-# Copyright (C) 2018 all contributors <meta@public-inbox.org>
+# Copyright (C) 2018-2020 all contributors <meta@public-inbox.org>
# License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
# This interface wraps and mimics PublicInbox::Import
+# Used to write to V2 inboxes (see L<public-inbox-v2-format(5)>).
package PublicInbox::V2Writable;
use strict;
-use warnings;
-use base qw(PublicInbox::Lock);
-use PublicInbox::SearchIdxPart;
-use PublicInbox::MIME;
+use v5.10.1;
+use parent qw(PublicInbox::Lock);
+use PublicInbox::SearchIdxShard;
+use PublicInbox::Eml;
use PublicInbox::Git;
use PublicInbox::Import;
-use PublicInbox::MID qw(mids);
-use PublicInbox::ContentId qw(content_id content_digest);
-use PublicInbox::Inbox;
+use PublicInbox::MID qw(mids references);
+use PublicInbox::ContentHash qw(content_hash content_digest);
+use PublicInbox::InboxWritable;
use PublicInbox::OverIdx;
use PublicInbox::Msgmap;
-use PublicInbox::Spawn qw(spawn);
-use PublicInbox::SearchIdx;
-use IO::Handle;
+use PublicInbox::Spawn qw(spawn popen_rd);
+use PublicInbox::SearchIdx qw(log2stack crlf_adjust is_ancestor check_size);
+use IO::Handle; # ->autoflush
+use File::Temp ();
+my $OID = qr/[a-f0-9]{40,}/;
# an estimate of the post-packed size to the raw uncompressed size
my $PACKING_FACTOR = 0.4;
-# assume 2 cores if GNU nproc(1) is not available
-sub nproc_parts () {
- my $n = int($ENV{NPROC} || `nproc 2>/dev/null` || 2);
+# SATA storage lags behind what CPUs are capable of, so relying on
+# nproc(1) can be misleading and having extra Xapian shards is a
+# waste of FDs and space. It can also lead to excessive IO latency
+# and slow things down. Users on NVME or other fast storage can
+# use the NPROC env or switches in our script/public-inbox-* programs
+# to increase Xapian shards
+our $NPROC_MAX_DEFAULT = 4;
+
+sub detect_nproc () {
+ # getconf(1) is POSIX, but *NPROCESSORS* vars are not
+ for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
+ `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
+ }
+ for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
+ `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
+ }
+
+ # should we bother with `sysctl hw.ncpu`? Those only give
+ # us total processor count, not online processor count.
+ undef
+}
+
+sub nproc_shards ($) {
+ my ($creat_opt) = @_;
+ my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
+ $n //= $ENV{NPROC};
+ if (!$n) {
+ # assume 2 cores if not detectable or zero
+ state $NPROC_DETECTED = detect_nproc() || 2;
+ $n = $NPROC_DETECTED;
+ $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
+ }
+
# subtract for the main process and git-fast-import
$n -= 1;
$n < 1 ? 1 : $n;
}
-sub count_partitions ($) {
+sub count_shards ($) {
my ($self) = @_;
- my $nparts = 0;
+ my $n = 0;
my $xpfx = $self->{xpfx};
- # always load existing partitions in case core count changes:
- # Also, partition count may change while -watch is running
- # due to -compact
+ # always load existing shards in case core count changes:
+ # Also, shard count may change while -watch is running
+ # due to "xcpdb --reshard"
if (-d $xpfx) {
- foreach my $part (<$xpfx/*>) {
- -d $part && $part =~ m!/\d+\z! or next;
+ my $XapianDatabase;
+ foreach my $shard (<$xpfx/*>) {
+ -d $shard && $shard =~ m!/[0-9]+\z! or next;
+ $XapianDatabase //= do {
+ require PublicInbox::Search;
+ PublicInbox::Search::load_xapian();
+ $PublicInbox::Search::X{Database};
+ };
eval {
- Search::Xapian::Database->new($part)->close;
- $nparts++;
+ $XapianDatabase->new($shard)->close;
+ $n++;
};
}
}
- $nparts;
+ $n;
}
sub new {
+ # $creat may be any true value, or 0/undef. A hashref is true,
+ # and $creat->{nproc} may be set to an integer
my ($class, $v2ibx, $creat) = @_;
- my $dir = $v2ibx->{mainrepo} or die "no mainrepo in inbox\n";
+ $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
+ my $dir = $v2ibx->assert_usable_dir;
unless (-d $dir) {
if ($creat) {
require File::Path;
die "$dir does not exist\n";
}
}
-
- $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
$v2ibx->umask_prepare;
my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
my $self = {
- -inbox => $v2ibx,
+ ibx => $v2ibx,
im => undef, # PublicInbox::Import
parallel => 1,
transact_bytes => 0,
+ total_bytes => 0,
+ current_info => '',
xpfx => $xpfx,
- over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3", 1),
+ over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
lock_path => "$dir/inbox.lock",
# limit each git repo (epoch) to 1GB or so
rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
- last_commit => [], # git repo -> commit
+ last_commit => [], # git epoch -> commit
};
- $self->{partitions} = count_partitions($self) || nproc_parts();
+ $self->{over}->{-no_fsync} = 1 if $v2ibx->{-no_fsync};
+ $self->{shards} = count_shards($self) || nproc_shards($creat);
bless $self, $class;
}
+# public (for now?)
sub init_inbox {
- my ($self, $parallel) = @_;
- $self->{parallel} = $parallel;
+ my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
+ if (defined $shards) {
+ $self->{parallel} = 0 if $shards == 0;
+ $self->{shards} = $shards if $shards > 0;
+ }
$self->idx_init;
+ $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
my $epoch_max = -1;
git_dir_latest($self, \$epoch_max);
+ if (defined $skip_epoch && $epoch_max == -1) {
+ $epoch_max = $skip_epoch;
+ }
$self->git_init($epoch_max >= 0 ? $epoch_max : 0);
$self->done;
}
# returns undef on duplicate or spam
# mimics Import::add and wraps it for v2
sub add {
- my ($self, $mime, $check_cb) = @_;
- $self->{-inbox}->with_umask(sub {
- _add($self, $mime, $check_cb)
- });
+ my ($self, $eml, $check_cb) = @_;
+ $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
+}
+
+# indexes a message, returns true if checkpointing is needed
+sub do_idx ($$$$) {
+ my ($self, $msgref, $mime, $smsg) = @_;
+ $smsg->{bytes} = $smsg->{raw_bytes} + crlf_adjust($$msgref);
+ $self->{over}->add_overview($mime, $smsg);
+ my $idx = idx_shard($self, $smsg->{num} % $self->{shards});
+ $idx->index_raw($msgref, $mime, $smsg);
+ my $n = $self->{transact_bytes} += $smsg->{raw_bytes};
+ $n >= $self->{batch_bytes};
}
sub _add {
# spam check:
if ($check_cb) {
- $mime = $check_cb->($mime) or return;
+ $mime = $check_cb->($mime, $self->{ibx}) or return;
