X-Git-Url: https://iankelling.org/git/?a=blobdiff_plain;f=btrfsmaint;h=d2263257d412b4da2d9e37e5c1678641e301e8f6;hb=3ca4714d6b02ff279a0c724415f3e0a3a6f49059;hp=1639a8dd48f450e69fdf67558a4780059583ec1d;hpb=40dd151ec6ba75633c74568da59e35a45351f194;p=distro-setup

diff --git a/btrfsmaint b/btrfsmaint
index 1639a8d..d226325 100755
--- a/btrfsmaint
+++ b/btrfsmaint
@@ -41,13 +41,13 @@ check-idle() {
   fi
 }
 
-
 usage() {
   cat <<EOF
 Usage: ${0##*/} [ARGS]
 Do btrfs maintence or stop if xprintidle shows a user
 
-force  Run regardless of user idle status
+force  Run regardless of user idle status on all disks and do scrub
+       regardless of when it was last run.
 check  Only check if an existing maintence should be cancelled due to
          nonidle user. Also, runs in a loop every 20 seconds for 10
          minutes.
@@ -85,31 +85,64 @@ 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
-      fi
-    fi
+    #### begin look for diff in stats, eg: increasing error count ####
 
-    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 ||:
-        continue
+    # Only run for $check, since it runs in parallel to non-check, avoid
+    # race condition.
+    if $check; then
+      tmp=$(mktemp)
+      # if mnt is /, 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 $tmp; then
+              exim -t <<EOF
+From: root@$HOSTNAME.b8.nz
+To: alerts@iankelling.org
+Subject: btrfsmaintstop: btrfs dev stats -c $mnt
+
+$(diff -u $stats_path $tmp ||:)
+EOF
+          mv $stats_path $stats_path.1
+          cat $tmp >$stats_path
+        fi
       fi
+      rm -f $tmp
     fi
+    #### end look for diff in stats, eg: increasing error count ####
+
     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
 
@@ -143,24 +176,30 @@ main() {
           sed -rn 's/^\s*scrub started at (.*) and finished.*/\1/p'
           )
     fi
-    if [[ $date ]]; then
+    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
+      # 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 $(( ($(date +%s) - date) / 60/60/24 )) days ago, < 30 days"
+          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 while I sleep
+    # and if we try to do all disks, we invariably end up doing a scrub still
+    # after I've woken up. So, just do one per day.
+    if ! $force; then
+      return 0
+    fi
   done
 }