X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FLeiMirror.pm;h=e284f55deb6133cb25eaaf0113426ff7517434d6;hb=9b285533321b533a5257fd071e4200723c37aa8b;hp=0a93ed440c868e73d375bd65cbc09c958bec2367;hpb=cfabf09dabe2f6780e2e3d29b44a2e8cccab8b77;p=public-inbox.git diff --git a/lib/PublicInbox/LeiMirror.pm b/lib/PublicInbox/LeiMirror.pm index 0a93ed44..e284f55d 100644 --- a/lib/PublicInbox/LeiMirror.pm +++ b/lib/PublicInbox/LeiMirror.pm @@ -3,8 +3,7 @@ # "lei add-external --mirror" support (also "public-inbox-clone"); package PublicInbox::LeiMirror; -use strict; -use v5.10.1; +use v5.12; use parent qw(PublicInbox::IPC); use IO::Uncompress::Gunzip qw(gunzip $GunzipError); use IO::Compress::Gzip qw(gzip $GzipError); @@ -19,22 +18,28 @@ use PublicInbox::Config; use PublicInbox::Inbox; use PublicInbox::LeiCurl; use PublicInbox::OnDestroy; +use Digest::SHA qw(sha256_hex sha1_hex); +use POSIX qw(strftime); our $LIVE; # pid => callback +our $FGRP_TODO; # objstore -> [ fgrp mirror objects ] +our $TODO; # reference => [ non-fgrp mirror objects ] + +sub keep_going ($) { + $LIVE && (!$_[0]->{lei}->{child_error} || + $_[0]->{lei}->{opt}->{'keep-going'}); +} sub _wq_done_wait { # dwaitpid callback (via wq_eof) my ($arg, $pid) = @_; my ($mrr, $lei) = @$arg; - my $f = "$mrr->{dst}/mirror.done"; if ($?) { $lei->child_error($?); - } elsif (!$mrr->{dry_run} && !unlink($f)) { - warn("unlink($f): $!\n") unless $!{ENOENT}; - } else { + } elsif (!$lei->{child_error}) { if (!$mrr->{dry_run} && $lei->{cmd} ne 'public-inbox-clone') { - # calls _finish_add_external - $lei->lazy_cb('add-external', '_finish_' - )->($lei, $mrr->{dst}); + require PublicInbox::LeiAddExternal; + PublicInbox::LeiAddExternal::_finish_add_external( + $lei, $mrr->{dst}); } $lei->qerr("# mirrored $mrr->{src} => $mrr->{dst}"); } @@ -43,7 +48,7 @@ sub _wq_done_wait { # dwaitpid callback (via wq_eof) # for old installations without manifest.js.gz sub try_scrape { - my ($self) = @_; + my ($self, $fallback_manifest) = @_; my $uri = URI->new($self->{src}); my $lei = $self->{lei}; my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return; @@ -54,9 +59,17 @@ sub try_scrape { close($fh) or return $lei->child_error($?, "@$cmd failed"); # we grep with URL below, we don't want Subject/From headers - # making us clone random URLs + # making us clone random URLs. This assumes remote instances + # prior to public-inbox 1.7.0 + # 5b96edcb1e0d8252 (www: move mirror instructions to /text/, 2021-08-28) my @html = split(/
/, $html); my @urls = ($html[-1] =~ m!\bgit clone --mirror ([a-z\+]+://\S+)!g); + if (!@urls && $fallback_manifest) { + warn <as_string; chop($url) eq '/' or die "BUG: $uri not canonicalized"; @@ -67,8 +80,9 @@ sub try_scrape { my ($n) = (m!/([0-9]+)\z!); $n => [ URI->new($_), '' ] } @v2_urls; # uniq - clone_v2($self, \%v2_epochs); - reap_live() while keys(%$LIVE); + clone_v2_prep($self, \%v2_epochs); + delete local $lei->{opt}->{epoch}; + clone_all($self); return; } @@ -98,21 +112,36 @@ sub clone_cmd { ($lei->{opt}->{jobs} // 1) > 1; push @cmd, '-v' if $lei->{opt}->{verbose}; # XXX any other options to support? - # --reference is tricky with multiple epochs... + # --reference is tricky with multiple epochs, but handled + # automatically