}
# All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
# leaking FDs to it...
$self->idx_init;
- my $mid0;
- my $num = num_for($self, $mime, \$mid0);
+ my ($num, $mid0) = v2_num_for($self, $mime);
defined $num or return; # duplicate
- defined $mid0 or die "BUG: $mid0 undefined\n";
+ defined $mid0 or die "BUG: \$mid0 undefined\n";
my $im = $self->importer;
- my $cmt = $im->add($mime);
+ my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
+ my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
$cmt = $im->get_mark($cmt);
$self->{last_commit}->[$self->{epoch_max}] = $cmt;
- my ($oid, $len, $msgref) = @{$im->{last_object}};
- $self->{over}->add_overview($mime, $len, $num, $oid, $mid0);
- my $nparts = $self->{partitions};
- my $part = $num % $nparts;
- my $idx = $self->idx_part($part);
- $idx->index_raw($len, $msgref, $num, $oid, $mid0, $mime);
- my $n = $self->{transact_bytes} += $len;
- if ($n > (PublicInbox::SearchIdx::BATCH_BYTES * $nparts)) {
+ my $msgref = delete $smsg->{-raw_email};
+ if (do_idx($self, $msgref, $mime, $smsg)) {
$self->checkpoint;
}
$cmt;
}
-sub num_for {
- my ($self, $mime, $mid0) = @_;
- my $mids = mids($mime->header_obj);
+sub v2_num_for {
+ my ($self, $mime) = @_;
+ my $mids = mids($mime);
if (@$mids) {
my $mid = $mids->[0];
my $num = $self->{mm}->mid_insert($mid);
if (defined $num) { # common case
- $$mid0 = $mid;
- return $num;
- };
+ return ($num, $mid);
+ }
# crap, Message-ID is already known, hope somebody just resent:
foreach my $m (@$mids) {
# read-only lookup now safe to do after above barrier
- my $existing = $self->lookup_content($mime, $m);
# easy, don't store duplicates
# note: do not add more diagnostic info here since
# it gets noisy on public-inbox-watch restarts
- return if $existing;
+ return () if content_exists($self, $mime, $m);
+ }
+
+ # AltId may pre-populate article numbers (e.g. X-Mail-Count
+ # or NNTP article number), use that article number if it's
+ # not in Over.
+ my $altid = $self->{ibx}->{altid};
+ if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
+ my $num = $self->{mm}->num_for($mid);
+
+ if (defined $num && !$self->{over}->get_art($num)) {
+ return ($num, $mid);
+ }
}
# very unlikely:
$num = $self->{mm}->mid_insert($m);
if (defined $num) {
warn "alternative <$m> for <$mid> found\n";
- $$mid0 = $m;
- return $num;
+ return ($num, $m);
}
}
}
# none of the existing Message-IDs are good, generate a new one:
- num_for_harder($self, $mime, $mid0);
+ v2_num_for_harder($self, $mime);
}
-sub num_for_harder {
- my ($self, $mime, $mid0) = @_;
+sub v2_num_for_harder {
+ my ($self, $eml) = @_;
- my $hdr = $mime->header_obj;
- my $dig = content_digest($mime);
- $$mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
- my $num = $self->{mm}->mid_insert($$mid0);
+ my $dig = content_digest($eml);
+ my $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
+ my $num = $self->{mm}->mid_insert($mid0);
unless (defined $num) {
# it's hard to spoof the last Received: header
- my @recvd = $hdr->header_raw('Received');
+ my @recvd = $eml->header_raw('Received');
$dig->add("Received: $_") foreach (@recvd);
- $$mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
- $num = $self->{mm}->mid_insert($$mid0);
+ $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
+ $num = $self->{mm}->mid_insert($mid0);
# fall back to a random Message-ID and give up determinism:
until (defined($num)) {
$dig->add(rand);
- $$mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
- warn "using random Message-ID <$$mid0> as fallback\n";
- $num = $self->{mm}->mid_insert($$mid0);
+ $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
+ warn "using random Message-ID <$mid0> as fallback\n";
+ $num = $self->{mm}->mid_insert($mid0);
}
}
- PublicInbox::Import::append_mid($hdr, $$mid0);
- $num;
+ PublicInbox::Import::append_mid($eml, $mid0);
+ ($num, $mid0);
+}
+
+sub idx_shard {
+ my ($self, $shard_i) = @_;
+ $self->{idx_shards}->[$shard_i];
}
-sub idx_part {
- my ($self, $part) = @_;
- $self->{idx_parts}->[$part];
+sub _idx_init { # with_umask callback
+ my ($self, $opt) = @_;
+ $self->lock_acquire unless $opt && $opt->{-skip_lock};
+ $self->{over}->create;
+
+ # xcpdb can change shard count while -watch is idle
+ my $nshards = count_shards($self);
+ $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
+ $self->{batch_bytes} = $opt->{batch_size} //
+ $PublicInbox::SearchIdx::BATCH_BYTES;
+ $self->{batch_bytes} *= $self->{shards} if $self->{parallel};
+
+ # need to create all shards before initializing msgmap FD
+ # idx_shards must be visible to all forked processes
+ my $max = $self->{shards} - 1;
+ my $idx = $self->{idx_shards} = [];
+ push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
+
+ # Now that all subprocesses are up, we can open the FDs
+ # for SQLite:
+ my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
+ "$self->{ibx}->{inboxdir}/msgmap.sqlite3",
+ $self->{ibx}->{-no_fsync} ? 2 : 1);
+ $mm->{dbh}->begin_work;
}
# idempotent
sub idx_init {
- my ($self) = @_;
- return if $self->{idx_parts};
- my $ibx = $self->{-inbox};
+ my ($self, $opt) = @_;
+ return if $self->{idx_shards};
+ my $ibx = $self->{ibx};
# do not leak read-only FDs to child processes, we only have these
# FDs for duplicate detection so they should not be
# frequently activated.
- delete $ibx->{$_} foreach (qw(git mm search));
-
- if ($self->{parallel}) {
- pipe(my ($r, $w)) or die "pipe failed: $!";
- $self->{bnote} = [ $r, $w ];
- $w->autoflush(1);
- }
+ delete @$ibx{qw(mm search)};
+ $ibx->git->cleanup;
+
+ $self->{parallel} = 0 if ($ibx->{indexlevel}//'') eq 'basic';
+ if ($self->{parallel}) {
+ pipe(my ($r, $w)) or die "pipe failed: $!";
+ # pipe for barrier notifications doesn't need to be big,
+ # 1031: F_SETPIPE_SZ
+ fcntl($w, 1031, 4096) if $^O eq 'linux';
+ $self->{bnote} = [ $r, $w ];
+ $w->autoflush(1);
+ }
- my $over = $self->{over};
$ibx->umask_prepare;
- $ibx->with_umask(sub {
- $self->lock_acquire;
- $over->create;
-
- # -compact can change partition count while -watch is idle
- my $nparts = count_partitions($self);
- if ($nparts && $nparts != $self->{partitions}) {
- $self->{partitions} = $nparts;
- }
-
- # need to create all parts before initializing msgmap FD
- my $max = $self->{partitions} - 1;
-
- # idx_parts must be visible to all forked processes
- my $idx = $self->{idx_parts} = [];
- for my $i (0..$max) {
- push @$idx, PublicInbox::SearchIdxPart->new($self, $i);
- }
-
- # Now that all subprocesses are up, we can open the FDs
- # for SQLite:
- my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
- "$self->{-inbox}->{mainrepo}/msgmap.sqlite3", 1);
- $mm->{dbh}->begin_work;
- });
+ $ibx->with_umask(\&_idx_init, $self, $opt);
}
-sub purge_oids {
- my ($self, $purge) = @_; # $purge = { $object_id => 1, ... }
+# returns an array mapping [ epoch => latest_commit ]
+# latest_commit may be undef if nothing was done to that epoch
+# $replace_map = { $object_id => $strref, ... }
