Changeset 42444 for trunk/tools/eam/rawfix.20230221/src/fix_chip_locations.pl

Timestamp:

Apr 6, 2023, 6:00:06 PM (3 years ago)

Author:

eugene

Message:

big re-work to allow repeat, iterated fixes

File:

• trunk/tools/eam/rawfix.20230221/src/fix_chip_locations.pl (modified) (11 diffs)

trunk/tools/eam/rawfix.20230221/src/fix_chip_locations.pl

@@ -12 +12 @@
 use Getopt::Long qw( GetOptions :config auto_help auto_version );
 
-# USAGE: fix_chip_locations.pl --dateobs (night) --chiplist (file) --stage (v0,v1,etc)
+# USAGE: fix_chip_locations.pl --dateobs (night) --chiplist (file) --stage (0,1,etc)
 
 # we need to select targets for ITC copies of files currently only at ATRC:
@@ -25 +25 @@
 
 # choose ITC machines on which to place backup copies:
-my @volarray = &get_neb_volumes;
+my @vol_copy = &get_neb_volumes_copy; # array of neb hosts to copy data to at ITC
+my %vol_good = &get_neb_volumes_good; # hash of valid neb hosts
 my $volnum = 0;
 
 my $Ngood = 0;
+my $Ndead = 0;
 my $Ngone = 0;
-my $Ndead = 0;
 my $Nsing = 0;
 my $Ndups = 0;
@@ -40 +41 @@
 my $Ncul3 = 0;
 
+# load the lists of good and fail chips
+my $prev = $stage - 1;
+my $goodfile = "$topdir/$dateword/goodchips.$prev.txt";
+my %goodchips;
+if (-e $goodfile) {
+    open (GOODFILE,  $goodfile);
+    my @lines = <GOODFILE>;
+    close (GOODFILE);
+    
+    foreach my $line (@lines) {
+        chomp $line;
+        $goodchips{$line} = 1;
+    }
+}
+my $failfile = "$topdir/$dateword/failchips.$prev.txt";
+my %failchips;
+if (-e $failfile) {
+    open (FAILFILE,  $failfile);
+    my @lines = <FAILFILE>;
+    close (FAILFILE);
+    
+    foreach my $line (@lines) {
+        chomp $line;
+        $failchips{$line} = 1;
+    }
+}
+
 foreach my $chip_id (@chip_ids) {
 
-    my $urifile = "$topdir/$dateword/$chip_id.uris.txt";
-    open (FILE,  $urifile);
-    my @lines = <FILE>;
-    close (FILE);
+    # read from chip_id file, update to next stage number
+    my $urifile = "$topdir/$dateword/$chip_id.uris.$stage.txt";
+    if (! -e $urifile) { die "missing file $urifile\n"; }
+    open (URIFILE,  $urifile);
+    my @lines = <URIFILE>;
+    close (URIFILE);
 
     foreach my $line (@lines) {
@@ -61 +91 @@
         my $md5sum = $words[3];
 
+        my $haveMD5 = 0;
+        my $goodMD5 = 0;
+        if ($goodchips{$ext_id}) { $haveMD5 = 1; $goodMD5 = 1; }
+        if ($failchips{$ext_id}) { $haveMD5 = 1; }
+
+
+        # nothing to be done, or need be done for these:
         if ($state eq "GOOD:") { $Ngood++; next; }
         if ($state eq "DEAD:") { $Ndead++; next; }
@@ -66 +103 @@
 
         # one copy at ITC: neb-replicate will create a copy at ATRC
+        # does not depend on existence of MD5sum
         if ($state eq "SING:") { 
             my $status = &vsystem ("neb-replicate $ext_id");
@@ -79 +117 @@
 
         # one copy at ATRC: neb-replicate will create a copy at ITC iff we supply the target volume
+        # does not depend on existence of MD5sum
         if ($state eq "SBCK:") { 
-            my $tgtvol = $volarray[$volnum];
+            my $tgtvol = $vol_copy[$volnum];
             my $status = &vsystem ("neb-replicate --volume $tgtvol $ext_id");
             if ($status) { print STDERR "FAIL SBCK: neb-replicate : $ext_id on $tgtvol\n"; next; }
@@ -92 +131 @@
         }
 