if using manifest.js.gz @cmd; } -sub ft_rename ($$$) { - my ($ft, $dst, $open_mode) = @_; - my $fn = $ft->filename; - my @st = stat($dst); +sub ft_rename ($$$;$) { + my ($ft, $dst, $open_mode, $fh) = @_; + my @st = stat($fh // $dst); my $mode = @st ? ($st[2] & 07777) : ($open_mode & ~umask); - chmod($mode, $ft) or croak "E: chmod $fn: $!"; + chmod($mode, $ft) or croak "E: chmod($ft): $!"; require File::Copy; - File::Copy::mv($fn, $dst) or croak "E: mv($fn => $ft): $!"; + File::Copy::mv($ft->filename, $dst) or croak "E: mv($ft => $dst): $!"; $ft->unlink_on_destroy(0); } +sub do_reap ($;$) { + my ($self, $jobs) = @_; + $jobs //= $self->{-jobs} //= $self->{lei}->{opt}->{jobs} // 1; + $jobs = 1 if $jobs < 1; + while (keys(%$LIVE) >= $jobs) { + my $pid = waitpid(-1, 0) // die "waitpid(-1): $!"; + if (my $x = delete $LIVE->{$pid}) { + my $cb = shift @$x; + $cb->(@$x) if $cb; + } else { + warn "reaped unknown PID=$pid ($?)\n"; + } + } +} + sub _get_txt_start { # non-fatal my ($self, $endpoint, $fini) = @_; my $uri = URI->new($self->{cur_src} // $self->{src}); @@ -125,8 +154,7 @@ sub _get_txt_start { # non-fatal my $opt = { 0 => $lei->{0}, 1 => $lei->{1}, 2 => $lei->{2} }; my $cmd = $self->{curl}->for_uri($lei, $uri, qw(--compressed -R -o), $ft->filename); - my $jobs = $lei->{opt}->{jobs} // 1; - reap_live() while keys(%$LIVE) >= $jobs; + do_reap($self); $lei->qerr("# @$cmd"); return if $self->{dry_run}; $self->{"-get_txt.$endpoint"} = [ $ft, $cmd, $uri ]; @@ -173,28 +201,19 @@ sub _write_inbox_config { sub set_description ($) { my ($self) = @_; my $dst = $self->{cur_dst} // $self->{dst}; - my $f = "$dst/description"; - open my $fh, '+>>', $f or die "open($f): $!"; - seek($fh, 0, SEEK_SET) or die "seek($f): $!"; - my $d = do { local $/; <$fh> } // die "read($f): $!"; - my $orig = $d; + chomp(my $orig = PublicInbox::Git::try_cat("$dst/description")); + my $d = $orig; while (defined($d) && ($d =~ m!^\(\$INBOX_DIR/description missing\)! || $d =~ /^Unnamed repository/ || $d !~ /\S/)) { $d = delete($self->{'txt.description'}); } - $d //= 'mirror of '.($self->{cur_src} // $self->{src})."\n"; - return if $d eq $orig; - seek($fh, 0, SEEK_SET) or die "seek($f): $!"; - truncate($fh, 0) or die "truncate($f): $!"; - print $fh $d or die "print($f): $!"; - close $fh or die "close($f): $!"; + $d //= 'mirror of '.($self->{cur_src} // $self->{src}); + atomic_write($dst, 'description', $d."\n") if $d ne $orig; } sub index_cloned_inbox { my ($self, $iv) = @_; my $lei = $self->{lei}; - eval { set_description($self) }; - warn $@ if $@; # n.b. public-inbox-clone works w/o (SQLite || Xapian) # lei is useless without Xapian + SQLite @@ -221,8 +240,7 @@ sub index_cloned_inbox { PublicInbox::Admin::progress_prepare($opt, $lei->{2}); PublicInbox::Admin::index_inbox($ibx, undef, $opt); } - return if defined $self->{cur_dst}; - open my $x, '>', "$self->{dst}/mirror.done"; # for _wq_done_wait + return if defined $self->{cur_dst}; # one of many repos to clone } sub run_reap { @@ -234,38 +252,317 @@ sub run_reap { $ret; } -sub start_clone { +sub start_cmd { my ($self, $cmd, $opt, $fini) = @_; - my $jobs = $self->{lei}->{opt}->{jobs} // 1; - reap_live() while keys(%$LIVE) >= $jobs; + do_reap($self); $self->{lei}->qerr("# @$cmd"); return