+sub _replace_oids ($$$) {
+ my ($self, $mime, $replace_map) = @_;
$self->done;
- my $pfx = "$self->{-inbox}->{mainrepo}/git";
- my $purges = [];
- foreach my $i (0..$self->{epoch_max}) {
- my $git = PublicInbox::Git->new("$pfx/$i.git");
+ my $pfx = "$self->{ibx}->{inboxdir}/git";
+ my $rewrites = []; # epoch => commit
+ my $max = $self->{epoch_max};
+
+ unless (defined($max)) {
+ defined(my $latest = git_dir_latest($self, \$max)) or return;
+ $self->{epoch_max} = $max;
+ }
+
+ foreach my $i (0..$max) {
+ my $git_dir = "$pfx/$i.git";
+ -d $git_dir or next;
+ my $git = PublicInbox::Git->new($git_dir);
my $im = $self->import_init($git, 0, 1);
- $purges->[$i] = $im->purge_oids($purge);
+ $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
+ $im->done;
}
- $purges;
+ $rewrites;
}
-sub content_ids ($) {
+sub content_hashes ($) {
my ($mime) = @_;
- my @cids = ( content_id($mime) );
+ my @chashes = ( content_hash($mime) );
+ # We still support Email::MIME, here, and
# Email::MIME->as_string doesn't always round-trip, so we may
- # use a second content_id
- my $rt = content_id(PublicInbox::MIME->new(\($mime->as_string)));
- push @cids, $rt if $cids[0] ne $rt;
- \@cids;
+ # use a second content_hash
+ my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
+ push @chashes, $rt if $chashes[0] ne $rt;
+ \@chashes;
}
sub content_matches ($$) {
- my ($cids, $existing) = @_;
- my $cid = content_id($existing);
- foreach (@$cids) {
- return 1 if $_ eq $cid
+ my ($chashes, $existing) = @_;
+ my $chash = content_hash($existing);
+ foreach (@$chashes) {
+ return 1 if $_ eq $chash
}
0
}
-sub remove_internal {
- my ($self, $mime, $cmt_msg, $purge) = @_;
+# used for removing or replacing (purging)
+sub rewrite_internal ($$;$$$) {
+ my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
$self->idx_init;
- my $im = $self->importer unless $purge;
+ my ($im, $need_reindex, $replace_map);
+ if ($sref) {
+ $replace_map = {}; # oid => sref
+ $need_reindex = [] if $new_eml;
+ } else {
+ $im = $self->importer;
+ }
my $over = $self->{over};
- my $cids = content_ids($mime);
- my $parts = $self->{idx_parts};
- my $mm = $self->{mm};
- my $removed;
- my $mids = mids($mime->header_obj);
+ my $chashes = content_hashes($old_eml);
+ my $removed = [];
+ my $mids = mids($old_eml);
# We avoid introducing new blobs into git since the raw content
# can be slightly different, so we do not need the user-supplied
- # message now that we have the mids and content_id
- $mime = undef;
+ # message now that we have the mids and content_hash
+ $old_eml = undef;
my $mark;
foreach my $mid (@$mids) {
- my %gone;
+ my %gone; # num => [ smsg, $mime, raw ]
my ($id, $prev);
while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
my $msg = get_blob($self, $smsg);
next; # continue
}
my $orig = $$msg;
- my $cur = PublicInbox::MIME->new($msg);
- if (content_matches($cids, $cur)) {
- $smsg->{mime} = $cur;
- $gone{$smsg->{num}} = [ $smsg, \$orig ];
+ my $cur = PublicInbox::Eml->new($msg);
+ if (content_matches($chashes, $cur)) {
+ $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
}
}
my $n = scalar keys %gone;
join(',', sort keys %gone), "\n";
}
foreach my $num (keys %gone) {
- my ($smsg, $orig) = @{$gone{$num}};
- $mm->num_delete($num);
+ my ($smsg, $mime, $orig) = @{$gone{$num}};
# $removed should only be set once assuming
# no bugs in our deduplication code:
- $removed = $smsg;
+ $removed = [ undef, $mime, $smsg ];
my $oid = $smsg->{blob};
- if ($purge) {
- $purge->{$oid} = 1;
+ if ($replace_map) {
+ $replace_map->{$oid} = $sref;
} else {
($mark, undef) = $im->remove($orig, $cmt_msg);
+ $removed->[0] = $mark;
}
$orig = undef;
- $self->unindex_oid_remote($oid, $mid);
+ if ($need_reindex) { # ->replace
+ push @$need_reindex, $smsg;
+ } else { # ->purge or ->remove
+ $self->{mm}->num_delete($num);
+ }
+ unindex_oid_remote($self, $oid, $mid);
}
}
my $cmt = $im->get_mark($mark);
$self->{last_commit}->[$self->{epoch_max}] = $cmt;
}
- if ($purge && scalar keys %$purge) {
- return purge_oids($self, $purge);
+ if ($replace_map && scalar keys %$replace_map) {
+ my $rewrites = _replace_oids($self, $new_eml, $replace_map);
+ return { rewrites => $rewrites, need_reindex => $need_reindex };
}
- $removed;
+ defined($mark) ? $removed : undef;
}
+# public (see PublicInbox::Import->remove), but note the 3rd element
+# (retval[2]) is not part of the stable API shared with Import->remove
sub remove {
- my ($self, $mime, $cmt_msg) = @_;
- $self->{-inbox}->with_umask(sub {
- remove_internal($self, $mime, $cmt_msg);
- });
+ my ($self, $eml, $cmt_msg) = @_;
+ my $r = $self->{ibx}->with_umask(\&rewrite_internal,
+ $self, $eml, $cmt_msg);
+ defined($r) && defined($r->[0]) ? @$r: undef;
}
+sub _replace ($$;$$) {
+ my ($self, $old_eml, $new_eml, $sref) = @_;
+ my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
+ my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
+ $self, $old_eml, undef, $new_eml, $sref) or return;
+
+ my $rewrites = $rewritten->{rewrites};
+ # ->done is called if there are rewrites since we gc+prune from git
+ $self->idx_init if @$rewrites;
+
+ for my $i (0..$#$rewrites) {
+ defined(my $cmt = $rewrites->[$i]) or next;
+ $self->{last_commit}->[$i] = $cmt;
+ }
+ $rewritten;
+}
+
+# public
sub purge {
my ($self, $mime) = @_;
- my $purges = $self->{-inbox}->with_umask(sub {
- remove_internal($self, $mime, undef, {});
- });
- $self->idx_init if @$purges; # ->done is called on purges
- for my $i (0..$#$purges) {
- defined(my $cmt = $purges->[$i]) or next;
- $self->{last_commit}->[$i] = $cmt;
+ my $rewritten = _replace($self, $mime, undef, \'') or return;
+ $rewritten->{rewrites}
+}
+
+# returns the git object_id of $fh, does not write the object to FS
+sub git_hash_raw ($$) {
+ my ($self, $raw) = @_;
+ # grab the expected OID we have to reindex:
+ pipe(my($in, $w)) or die "pipe: $!";
+ my $git_dir = $self->{ibx}->git->{git_dir};
+ my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
+ my $r = popen_rd($cmd, undef, { 0 => $in });
+ print $w $$raw or die "print \$w: $!";
+ close $w or die "close \$w: $!";
+ local $/ = "\n";
+ chomp(my $oid = <$r>);
+ close $r or die "git hash-object failed: $?";
+ $oid =~ /\A$OID\z/ or die "OID not expected: $oid";
+ $oid;
+}
+
+sub _check_mids_match ($$$) {
+ my ($old_list, $new_list, $hdrs) = @_;
+ my %old_mids = map { $_ => 1 } @$old_list;
+ my %new_mids = map { $_ => 1 } @$new_list;
+ my @old = keys %old_mids;
+ my @new = keys %new_mids;
+ my $err = "$hdrs may not be changed when replacing\n";
+ die $err if scalar(@old) != scalar(@new);
+ delete @new_mids{@old};
+ delete @old_mids{@new};
+ die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
+}
+
+# Changing Message-IDs or References with ->replace isn't supported.
+# The rules for dealing with messages with multiple or conflicting
+# Message-IDs are pretty complex and rethreading hasn't been fully
+# implemented, yet.