-        # these are done in stage 2:
-        if ($state eq "DUPS:") { $Ndups++; next; }
-        if ($state eq "XBCK:") { $Nxbck++; next; }
-
-        # these are done in stage 3:
-        if ($state eq "CUL1:") { $Ncul1++; next; }
-        if ($state eq "CUL2:") { $Ncul2++; next; }
-        if ($state eq "CUL3:") { $Ncul3++; next; }
+        # multiple copies @ ITC, but none at ATRC. use neb-replicate to make copies
+        # skip if we do not know the state of the md5sum
+        if ($state eq "DUPS:") { 
+            if ($haveMD5 && $goodMD5) { 
+                my $status = &vsystem ("neb-replicate $ext_id");
+                if ($status) { print STDERR "FAIL DUPS: neb-replicate : $ext_id\n"; next; }
+                my ($newfile, $newhost, $newxattr) = &get_new_instance ($ext_id);
+                unless ($newxattr == 1) { die "new instance is not at ATRC? $ext_id : $newfile : $newhost : $newxattr\n"; }
+                open (OUTFILE, ">>$topdir/$dateword/$newhost.inst.$stage.txt");
+                print OUTFILE "$newfile $datast $md5sum\n";
+                close (OUTFILE);
+            }
+            $Ndups++; 
+            next; 
+        }
+
+        # multiple copies @ ATRC, but none @ ITC: neb-replicate will create a copy at ITC iff we supply the target volume
+        # skip if we do not know the state of the md5sum
+        if ($state eq "XBCK:") { 
+            if ($haveMD5 && $goodMD5) {
+                my $tgtvol = $vol_copy[$volnum];
+                my $status = &vsystem ("neb-replicate --volume $tgtvol $ext_id");
+                if ($status) { print STDERR "FAIL XBCK: neb-replicate : $ext_id on $tgtvol\n"; next; }
+                my ($newfile, $newhost, $newxattr) = &get_new_instance ($ext_id);
+                unless ($newxattr == 0) { die "new instance is not at ITC? $ext_id : $newfile : $newhost : $newxattr\n"; }
+                open (OUTFILE, ">>$topdir/$dateword/$newhost.inst.$stage.txt");
+                print OUTFILE "$newfile $datast $md5sum\n";
+                close (OUTFILE);
+            }
+            $Nxbck++; 
+            next; 
+        }
+
+        # one copy at ITC: neb-replicate will create a copy at ATRC
+        # skip if we do not know the state of the md5sum
+        if (($state eq "CUL1:") || ($state eq "CUL2:") || ($state eq "CUL3:")) { 
+
+            if ($haveMD5 && $goodMD5) {
+                # get info for all instances of this chip
+                # returns a reference to the array of instance data
+                # also strips neb://hostname.volnum/ from nebname and returns as ext_id
+                my ($inforows, $ext_id) =  &get_chip_instances ($ext_id);
+
+                my ($main_keep, $back_keep, $cull_list_r) = &check_instance_xattr ($inforows, $ext_id);
+
+                if ($main_keep eq "") { print STDERR "FAIL $state: missing ITC copy for $ext_id\n";  next; }
+                if ($back_keep eq "") { print STDERR "FAIL $state: missing ATRC copy for $ext_id\n"; next; }
+
+                my $NcullTotal = @$cull_list_r;
+                foreach my $instance (@$cull_list_r) {
+                    my ($volname) = $instance =~ m|file:///data/([0-9a-zA-Z]*\.[0-9])/\S*|;
+                    unless (defined $vol_good{$volname}) { die "ERROR: invalid volume name $volname\n"; }
+                    # print STDERR "neb-cull $state: $ext_id $volname ($main_keep $back_keep)\n";
+                    my $status = &vsystem ("neb-cull --volume $volname $ext_id");
+                    if ($status) { print STDERR "FAIL CULL: neb-replicate : $ext_id\n"; next; }
+                }
+                # print STDERR "Ncull Total $NcullTotal\n";
+            }
+
+            # these are done in stage 3:
+            if ($state eq "CUL1:") { $Ncul1++; next; }
+            if ($state eq "CUL2:") { $Ncul2++; next; }
+            if ($state eq "CUL3:") { $Ncul3++; next; }
+        }
     }
     $volnum ++;
@@ -136 +231 @@
 }
 