if $self->{dry_run}; - $LIVE->{spawn($cmd, undef, $opt)} = [ \&reap_clone, $self, $cmd, $fini ] + $LIVE->{spawn($cmd, undef, $opt)} = [ \&reap_cmd, $self, $cmd, $fini ] +} + +sub fetch_args ($$) { + my ($lei, $opt) = @_; + my @cmd; # (git --git-dir=...) to be added by caller + $opt->{$_} = $lei->{$_} for (0..2); + # we support "-c $key=$val" for arbitrary git config options + # e.g.: git -c http.proxy=socks5h://127.0.0.1:9050 + push(@cmd, '-c', $_) for @{$lei->{opt}->{c} // []}; + push @cmd, 'fetch'; + push @cmd, '-q' if $lei->{opt}->{quiet} || + ($lei->{opt}->{jobs} // 1) > 1; + push @cmd, '-v' if $lei->{opt}->{verbose}; + push(@cmd, '-p') if $lei->{opt}->{prune}; + @cmd; +} + +sub upr { # feed `git update-ref --stdin -z' verbosely + my ($lei, $w, $op, @rest) = @_; # ($ref, $oid) = @rest + $lei->qerr("# $op @rest") if $lei->{opt}->{verbose}; + print $w "$op ", join("\0", @rest, '') or die "print(w): $!"; +} + +sub fgrp_update { + my ($fgrp) = @_; + return if !keep_going($fgrp); + my $srcfh = delete $fgrp->{srcfh} or return; + my $dstfh = delete $fgrp->{dstfh} or return; + seek($srcfh, SEEK_SET, 0) or die "seek(src): $!"; + seek($dstfh, SEEK_SET, 0) or die "seek(dst): $!"; + my %src = map { chomp; split(/\0/) } (<$srcfh>); + close $srcfh; + my %dst = map { chomp; split(/\0/) } (<$dstfh>); + close $dstfh; + pipe(my ($r, $w)) or die "pipe: $!"; + my $cmd = [ 'git', "--git-dir=$fgrp->{cur_dst}", + qw(update-ref --stdin -z) ]; + my $lei = $fgrp->{lei}; + my $pack = PublicInbox::OnDestroy->new($$, \&pack_dst, $fgrp); + start_cmd($fgrp, $cmd, { 0 => $r, 2 => $lei->{2} }, $pack); + close $r or die "close(r): $!"; + return if $fgrp->{dry_run}; + for my $ref (keys %dst) { + my $new = delete $src{$ref}; + my $old = $dst{$ref}; + if (defined $new) { + $new eq $old or + upr($lei, $w, 'update', $ref, $new, $old); + } else { + upr($lei, $w, 'delete', $ref, $old); + } + } + while (my ($ref, $oid) = each %src) { + upr($lei, $w, 'create', $ref, $oid); + } + close($w) or warn "E: close(update-ref --stdin): $! (need git 1.8.5+)\n"; +} + +sub pack_dst { # packs lightweight satellite repos + my ($fgrp) = @_; + pack_refs($fgrp, $fgrp->{cur_dst}); +} + +sub pack_refs { + my ($self, $git_dir) = @_; + my $cmd = [ 'git', "--git-dir=$git_dir", qw(pack-refs --all --prune) ]; + start_cmd($self, $cmd, { 2 => $self->{lei}->{2} }); +} + +sub fgrpv_done { + my ($fgrpv) = @_; + return if !$LIVE; + my $first = $fgrpv->[0] // die 'BUG: no fgrpv->[0]'; + return if !keep_going($first); + pack_refs($first, $first->{-osdir}); # objstore refs always packed + for my $fgrp (@$fgrpv) { + my $rn = $fgrp->{-remote}; + my %opt = ( 2 => $fgrp->{lei}->{2} ); + + my $update_ref = PublicInbox::OnDestroy->new($$, + \&fgrp_update, $fgrp); + + my $src = [ 'git', "--git-dir=$fgrp->{-osdir}", 'for-each-ref', + "--format=refs/%(refname:lstrip=3)%00%(objectname)", + "refs/remotes/$rn/" ]; + open(my $sfh, '+>', undef) or die "open(src): $!"; + $fgrp->{srcfh} = $sfh; + start_cmd($fgrp, $src, { %opt, 1 => $sfh }, $update_ref); + my $dst = [ 'git', "--git-dir=$fgrp->{cur_dst}", 'for-each-ref', + '--format=%(refname)%00%(objectname)' ]; + open(my $dfh, '+>', undef) or die "open(dst): $!"; + $fgrp->{dstfh} = $dfh; + start_cmd($fgrp, $dst, { %opt, 