+sub check_mids_match ($$) {
+ my ($old, $new) = @_;
+ _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
+ _check_mids_match(references($old), references($new),
+ 'References/In-Reply-To');
+}
+
+# public
+sub replace ($$$) {
+ my ($self, $old_mime, $new_mime) = @_;
+
+ check_mids_match($old_mime, $new_mime);
+
+ # mutt will always add Content-Length:, Status:, Lines: when editing
+ PublicInbox::Import::drop_unwanted_headers($new_mime);
+
+ my $raw = $new_mime->as_string;
+ my $expect_oid = git_hash_raw($self, \$raw);
+ my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
+ my $need_reindex = $rewritten->{need_reindex};
+
+ # just in case we have bugs in deduplication code:
+ my $n = scalar(@$need_reindex);
+ if ($n > 1) {
+ my $list = join(', ', map {
+ "$_->{num}: <$_->{mid}>"
+ } @$need_reindex);
+ warn <<"";
+W: rewritten $n messages matching content of original message (expected: 1).
+W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
+W: $list
+
+ }
+
+ # make sure we really got the OID:
+ my ($blob, $type, $bytes) = $self->{ibx}->git->check($expect_oid);
+ $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
+
+ # don't leak FDs to Xapian:
+ $self->{ibx}->git->cleanup;
+
+ # reindex modified messages:
+ for my $smsg (@$need_reindex) {
+ my $new_smsg = bless {
+ blob => $blob,
+ raw_bytes => $bytes,
+ num => $smsg->{num},
+ mid => $smsg->{mid},
+ }, 'PublicInbox::Smsg';
+ my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
+ $new_smsg->populate($new_mime, $sync);
+ do_idx($self, \$raw, $new_mime, $new_smsg);
}
- $purges;
+ $rewritten->{rewrites};
}
-sub last_commit_part ($$;$) {
+sub last_epoch_commit ($$;$) {
my ($self, $i, $cmt) = @_;
my $v = PublicInbox::Search::SCHEMA_VERSION();
$self->{mm}->last_commit_xap($v, $i, $cmt);
foreach my $i (0..$epoch_max) {
defined(my $cmt = $last_commit->[$i]) or next;
$last_commit->[$i] = undef;
- last_commit_part($self, $i, $cmt);
+ last_epoch_commit($self, $i, $cmt);
}
}
my $bnote = $self->{bnote} or return;
my $r = $bnote->[0];
while (scalar keys %$barrier) {
- defined(my $l = $r->getline) or die "EOF on barrier_wait: $!";
+ defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
$l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
- delete $barrier->{$1} or die "bad part[$1] on barrier wait";
+ delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
}
}
+# public
sub checkpoint ($;$) {
my ($self, $wait) = @_;
$im->checkpoint;
}
}
- my $parts = $self->{idx_parts};
- if ($parts) {
+ my $shards = $self->{idx_shards};
+ if ($shards) {
my $dbh = $self->{mm}->{dbh};
# SQLite msgmap data is second in importance
# Now deal with Xapian
if ($wait) {
- my $barrier = $self->barrier_init(scalar @$parts);
+ my $barrier = $self->barrier_init(scalar @$shards);
- # each partition needs to issue a barrier command
- $_->remote_barrier for @$parts;
+ # each shard needs to issue a barrier command
+ $_->shard_barrier for @$shards;
- # wait for each Xapian partition
+ # wait for each Xapian shard
$self->barrier_wait($barrier);
} else {
- $_->remote_commit for @$parts;
+ $_->shard_commit for @$shards;
}
# last_commit is special, don't commit these until
- # remote partitions are done:
+ # remote shards are done:
$dbh->begin_work;
set_last_commits($self);
$dbh->commit;
$dbh->begin_work;
}
+ $self->{total_bytes} += $self->{transact_bytes};
$self->{transact_bytes} = 0;
}
# issue a write barrier to ensure all data is visible to other processes
# and read-only ops. Order of data importance is: git > SQLite > Xapian
+# public
sub barrier { checkpoint($_[0], 1) };
+# true if locked and active
+sub active { !!$_[0]->{im} }
+
+# public
sub done {
my ($self) = @_;
- my $im = delete $self->{im};
- $im->done if $im; # PublicInbox::Import::done
- checkpoint($self);
- my $mm = delete $self->{mm};
- $mm->{dbh}->commit if $mm;
- my $parts = delete $self->{idx_parts};
- if ($parts) {
- $_->remote_close for @$parts;
- }
- $self->{over}->disconnect;
+ my $err = '';
+ if (my $im = delete $self->{im}) {
+ eval { $im->done }; # PublicInbox::Import::done
+ $err .= "import done: $@\n" if $@;
+ }
+ if (!$err) {
+ eval { checkpoint($self) };
+ $err .= "checkpoint: $@\n" if $@;
+ }
+ if (my $mm = delete $self->{mm}) {
+ my $m = $err ? 'rollback' : 'commit';
+ eval { $mm->{dbh}->$m };
+ $err .= "msgmap $m: $@\n" if $@;
+ }
+ my $shards = delete $self->{idx_shards};
+ if ($shards) {
+ for (@$shards) {
+ eval { $_->shard_close };
+ $err .= "shard close: $@\n" if $@;
+ }
+ }
+ eval { $self->{over}->dbh_close };
+ $err .= "over close: $@\n" if $@;
delete $self->{bnote};
- $self->{transact_bytes} = 0;
- $self->lock_release if $parts;
+ my $nbytes = $self->{total_bytes};
+ $self->{total_bytes} = 0;
+ $self->lock_release(!!$nbytes) if $shards;
+ $self->{ibx}->git->cleanup;
+ die $err if $err;
}
-sub git_init {
+sub fill_alternates ($$) {
my ($self, $epoch) = @_;
- my $pfx = "$self->{-inbox}->{mainrepo}/git";
- my $git_dir = "$pfx/$epoch.git";
- my @cmd = (qw(git init --bare -q), $git_dir);
- PublicInbox::Import::run_die(\@cmd);
-
- my $all = "$self->{-inbox}->{mainrepo}/all.git";
- unless (-d $all) {
- @cmd = (qw(git init --bare -q), $all);
- PublicInbox::Import::run_die(\@cmd);
- @cmd = (qw/git config/, "--file=$all/config",
- 'repack.writeBitmaps', 'true');
- PublicInbox::Import::run_die(\@cmd);
- }
-
- @cmd = (qw/git config/, "--file=$git_dir/config",
- 'include.path', '../../all.git/config');
- PublicInbox::Import::run_die(\@cmd);
- my $alt = "$all/objects/info/alternates";
- my %alts;
- my @add;
+ my $pfx = "$self->{ibx}->{inboxdir}/git";
+ my $all = "$self->{ibx}->{inboxdir}/all.git";
+ PublicInbox::Import::init_bare($all) unless -d $all;
+ my $info_dir = "$all/objects/info";
+ my $alt = "$info_dir/alternates";
+ my (%alt, $new);
+ my $mode = 0644;
if (-e $alt) {
open(my $fh, '<', $alt) or die "open < $alt: $!\n";
- %alts = map { chomp; $_ => 1 } (<$fh>);
+ $mode = (stat($fh))[2] & 07777;
+
+ # we assign a sort score to every alternate and favor
+ # the newest (highest numbered) one because loose objects
+ # require scanning epochs and only the latest epoch is
+ # expected to see loose objects
+ my $score;
+ my $other = 0; # in case admin adds non-epoch repos
+ %alt = map {;
+ if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
+ $score = $1 + 0;
+ } else {
+ $score = --$other;
+ }
+ $_ => $score;
+ } split(/\n+/, do { local $/; <$fh> });
}
+
foreach my $i (0..$epoch) {
my $dir = "../../git/$i.git/objects";
- push @add, $dir if !$alts{$dir} && -d "$pfx/$i.git";
- }
- return $git_dir unless @add;
- open my $fh, '>>', $alt or die "open >> $alt: $!\n";
