stats
+##### end command line parsing ########
 
 
 main() {
   idle=true
   if ! $force; then
     check-idle
+    if ! $check; then
+      min=0
+      max_min=300
+      # When the cron kicks in, we may not be idle (physically sleeping) yet, so
+      # wait.
+      while ! $idle && (( min < max_min )); do
+        min=$(( min + 1 ))
+        sleep 60
+        check-idle
+      done
+      # If we've waited a really long time for idle, just give up.
+      if (( min == max_min )); then
+        return
+      fi
+    fi
   fi
 
-  tmp=$(mktemp)
 
   fnd="findmnt --types btrfs --noheading"
   for x in $($fnd --output "SOURCE" --nofsroot | sort -u); do
     mnt=$($fnd --output "TARGET" --first-only --source $x)
     [[ $mnt ]] || continue
 
-    if ! btrfs dev stats -c $mnt >$tmp; then
-      if diff -q $mnt/btrfs-dev-stats $tmp; then
-        diff -u $mnt/btrfs-dev-stats $tmp | mail -s "$HOSTNAME: error: btrfs dev stats -c $mnt" root@localhost
-        cat $tmp >$mnt/btrfs-dev-stats
+    #### begin look for diff in stats, eg: increasing error count ####
+    if $stats; then
+      tmp=$(mktemp)
+      # ${mnt%/} so that if mnt is / we avoid making a buggy looking path
+      stats_path=${mnt%/}/btrfs-dev-stats
+      if [[ ! -e $stats_path ]]; then
+        btrfs dev stats -c $mnt >$stats_path ||: # populate initial reading
+      elif ! btrfs dev stats -c $mnt >$tmp; then
+        if ! diff -q $stats_path $tmp; then
+          mv $stats_path $stats_path.1
+          cat $tmp >$stats_path
+          diff=$(diff -u $stats_path $tmp 2>&1 ||:)
+          printf "diff of: btrfs dev stats -c %s\n%s\n" "$mnt" "$diff"
+          exim -odf -i root <<EOF
+From: root@$(hostname -f)
+To: root@$(hostname -f)
+Subject: btrfsmaint: device stats changed for $mnt
+
+diff of: btrfs dev stats -c $mnt
+$diff
+EOF
+        fi
       fi
+      rm -f $tmp
     fi
+    #### end look for diff in stats, eg: increasing error count ####
 
-    if ! $idle; then
-      btrfs scrub cancel $mnt &>/dev/null ||:
-      continue
-    fi
     if $check; then
+      if ! $idle; then
+        if $dryrun; then
+          echo "$0: not idle. if this wasnt a dry run, btrfs scrub cancel $mnt"
+        else
+          btrfs scrub cancel $mnt &>/dev/null ||:
+        fi
+      fi
       continue
     fi
 
@@ -118,34 +189,54 @@ main() {
         e ionice -c 3 btrfs balance start -musage=$usage $mnt
       done
     fi
-    # e btrfs filesystem df $mnt
-    # e df -H $mnt
-    date=$(
-      btrfs scrub status $mnt | \
-        sed -rn 's/^\s*scrub started at (.*) and finished.*/\1/p'
-        )
-    if [[ $date ]]; then
+    date=
+    scrub_status=$(btrfs scrub status $mnt)
+    if printf "%s\n" "$scrub_status" | grep -i '^status:[[:space:]]*finished$' &>/dev/null; then
+      date=$(printf "%s\n" "$scrub_status" | sed -rn 's/^Scrub started:[[:space:]]*(.*)/\1/p')
+    fi
+    if [[ ! $date ]]; then
+      # output from older versions, at least btrfs v4.15.1
+      date=$(
+        printf "%s\n" "$scrub_status" | \
+          sed -rn 's/^\s*scrub started at (.*) and finished.*/\1/p'
+          )
+    fi
+    if ! $force && [[ $date ]]; then
+      if $dryrun; then
+        echo "$0: last scrub finish for $mnt: $date"
+      fi
       date=$(date --date="$date" +%s)
-      # if date is sooner than 90 days ago
+      # if date is sooner than 60 days ago
       # the wiki recommends 30 days or so, but
-      # it makes the comp lag like shit for a day,
-      # so I'm going with 90 days.
-      if (( date > $(date +%s) - 60*60*24*30 )); then
-        echo "cron: skiping scrub of $mnt"
+      # I'm going with 60 days.
+      if (( date > EPOCHSECONDS - 60*60*24*60 )); then
+        if $dryrun; then
+          echo "$0: skiping scrub of $mnt, last was $(( (EPOCHSECONDS - date) / 60/60/24 )) days ago, < 30 days"
+        fi
         continue
       fi
     fi
-    # -c 2 -n 4 is from btrfsmaintenance, does ionice
-    e btrfs scrub start -Bd -c 2 -n 4 $mnt
+    # btrfsmaintenance does -c 2 -n 4, but I want lowest pri.
+    e btrfs scrub start -Bd -c 3 $mnt
+
+    # We normally only do one disk since this is meant to be run in
+    # downtime and if we try to do all disks, we invariably end up doing
+    # a scrub after downtime. So, just do one disk per day.
+    if ! $force; then
+      return 0
+    fi
   done
 }
 
-if $check; then
-  # this is to prevent systemd from filling up the journal
-  for (( runcount=0; runcount < 90; runcount++ )); do
+loop-main() {
+  while true; do
     main
     sleep 60
   done
+}
+
+if $check; then
+  loop-main
 else
   main
 fi