1 => $dfh }, $update_ref); + } +} + +sub fgrp_fetch_all { + my ($self) = @_; + my $todo = $FGRP_TODO; + $FGRP_TODO = \'BUG on further use'; + keys(%$todo) or return; + + # Rely on the fgrptmp remote groups in the config file rather + # than listing all remotes since the remote name list may exceed + # system argv limits: + my $grp = 'fgrptmp'; + + my @git = (@{$self->{-torsocks}}, 'git'); + my $j = $self->{lei}->{opt}->{jobs}; + my $opt = {}; + my @fetch = do { + local $self->{lei}->{opt}->{jobs} = 1; + (fetch_args($self->{lei}, $opt), + qw(--no-tags --multiple)); + }; + push(@fetch, "-j$j") if $j; + while (my ($osdir, $fgrpv) = each %$todo) { + my $f = "$osdir/config"; + + # clobber group from previous run atomically + my $cmd = ['git', "--git-dir=$osdir", qw(config -f), + $f, '--unset-all', "remotes.$grp"]; + $self->{lei}->qerr("# @$cmd"); + if (!$self->{dry_run}) { + my $pid = spawn($cmd, undef, { 2 => $self->{lei}->{2} }); + waitpid($pid, 0) // die "waitpid: $!"; + die "E: @$cmd: \$?=$?" if ($? && ($? >> 8) != 5); + + # update the config atomically via O_APPEND while + # respecting git-config locking + sysopen(my $lk, "$f.lock", O_CREAT|O_EXCL|O_WRONLY) + or die "open($f.lock): $!"; + open my $fh, '>>', $f or die "open(>>$f): $!"; + $fh->autoflush(1); + my $buf = join('', "[remotes]\n", + map { "\t$grp = $_->{-remote}\n" } @$fgrpv); + print $fh $buf or die "print($f): $!"; + close $fh or die "close($f): $!"; + unlink("$f.lock") or die "unlink($f.lock): $!"; + } + + $cmd = [ @git, "--git-dir=$osdir", @fetch, $grp ]; + my $end = PublicInbox::OnDestroy->new($$, \&fgrpv_done, $fgrpv); + start_cmd($self, $cmd, $opt, $end); + } +} + +# keep this idempotent for future use by public-inbox-fetch +sub forkgroup_prep { + my ($self, $uri) = @_; + $self->{-ent} // return; + my $os = $self->{-objstore} // return; + my $fg = $self->{-ent}->{forkgroup} // return; + my $dir = "$os/$fg.git"; + if (!-d $dir && !$self->{dry_run}) { + PublicInbox::Import::init_bare($dir); + my @cmd = ('git', "--git-dir=$dir", 'config'); + my $opt = { 2 => $self->{lei}->{2} }; + for ('repack.useDeltaIslands=true', + 'pack.island=refs/remotes/([^/]+)/') { + run_die([@cmd, split(/=/, $_, 2)], undef, $opt); + } + } + my $key = $self->{-key} // die 'BUG: no -key'; + my $rn = substr(sha256_hex($key), 0, 16); + if (!-d $self->{cur_dst} && !$self->{dry_run}) { + my $alt = File::Spec->rel2abs("$dir/objects"); + PublicInbox::Import::init_bare($self->{cur_dst}); + my $o = "$self->{cur_dst}/objects"; + my $f = "$o/info/alternates"; + my $l = File::Spec->abs2rel($alt, File::Spec->rel2abs($o)); + open my $fh, '+>>', $f or die "open($f): $!"; + seek($fh, SEEK_SET, 0) or die "seek($f): $!"; + chomp(my @cur = <$fh>); + if (!grep(/\A\Q$l\E\z/, @cur)) { + say $fh $l or die "say($f): $!"; + } + close $fh or die "close($f): $!"; + $f = "$self->{cur_dst}/config"; + open $fh, '+>>', $f or die "open:($f): $!"; + print $fh < $dir, -remote => $rn, -uri => $uri + }, __PACKAGE__; +} + +sub fp_done { + my ($self, $cb, @arg) = @_; + return if !keep_going($self); + my $fh = delete $self->{-show_ref} // die 'BUG: no show-ref output'; + seek($fh, SEEK_SET, 0) or die "seek(show_ref): $!"; + $self->{-ent} // die 