- foreach my $dir (@add) {
- print $fh "$dir\n" or die "print >> $alt: $!\n";
+ if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
+ $alt{$dir} = $i;
+ $new = 1;
+ }
}
- close $fh or die "close $alt: $!\n";
+ return unless $new;
+
+ my $fh = File::Temp->new(TEMPLATE => 'alt-XXXXXXXX', DIR => $info_dir);
+ my $tmp = $fh->filename;
+ print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"
+ or die "print $tmp: $!\n";
+ chmod($mode, $fh) or die "fchmod $tmp: $!\n";
+ close $fh or die "close $tmp $!\n";
+ rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
+ $fh->unlink_on_destroy(0);
+}
+
+sub git_init {
+ my ($self, $epoch) = @_;
+ my $git_dir = "$self->{ibx}->{inboxdir}/git/$epoch.git";
+ PublicInbox::Import::init_bare($git_dir);
+ my @cmd = (qw/git config/, "--file=$git_dir/config",
+ 'include.path', '../../all.git/config');
+ PublicInbox::Import::run_die(\@cmd);
+ fill_alternates($self, $epoch);
$git_dir
}
sub git_dir_latest {
my ($self, $max) = @_;
$$max = -1;
- my $pfx = "$self->{-inbox}->{mainrepo}/git";
+ my $pfx = "$self->{ibx}->{inboxdir}/git";
return unless -d $pfx;
my $latest;
opendir my $dh, $pfx or die "opendir $pfx: $!\n";
while (defined(my $git_dir = readdir($dh))) {
- $git_dir =~ m!\A(\d+)\.git\z! or next;
+ $git_dir =~ m!\A([0-9]+)\.git\z! or next;
if ($1 > $$max) {
$$max = $1;
$latest = "$pfx/$git_dir";
if (defined $latest) {
my $git = PublicInbox::Git->new($latest);
my $packed_bytes = $git->packed_bytes;
- if ($packed_bytes >= $self->{rotate_bytes}) {
+ my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
+
+ if ($unpacked_bytes >= $self->{rotate_bytes}) {
$epoch = $max + 1;
} else {
$self->{epoch_max} = $max;
sub import_init {
my ($self, $git, $packed_bytes, $tmp) = @_;
- my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox});
+ my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx});
$im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
- $im->{want_object_info} = 1;
$im->{lock_path} = undef;
$im->{path_type} = 'v2';
$self->{im} = $im unless $tmp;
# XXX experimental
sub diff ($$$) {
my ($mid, $cur, $new) = @_;
- use File::Temp qw(tempfile);
- my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
+ my $ah = File::Temp->new(TEMPLATE => 'email-cur-XXXXXXXX', TMPDIR => 1);
print $ah $cur->as_string or die "print: $!";
- close $ah or die "close: $!";
- my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1);
+ $ah->flush or die "flush: $!";
PublicInbox::Import::drop_unwanted_headers($new);
+ my $bh = File::Temp->new(TEMPLATE => 'email-new-XXXXXXXX', TMPDIR => 1);
print $bh $new->as_string or die "print: $!";
- close $bh or die "close: $!";
- my $cmd = [ qw(diff -u), $an, $bn ];
+ $bh->flush or die "flush: $!";
+ my $cmd = [ qw(diff -u), $ah->filename, $bh->filename ];
print STDERR "# MID conflict <$mid>\n";
my $pid = spawn($cmd, undef, { 1 => 2 });
- defined $pid or die "diff failed to spawn $!";
waitpid($pid, 0) == $pid or die "diff did not finish";
- unlink($an, $bn);
}
sub get_blob ($$) {
return $msg if $msg;
}
# older message, should be in alternates
- my $ibx = $self->{-inbox};
- $ibx->msg_by_smsg($smsg);
+ $self->{ibx}->msg_by_smsg($smsg);
}
-sub lookup_content {
+sub content_exists ($$$) {
my ($self, $mime, $mid) = @_;
my $over = $self->{over};
- my $cids = content_ids($mime);
+ my $chashes = content_hashes($mime);
my ($id, $prev);
while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
my $msg = get_blob($self, $smsg);
warn "broken smsg for $mid\n";
next;
}
- my $cur = PublicInbox::MIME->new($msg);
- if (content_matches($cids, $cur)) {
- $smsg->{mime} = $cur;
- return $smsg;
- }
-
+ my $cur = PublicInbox::Eml->new($msg);
+ return 1 if content_matches($chashes, $cur);
# XXX DEBUG_DIFF is experimental and may be removed
diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
sub atfork_child {
my ($self) = @_;
- my $fh = delete $self->{reindex_pipe};
- close $fh if $fh;
- if (my $parts = $self->{idx_parts}) {
- $_->atfork_child foreach @$parts;
+ if (my $shards = $self->{idx_shards}) {
+ $_->atfork_child foreach @$shards;
}
if (my $im = $self->{im}) {
$im->atfork_child;
$self->{bnote}->[1];
}
-sub mark_deleted {
- my ($self, $D, $git, $oid) = @_;
- my $msgref = $git->cat_file($oid);
- my $mime = PublicInbox::MIME->new($$msgref);
- my $mids = mids($mime->header_obj);
- my $cid = content_id($mime);
- foreach my $mid (@$mids) {
- $D->{"$mid\0$cid"} = $oid;
+sub reindex_checkpoint ($$) {
+ my ($self, $sync) = @_;
+
+ $self->{ibx}->git->cleanup; # *async_wait
+ ${$sync->{need_checkpoint}} = 0;
+ my $mm_tmp = $sync->{mm_tmp};
+ $mm_tmp->atfork_prepare if $mm_tmp;
+ $self->done; # release lock
+
+ if (my $pr = $sync->{-opt}->{-progress}) {
+ $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}}));
}
+
+ # allow -watch or -mda to write...
+ $self->idx_init($sync->{-opt}); # reacquire lock
+ $mm_tmp->atfork_parent if $mm_tmp;
}
-sub reindex_oid {
- my ($self, $mm_tmp, $D, $git, $oid, $regen, $reindex) = @_;
- my $len;
- my $msgref = $git->cat_file($oid, \$len);
- my $mime = PublicInbox::MIME->new($$msgref);
- my $mids = mids($mime->header_obj);
- my $cid = content_id($mime);
+sub index_oid { # cat_async callback
+ my ($bref, $oid, $type, $size, $arg) = @_;
+ return if $size == 0; # purged
+ my ($num, $mid0);
+ my $eml = PublicInbox::Eml->new($$bref);
+ my $mids = mids($eml);
+ my $chash = content_hash($eml);
+ my $self = $arg->{v2w};
+
+ if (scalar(@$mids) == 0) {
+ warn "E: $oid has no Message-ID, skipping\n";
+ return;
+ }
- # get the NNTP article number we used before, highest number wins
- # and gets deleted from mm_tmp;
- my $mid0;
- my $num = -1;
- my $del = 0;
- foreach my $mid (@$mids) {
- $del += delete($D->{"$mid\0$cid"}) ? 1 : 0;
- my $n = $mm_tmp->num_for($mid);
- if (defined $n && $n > $num) {
- $mid0 = $mid;
- $num = $n;
+ # {unindexed} is unlikely
+ if ((my $unindexed = $arg->{unindexed}) && scalar(@$mids) == 1) {
+ $num = delete($unindexed->{$mids->[0]});
+ if (defined $num) {
+ $mid0 = $mids->[0];
$self->{mm}->mid_set($num, $mid0);
+ delete($arg->{unindexed}) if !keys(%$unindexed);
}
}
- if (!defined($mid0) && $regen && !$del) {
- $num = $$regen--;
- die "BUG: ran out of article numbers\n" if $num <= 0;
- my $mm = $self->{mm};
- foreach my $mid (reverse @$mids) {
- if ($mm->mid_set($num, $mid) == 1) {
- $mid0 = $mid;
+ if (!defined($num)) { # reuse if reindexing (or duplicates)
+ my $over = $self->{over};
+ for my $mid (@$mids) {
+ ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid);
+ last if defined $num;
+ }
+ }
+ $mid0 //= do { # is this a number we got before?