-# get the list of chip neb paths for one chip
-sub get_neb_volumes {
+sub get_chip_instances {
+
+    my $ext_id = $_[0];
+    
+    my $sql_chip_instances = "SELECT uri, host, available, xattr, cab_id from instance join storage_object using (so_id) join volume using (vol_id) where ext_id = '$ext_id'";
+    my $rowref   = $neb_dbh->selectall_arrayref( $sql_chip_instances );
+    
+    return ($rowref, $ext_id);
+}
+
+# select 1 instance with xattr = 0, 1 with xattr = 1.
+# the rest should be culled.
+
+sub check_instance_xattr {
+
+    my $rowref  = $_[0];
+    my $ext_id  = $_[1];
+
+    my @inforows = @$rowref;
+    my $Nrows    = @inforows;
+
+    if ($DEBUG > 5) { print "found $Nrows instances\n"; }
+
+    # get user_copies for this storage_object
+    my $main_keep = "";
+    my $back_keep = "";
+    my @cull_list = ();
+
+    # xattr = 0 means instance @ ITC  (main)
+    # xattr = 1 means instance @ ATRC (back)
+    for (my $j = 0; $j < $Nrows; $j ++) {
+        my $valref = $inforows[$j];
+        my @values = @$valref;
+        
+        # the number of returned values per row must match the query above (in get_chip_instances)
+        my $Nvalue = @values;
+        if ($Nvalue != 5) {
+            print STDERR "ERROR : INVALID DB XATTR RESPONSE: $ext_id\n";
+            next;
+        }
+
+        my $instance = $values[0];
+        my $hostname = $values[1];
+        my $xattr    = $values[3];
+        
+        if (($xattr == 0) && ($main_keep eq "")) { $main_keep = $instance; next; }
+        if (($xattr == 1) && ($back_keep eq "")) { $back_keep = $instance; next; }
+
+        # if we make it here, we already have an instance at this location
+        push (@cull_list, $instance);
+    }
+    return ($main_keep, $back_keep, \@cull_list);
+}
+
+# get the list of neb hosts to copy data to at ITC
+sub get_neb_volumes_copy {
 
     my @volarray;
@@ -164 +313 @@
 }
 
+# get the list of valid neb hosts (ITC or ATRC)
+sub get_neb_volumes_good {
+
+    my %volarray;
+
+    my $sql_neb_volumes = "select name from mountedvol where ((xattr = 0) or (xattr = 1))";
+    if ($DEBUG > 4) { print "sql: $sql_neb_volumes\n"; }
+
+    my $rowref   = $neb_dbh->selectall_arrayref( $sql_neb_volumes );
+    my @tmparray = @$rowref;
+    my $Nrows    = @tmparray;
+
+    if ($DEBUG > 4) { print "found $Nrows possible volumes\n"; }
+
+    foreach my $valref (@tmparray) {
+        my @values = @$valref;
+        # there should only be a single value (only name selected above)
+        my $Nvalues = @values;
+        if ($Nvalues == 0) { print "ERROR: missing uri\n"; next; }
+        if ($Nvalues  > 1) { print "ERROR: too many uri entries?\n"; next; }
+        
+        $volarray{$values[0]} = 1;
+        if ($DEBUG > 4) { print "$values[0]\n"; }
+    }
+    
+    return %volarray;
+}
+
 # read the list of chip_ids:
 sub load_chip_list {
@@ -194 +371 @@
     pod2usage( -msg => "Cannot determine target date, use --dateobs YYYY/MM/DD", -exitval => 2) unless defined $dateobs;
     pod2usage( -msg => "Provide a list of chips with --chiplist", -exitval => 2) unless defined $chiplist;
-    pod2usage( -msg => "Define the stage (v0, v1, etc) with --stage", -exitval => 2) unless defined $stage;
+    pod2usage( -msg => "Define the stage (0, 1, etc) with --stage", -exitval => 2) unless defined $stage;
     pod2usage( -msg => "Specify the top-level output directory with --topdir", -exitval => 2) unless defined $topdir;
     
+    # check that stage is an integer
+    unless (is_integer ($stage)) { die "stage must be a non-negative integer\n"; }
+    unless ($stage >= 0) { die "stage must be a non-negative integer\n"; }
+
     # check the date format (require YYYY/MM/DD and use that from 00:00:00 to 23:59:59 or take start and end?)
 
@@ -248 +429 @@
 sub vsystem {
     print STDERR "@_\n";
-    #return 0;
+    # return 0;
     my $status = system ("@_");
     return $status;
 }
+
+sub is_integer {
+   defined $_[0] && $_[0] =~ /^[+-]?\d+$/;
+}
+