+ $num = $arg->{mm_tmp}->num_for($mids->[0]);
+ defined($num) ? $mids->[0] : undef;
+ };
+ if (!defined($num)) {
+ for (my $i = $#$mids; $i >= 1; $i--) {
+ $num = $arg->{mm_tmp}->num_for($mids->[$i]);
+ if (defined($num)) {
+ $mid0 = $mids->[$i];
last;
}
}
- if (!defined($mid0)) {
- my $id = '<' . join('> <', @$mids) . '>';
- warn "Message-ID $id unusable for $num\n";
- foreach my $mid (@$mids) {
- defined(my $n = $mm->num_for($mid)) or next;
- warn "#$n previously mapped for <$mid>\n";
+ }
+ if (defined($num)) {
+ $arg->{mm_tmp}->num_delete($num);
+ } else { # never seen
+ $num = $self->{mm}->mid_insert($mids->[0]);
+ if (defined($num)) {
+ $mid0 = $mids->[0];
+ } else { # rare, try the rest of them, backwards
+ for (my $i = $#$mids; $i >= 1; $i--) {
+ $num = $self->{mm}->mid_insert($mids->[$i]);
+ if (defined($num)) {
+ $mid0 = $mids->[$i];
+ last;
+ }
}
}
}
-
- if (!defined($mid0) || $del) {
- if (!defined($mid0) && $del) { # expected for deletes
- $num = $$regen--;
- $self->{mm}->num_highwater($num) unless $reindex;
- return
- }
-
- my $id = '<' . join('> <', @$mids) . '>';
- defined($mid0) or
- warn "Skipping $id, no article number found\n";
- if ($del && defined($mid0)) {
- warn "$id was deleted $del " .
- "time(s) but mapped to article #$num\n";
- }
+ if (!defined($num)) {
+ warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n";
return;
-
}
- $mm_tmp->mid_delete($mid0) or
- die "failed to delete <$mid0> for article #$num\n";
-
- $self->{over}->add_overview($mime, $len, $num, $oid, $mid0);
- my $nparts = $self->{partitions};
- my $part = $num % $nparts;
- my $idx = $self->idx_part($part);
- $idx->index_raw($len, $msgref, $num, $oid, $mid0, $mime);
- my $n = $self->{transact_bytes} += $len;
- if ($n > (PublicInbox::SearchIdx::BATCH_BYTES * $nparts)) {
- $git->cleanup;
- $mm_tmp->atfork_prepare;
- $self->done; # release lock
- # allow -watch or -mda to write...
- $self->idx_init; # reacquire lock
- $mm_tmp->atfork_parent;
+ ++${$arg->{nr}};
+ my $smsg = bless {
+ raw_bytes => $size,
+ num => $num,
+ blob => $oid,
+ mid => $mid0,
+ }, 'PublicInbox::Smsg';
+ $smsg->populate($eml, $arg);
+ if (do_idx($self, $bref, $eml, $smsg)) {
+ ${$arg->{need_checkpoint}} = 1;
}
}
# only update last_commit for $i on reindex iff newer than current
-sub update_last_commit {
+sub update_last_commit ($$$$) {
my ($self, $git, $i, $cmt) = @_;
- my $last = last_commit_part($self, $i);
+ my $last = last_epoch_commit($self, $i);
if (defined $last && is_ancestor($git, $last, $cmt)) {
my @cmd = (qw(rev-list --count), "$last..$cmt");
chomp(my $n = $git->qx(@cmd));
return if $n ne '' && $n == 0;
}
- last_commit_part($self, $i, $cmt);
+ last_epoch_commit($self, $i, $cmt);
}
-sub git_dir_n ($$) { "$_[0]->{-inbox}->{mainrepo}/git/$_[1].git" }
+sub git_dir_n ($$) { "$_[0]->{ibx}->{inboxdir}/git/$_[1].git" }
-sub last_commits {
+sub last_commits ($$) {
my ($self, $epoch_max) = @_;
my $heads = [];
for (my $i = $epoch_max; $i >= 0; $i--) {
- $heads->[$i] = last_commit_part($self, $i);
+ $heads->[$i] = last_epoch_commit($self, $i);
}
$heads;
}
-*is_ancestor = *PublicInbox::SearchIdx::is_ancestor;
+# returns a revision range for git-log(1)
+sub log_range ($$$$$) {
+ my ($self, $sync, $git, $i, $tip) = @_;
+ my $opt = $sync->{-opt};
+ my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
+ my $cur = $sync->{ranges}->[$i] or do {
+ $pr->("$i.git indexing all of $tip\n") if $pr;
+ return $tip; # all of it
+ };
-sub index_prepare {
- my ($self, $opts, $epoch_max, $ranges) = @_;
- my $regen_max = 0;
- my $head = $self->{-inbox}->{ref_head} || 'refs/heads/master';
- for (my $i = $epoch_max; $i >= 0; $i--) {
- die "already indexing!\n" if $self->{index_pipe};
- my $git_dir = git_dir_n($self, $i);
- -d $git_dir or next; # missing parts are fine
- my $git = PublicInbox::Git->new($git_dir);
- chomp(my $tip = $git->qx('rev-parse', $head));
- my $range;
- if (defined(my $cur = $ranges->[$i])) {
- $range = "$cur..$tip";
- if (is_ancestor($git, $cur, $tip)) { # common case
- my $n = $git->qx(qw(rev-list --count), $range);
- chomp($n);
- if ($n == 0) {
- $ranges->[$i] = undef;
- next;
- }
- } else {
- warn <<"";
+ # fast equality check to avoid (v)fork+execve overhead
+ if ($cur eq $tip) {
+ $sync->{ranges}->[$i] = undef;
+ return;
+ }
+
+ my $range = "$cur..$tip";
+ $pr->("$i.git checking contiguity... ") if $pr;
+ if (is_ancestor($git, $cur, $tip)) { # common case
+ $pr->("OK\n") if $pr;
+ my $n = $git->qx(qw(rev-list --count), $range);
+ chomp($n);
+ if ($n == 0) {
+ $sync->{ranges}->[$i] = undef;
+ $pr->("$i.git has nothing new\n") if $pr;
+ return; # nothing to do
+ }
+ $pr->("$i.git has $n changes since $cur\n") if $pr;
+ } else {
+ $pr->("FAIL\n") if $pr;
+ warn <<"";
discontiguous range: $range
-Rewritten history? (in $git_dir)
-
- my $base = $git->qx('merge-base', $tip, $cur);
- chomp $base;
- if ($base) {
- $range = "$base..$tip";
- warn "found merge-base: $base\n"
- } else {
- $range = $tip;
- warn <<"";
-discarding history at $cur
+Rewritten history? (in $git->{git_dir})
- }
- warn <<"";
-reindexing $git_dir starting at
+ chomp(my $base = $git->qx('merge-base', $tip, $cur));
+ if ($base) {
+ $range = "$base..$tip";
+ warn "found merge-base: $base\n"
+ } else {
+ $range = $tip;
+ warn "discarding history at $cur\n";
+ }
+ warn <<"";
+reindexing $git->{git_dir} starting at
$range
- $self->{"unindex-range.$i"} = "$base..$cur";
- }
- } else {
- $range = $tip; # all of it
+ $sync->{unindex_range}->{$i} = "$base..$cur";
+ }
+ $range;
+}
+
+sub sync_prepare ($$$) {
+ my ($self, $sync, $epoch_max) = @_;
+ my $pr = $sync->{-opt}->{-progress};
+ my $regen_max = 0;
+ my $head = $self->{ibx}->{ref_head} || 'refs/heads/master';
+
+ # reindex stops at the current heads and we later rerun index_sync
+ # without {reindex}
+ my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
+
+ for (my $i = $epoch_max; $i >= 0; $i--) {
+ my $git_dir = git_dir_n($self, $i);
+ -d $git_dir or next; # missing epochs are fine
+ my $git = PublicInbox::Git->new($git_dir);
+ if ($reindex_heads) {
+ $head = $reindex_heads->[$i] or next;
}
- $ranges->[$i] = $range;
+ chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
+ next if $?; # new repo
+ my $range = log_range($self, $sync, $git, $i, $tip) or next;
# can't use 'rev-list --count' if we use --diff-filter
- my $fh = $git->popen(qw(log --pretty=tformat:%H
- --no-notes --no-color --no-renames
- --diff-filter=AM), $range, '--', 'm');
- ++$regen_max while <$fh>;
+ $pr->("$i.git counting $range ... ") if $pr;
+ # Don't bump num_highwater on --reindex by using {D}.
+ # We intentionally do NOT use {D} in the non-reindex case
+ # because we want NNTP article number gaps from unindexed
+ # messages to show up in mirrors, too.
+ $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR
+ my $stk = log2stack($sync, $git, $range, $self->{ibx});
+ my $nr = $stk ? $stk->num_records : 0;
+ $pr->("$nr\n") if $pr;
+ $sync->{stacks}->[$i] = $stk if $stk;
+ $regen_max += $nr;
+ }
+
+ # XXX this should not happen unless somebody bypasses checks in
+ # our code and blindly injects "d" file history into git repos
+ if (my @leftovers = keys %{delete($sync->{D}) // {}}) {
+ warn('W: unindexing '.scalar(@leftovers)." leftovers\n");
+ my $arg = { v2w => $self };
+ my $all = $self->{ibx}->git;
+ for my $oid (@leftovers) {
+ $oid = unpack('H*', $oid);
+ $self->{current_info} = "leftover $oid";
+ $all->cat_async($oid, \&unindex_oid, $arg);
+ }
+ $all->cat_async_wait;
+ }
+ if (!$regen_max && !keys(%{$self->{unindex_range}})) {
+ $sync->{-regen_fmt} = "%u/?\n";
+ return 0;
}
- \$regen_max;
+
+ # reindex should NOT see new commits anymore, if we do,
+ # it's a problem and we need to notice it via die()
+ my $pad = length($regen_max) + 1;
+ $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
+ $sync->{nr} = \(my $nr = 0);
+ return -1 if $sync->{reindex};
+ $regen_max + $self->{mm}->num_highwater() || 0;
}
-sub unindex_oid_remote {
+sub unindex_oid_remote ($$$) {
my ($self, $oid, $mid) = @_;
- $_->remote_remove($oid, $mid) foreach @{$self->{idx_parts}};
- $self->{over}->remove_oid($oid, $mid);
+ my @removed = $self->{over}->remove_oid($oid, $mid);
+ for my $num (@removed) {
+ my $idx = idx_shard($self, $num % $self->{shards});
+ $idx->shard_remove($oid, $num);
+ }
}
-sub unindex_oid {
- my ($self, $git, $oid) = @_;
- my $msgref = $git->cat_file($oid);
- my $mime = PublicInbox::MIME->new($msgref);
- my $mids = mids($mime->header_obj);
- $mime = $msgref = undef;
+sub unindex_oid ($$;$) { # git->cat_async callback
+ my ($bref, $oid, $type, $size, $sync) = @_;
+ my $self = $sync->{v2w};
+ my $unindexed = $sync->{in_unindex} ? $sync->{unindexed} : undef;
+ my $mm = $self->{mm};
+ my $mids = mids(PublicInbox::Eml->new($bref));
+ undef $$bref;
my $over = $self->{over};
foreach my $mid (@$mids) {
my %gone;
my ($id, $prev);
while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
$gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
- 1; # continue
}
- my $n = scalar keys %gone;
- next unless $n;
+ my $n = scalar(keys(%gone)) or next;
if ($n > 1) {
warn "BUG: multiple articles linked to $oid\n",
join(',',sort keys %gone), "\n";
}
foreach my $num (keys %gone) {
- $self->{unindexed}->{$_}++;
- $self->{mm}->num_delete($num);
+ if ($unindexed) {
+ my $mid0 = $mm->mid_for($num);
+ $unindexed->{$mid0} = $num;
+ }
+ $mm->num_delete($num);
}
- $self->unindex_oid_remote($oid, $mid);
+ unindex_oid_remote($self, $oid, $mid);
}
}
-my $x40 = qr/[a-f0-9]{40}/;
-sub unindex {
- my ($self, $opts, $git, $unindex_range) = @_;
- my $un = $self->{unindexed} ||= {}; # num => removal count
- my $before = scalar keys %$un;
+# this is rare, it only happens when we get discontiguous history in
+# a mirror because the source used -purge or -edit
+sub unindex ($$$$) {
+ my ($self, $sync, $git, $unindex_range) = @_;
+ my $unindexed = $sync->{unindexed} //= {}; # $mid0 => $num
+ my $before = scalar keys %$unindexed;
+ # order does not matter, here:
my @cmd = qw(log --raw -r
--no-notes --no-color --no-abbrev --no-renames);
- my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $unindex_range);
+ my $fh = $git->popen(@cmd, $unindex_range);
+ my $all = $self->{ibx}->git;
+ local $sync->{in_unindex} = 1;
while (<$fh>) {
- /\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o or next;
- $self->unindex_oid($git, $1);
+ /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
+ $all->cat_async($1, \&unindex_oid, $sync);
}
- delete $self->{reindex_pipe};
- $fh = undef;
+ close $fh or die "git log failed: \$?=$?";
+ $all->cat_async_wait;
- return unless $opts->{prune};
- my $after = scalar keys %$un;
+ return unless $sync->{-opt}->{prune};
+ my $after = scalar keys %$unindexed;
return if $before == $after;
# ensure any blob can not longer be accessed via dumb HTTP
PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
- qw(-c gc.reflogExpire=now gc --prune=all)]);
+ qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
}
+sub sync_ranges ($$$) {
+ my ($self, $sync, $epoch_max) = @_;
+ my $reindex = $sync->{reindex};
+
+ return last_commits($self, $epoch_max) unless $reindex;
+ return [] if ref($reindex) ne 'HASH';
+
+ my $ranges = $reindex->{from}; # arrayref;
+ if (ref($ranges) ne 'ARRAY') {
+ die 'BUG: $reindex->{from} not an ARRAY';
+ }
+ $ranges;
+}
+
+sub index_xap_only { # git->cat_async callback
+ my ($bref, $oid, $type, $size, $smsg) = @_;
+ my $self = $smsg->{v2w};
+ my $idx = idx_shard($self, $smsg->{num} % $self->{shards});
+ $smsg->{raw_bytes} = $size;
+ $idx->index_raw($bref, undef, $smsg);
+ $self->{transact_bytes} += $size;
+}
+
+sub index_xap_step ($$$;$) {
+ my ($self, $sync, $beg, $step) = @_;
+ my $end = $sync->{art_end};
+ return if $beg > $end; # nothing to do
+
+ $step //= $self->{shards};
+ my $ibx = $self->{ibx};
+ if (my $pr = $sync->{-opt}->{-progress}) {
+ $pr->("Xapian indexlevel=$ibx->{indexlevel} ".
+ "$beg..$end (% $step)\n");
+ }
+ for (my $num = $beg; $num <= $end; $num += $step) {
+ my $smsg = $ibx->over->get_art($num) or next;
+ $smsg->{v2w} = $self;
+ $ibx->git->cat_async($smsg->{blob}, \&index_xap_only, $smsg);
+ if ($self->{transact_bytes} >= $self->{batch_bytes}) {
+ ${$sync->{nr}} = $num;
+ reindex_checkpoint($self, $sync);
+ }
+ }
+}
+
+sub index_epoch ($$$) {
+ my ($self, $sync, $i) = @_;
+
+ my $git_dir = git_dir_n($self, $i);
+ -d $git_dir or return; # missing epochs are fine
+ my $git = PublicInbox::Git->new($git_dir);
+ if (my $unindex_range = delete $sync->{unindex_range}->{$i}) { # rare
+ unindex($self, $sync, $git, $unindex_range);
+ }
+ defined(my $stk = $sync->{stacks}->[$i]) or return;
+ $sync->{stacks}->[$i] = undef;
+ my $all = $self->{ibx}->git;
+ while (my ($f, $at, $ct, $oid) = $stk->pop_rec) {
+ $self->{current_info} = "$i.git $oid";
+ if ($f eq 'm') {
+ my $arg = { %$sync, autime => $at, cotime => $ct };
+ if ($sync->{max_size}) {
+ $all->check_async($oid, \&check_size, $arg);
+ } else {
+ $all->cat_async($oid, \&index_oid, $arg);
+ }
+ } elsif ($f eq 'd') {
+ $all->cat_async($oid, \&unindex_oid, $sync);
+ }
+ if (${$sync->{need_checkpoint}}) {
+ reindex_checkpoint($self, $sync);
+ }
+ }
+ $all->check_async_wait;
+ $all->cat_async_wait;
+ update_last_commit($self, $git, $i, $stk->{latest_cmt});
+}
+
+sub xapian_only {
+ my ($self, $opt, $sync, $art_beg) = @_;
+ my $seq = $opt->{sequential_shard};
+ $art_beg //= 0;
+ local $self->{parallel} = 0 if $seq;
+ $self->idx_init($opt); # acquire lock
+ if (my $art_end = $self->{ibx}->mm->max) {
+ $sync //= {
+ need_checkpoint => \(my $bool = 0),
+ -opt => $opt,
+ v2w => $self,
+ nr => \(my $nr = 0),
+ -regen_fmt => "%u/?\n",
+ };
+ $sync->{art_end} = $art_end;
+ if ($seq || !$self->{parallel}) {
+ my $shard_end = $self->{shards} - 1;
+ for my $i (0..$shard_end) {
+ index_xap_step($self, $sync, $art_beg + $i);
+ if ($i != $shard_end) {
+ reindex_checkpoint($self, $sync);
+ }
+ }
+ } else { # parallel (maybe)
+ index_xap_step($self, $sync, $art_beg, 1);
+ }
+ }
+ $self->{ibx}->git->cat_async_wait;
+ $self->done;
+}
+
+# public, called by public-inbox-index
sub index_sync {
- my ($self, $opts) = @_;
- $opts ||= {};
+ my ($self, $opt) = @_;
+ $opt //= $_[1] //= {};
+ goto \&xapian_only if $opt->{xapian_only};
+
+ my $pr = $opt->{-progress};
my $epoch_max;
my $latest = git_dir_latest($self, \$epoch_max);
return unless defined $latest;
- $self->idx_init; # acquire lock
- my $mm_tmp = $self->{mm}->tmp_clone;
- my $reindex = $opts->{reindex};
- my $ranges = $reindex ? [] : $self->last_commits($epoch_max);
-
- my $high = $self->{mm}->num_highwater();
- my $regen = $self->index_prepare($opts, $epoch_max, $ranges);
- $$regen += $high if $high;
- my $D = {}; # "$mid\0$cid" => $oid
- my @cmd = qw(log --raw -r --pretty=tformat:%H
- --no-notes --no-color --no-abbrev --no-renames);
- # work backwards through history
- my $last_commit = [];
- for (my $i = $epoch_max; $i >= 0; $i--) {
- my $git_dir = git_dir_n($self, $i);
- die "already reindexing!\n" if delete $self->{reindex_pipe};
- -d $git_dir or next; # missing parts are fine
- my $git = PublicInbox::Git->new($git_dir);
- my $unindex = delete $self->{"unindex-range.$i"};
- $self->unindex($opts, $git, $unindex) if $unindex;
- defined(my $range = $ranges->[$i]) or next;
- my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $range);
- my $cmt;
- while (<$fh>) {
- if (/\A$x40$/o && !defined($cmt)) {
- chomp($cmt = $_);
- } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o) {
- $self->reindex_oid($mm_tmp, $D, $git, $1,
- $regen, $reindex);
- } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) {
- $self->mark_deleted($D, $git, $1);
- }
+ my $seq = $opt->{sequential_shard};
+ my $art_beg; # the NNTP article number we start xapian_only at
+ my $idxlevel = $self->{ibx}->{indexlevel};
+ local $self->{ibx}->{indexlevel} = 'basic' if $seq;
+
+ $self->idx_init($opt); # acquire lock
+ fill_alternates($self, $epoch_max);
+ $self->{over}->rethread_prepare($opt);
+ my $sync = {
+ need_checkpoint => \(my $bool = 0),
+ unindex_range => {}, # EPOCH => oid_old..oid_new
+ reindex => $opt->{reindex},
+ -opt => $opt,
+ v2w => $self,
+ };
+ $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
+ if (sync_prepare($self, $sync, $epoch_max)) {
+ # tmp_clone seems to fail if inside a transaction, so
+ # we rollback here (because we opened {mm} for reading)
+ # Note: we do NOT rely on DBI transactions for atomicity;
+ # only for batch performance.
+ $self->{mm}->{dbh}->rollback;
+ $self->{mm}->{dbh}->begin_work;
+ $sync->{mm_tmp} =
+ $self->{mm}->tmp_clone($self->{ibx}->{inboxdir});
+
+ # xapian_only works incrementally w/o --reindex
+ if ($seq && !$opt->{reindex}) {
+ $art_beg = $sync->{mm_tmp}->max;
+ $art_beg++ if defined($art_beg);
}
- $fh = undef;
- delete $self->{reindex_pipe};
- $self->update_last_commit($git, $i, $cmt) if defined $cmt;
}
-
- # unindex is required for leftovers if "deletes" affect messages
- # in a previous fetch+index window:
- if (scalar keys %$D) {
- my $git = $self->{-inbox}->git;
- $self->unindex_oid($git, $_) for values %$D;
- $git->cleanup;
+ if ($sync->{max_size} = $opt->{max_size}) {
+ $sync->{index_oid} = \&index_oid;
}
+ # work forwards through history
+ index_epoch($self, $sync, $_) for (0..$epoch_max);
+ $self->{over}->rethread_done($opt);
$self->done;
+
+ if (my $nr = $sync->{nr}) {
+ my $pr = $sync->{-opt}->{-progress};
+ $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr;
+ }
+
+ # deal with Xapian shards sequentially
+ if ($seq && delete($sync->{mm_tmp})) {
+ $self->{ibx}->{indexlevel} = $idxlevel;
+ xapian_only($self, $opt, $sync, $art_beg);
+ }
+
+ # --reindex on the command-line
+ if ($opt->{reindex} && !ref($opt->{reindex}) && $idxlevel ne 'basic') {
+ $self->lock_acquire;
+ my $s0 = PublicInbox::SearchIdx->new($self->{ibx}, 0, 0);
+ if (my $xdb = $s0->idx_acquire) {
+ my $n = $xdb->get_metadata('has_threadid');
+ $xdb->set_metadata('has_threadid', '1') if $n ne '1';
+ }
+ $s0->idx_release;
+ $self->lock_release;
+ }
+
+ # reindex does not pick up new changes, so we rerun w/o it:
+ if ($opt->{reindex}) {
+ my %again = %$opt;
+ $sync = undef;
+ delete @again{qw(rethread reindex -skip_lock)};
+ index_sync($self, \%again);
+ }
}
1;