'BUG: no -ent'; + my $A = $self->{-ent}->{fingerprint} // die 'BUG: no fingerprint'; + my $B = sha1_hex(do { local $/; <$fh> } // die("read(show_ref): $!")); + return $cb->($self, @arg) if $A ne $B; + $self->{lei}->qerr("# $self->{-key} up-to-date"); +} + +sub cmp_fp_do { + my ($self, $cb, @arg) = @_; + # $cb is either resume_fetch or fgrp_enqueue + $self->{-ent} // return $cb->($self, @arg); + my $new = $self->{-ent}->{fingerprint} // return $cb->($self, @arg); + my $key = $self->{-key} // die 'BUG: no -key'; + if (my $cur_ent = $self->{-local_manifest}->{$key}) { + # runs go_fetch->DESTROY run if eq + return if $cur_ent->{fingerprint} eq $new; + } + my $dst = $self->{cur_dst} // $self->{dst}; + my $cmd = ['git', "--git-dir=$dst", 'show-ref']; + my $opt = { 2 => $self->{lei}->{2} }; + open($opt->{1}, '+>', undef) or die "open(tmp): $!"; + $self->{-show_ref} = $opt->{1}; + my $done = PublicInbox::OnDestroy->new($$, \&fp_done, $self, $cb, @arg); + start_cmd($self, $cmd, $opt, $done); +} + +sub resume_fetch { + my ($self, $uri, $fini) = @_; + return if !keep_going($self); + my $dst = $self->{cur_dst} // $self->{dst}; + my @git = ('git', "--git-dir=$dst"); + my $opt = { 2 => $self->{lei}->{2} }; + my $rn = 'origin'; # configurable? + for ("url=$uri", "fetch=+refs/*:refs/*", 'mirror=true') { + my @kv = split(/=/, $_, 2); + $kv[0] = "remote.$rn.$kv[0]"; + next if $self->{dry_run}; + run_die([@git, 'config', @kv], undef, $opt); + } + my $cmd = [ @{$self->{-torsocks}}, @git, + fetch_args($self->{lei}, $opt), $rn ]; + push @$cmd, '-P' if $self->{lei}->{prune}; # --prune-tags implied + start_cmd($self, $cmd, $opt, $fini); +} + +sub fgrp_enqueue { + my ($fgrp, $end) = @_; # $end calls fgrp_fetch_all + return if !keep_going($fgrp); + my $opt = { 2 => $fgrp->{lei}->{2} }; + # --no-tags is required to avoid conflicts + my $u = $fgrp->{-uri} // die 'BUG: no {-uri}'; + my $rn = $fgrp->{-remote} // die 'BUG: no {-remote}'; + my @cmd = ('git', "--git-dir=$fgrp->{-osdir}", 'config'); + for ("url=$u", "fetch=+refs/*:refs/remotes/$rn/*", 'tagopt=--no-tags') { + my @kv = split(/=/, $_, 2); + $kv[0] = "remote.$rn.$kv[0]"; + $fgrp->{dry_run} ? $fgrp->{lei}->qerr("# @cmd @kv") : + run_die([@cmd, @kv], undef, $opt); + } + push @{$FGRP_TODO->{$fgrp->{-osdir}}}, $fgrp; } sub clone_v1 { - my ($self, $nohang) = @_; + my ($self, $end) = @_; my $lei = $self->{lei}; my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return; my $uri = URI->new($self->{cur_src} // $self->{src}); + my $path = $uri->path; + $path =~ s!/*\z!! and $uri->path($path); defined($lei->{opt}->{epoch}) and die "$uri is a v1 inbox, --epoch is not supported\n"; - my $pfx = $curl->torsocks($lei, $uri) or return; + $self->{-torsocks} //= $curl->torsocks($lei, $uri) or return; my $dst = $self->{cur_dst} // $self->{dst}; my $fini = PublicInbox::OnDestroy->new($$, \&v1_done, $self); - my $cmd = [ @$pfx, clone_cmd($lei, my $opt = {}), "$uri", $dst ]; - my $ref = $self->{-ent} ? $self->{-ent}->{reference} : undef; - defined($ref) && -e "$self->{dst}$ref" and - push @$cmd, '--reference', "$self->{dst}$ref"; - start_clone($self, $cmd, $opt, $fini); - - $lei->{opt}->{'inbox-config'} =~ /\A(?:always|v1)\z/s and + my $resume = -d $dst; + if (my $fgrp = forkgroup_prep($self, $uri)) { + $fgrp->{-fini} = $fini; + $resume ? cmp_fp_do($fgrp, \&fgrp_enqueue, $end) + : fgrp_enqueue($fgrp, $end); + } elsif ($resume) { + cmp_fp_do($self, \&resume_fetch, $uri, $fini); + } else { # normal clone + my $cmd = [ @{$self->{-torsocks}}, + clone_cmd($lei, my $opt = {}), "$uri", $dst ]; + if (defined($self->{-ent})) { + if (defined(my $ref = $self->{-ent}->{reference})) { + -e "$self->{dst}$ref" and + push @$cmd, '--reference', + "$self->{dst}$ref"; + } + } + start_cmd($self, $cmd, $opt, $fini); + } + if (!$self->{-is_epoch} && $lei->{opt}->{'inbox-config'} =~ + /\A(?:always|v1)\z/s) { _get_txt_start($self, '_/text/config/raw', $fini); + } + my $d = $self->{-ent} ? $self->{-ent}->{description} : undef; - defined($d) ? ($self->{'txt.description'} = $d) : + $self->{'txt.description'} = $d if defined $d; + (!defined($d) && !$end) and _get_txt_start($self, 'description', $fini); - reap_live() until ($nohang || !keys(%$LIVE)); # for non-manifest clone + $end or do_reap($self, 1); # for non-manifest clone } sub parse_epochs ($$) { @@ -333,51 +630,146 @@ EOM EOM } close $fh or die "close($f): $!"; - if (defined $ent->{head}) { - $f = "$edst/HEAD"; - open $fh, '>', $f or die "open($f): $!"; - print $fh $ent->{head}, "\n" or die "print($f): $!"; - close $fh or die "close($f): $!"; + my %map = (head => 'HEAD', description => undef); + while (my ($key, $fn) = each %map) { + my $val = $ent->{$key} // next; + $fn //= $key; + $fn = "$edst/$fn"; + open $fh, '>', $fn or die "open($fn): $!"; + print $fh $val, "\n" or die "print($fn): $!"; + close $fh or die "close($fn): $!"; } } -sub reap_clone { # async, called via SIGCHLD +sub reap_cmd { # async, called via SIGCHLD my ($self, $cmd) = @_; my $cerr = $?; $? = 0; # don't let it influence normal exit - if ($cerr) { - kill('TERM', keys %$LIVE); - $self->{lei}->child_error($cerr, "@$cmd failed"); + $self->{lei}->child_error($cerr, "@$cmd failed (\$?=$cerr)") if $cerr; +} + +sub up_fp_done { + my ($self) = @_; + return if !keep_going($self); + my $fh = delete $self->{-show_ref_up} // die 'BUG: no show-ref output'; + seek($fh, SEEK_SET, 0) or die "seek(show_ref): $!"; + $self->{-ent} // die 'BUG: no -ent'; + my $A = $self->{-ent}->{fingerprint} // die 'BUG: no fingerprint'; + my $B = sha1_hex(do { local $/; <$fh> } // die("read(show_ref): $!")); + return if $A eq $B; + $self->{-ent}->{fingerprint} = $B; + push @{$self->{chg}->{fp_mismatch}}, $self->{-key}; +} + +sub atomic_write ($$$) { + my ($dn, $bn, $raw) = @_; + my $ft = File::Temp->new(DIR => $dn, TEMPLATE => "$bn-XXXX"); + print $ft $raw or die "print($ft): $!"; + $ft->flush or die "flush($ft): $!"; + ft_rename($ft, "$dn/$bn", 0666); +} + +# modifies the to-be-written manifest entry, and sets values from it, too +sub update_ent { + my ($self) = @_; + my $key = $self->{-key} // die 'BUG: no -key'; + my $new = $self->{-ent}->{fingerprint}; + my $cur = $self->{-local_manifest}->{$key}->{fingerprint} // "\0"; + my $dst = $self->{cur_dst} // $self->{dst}; + if (defined($new) && $new ne $cur) { + my $cmd = ['git', "--git-dir=$dst", 'show-ref']; + my $opt = { 2 => $self->{lei}->{2} }; + open($opt->{1}, '+>', undef) or die "open(tmp): $!"; + $self->{-show_ref_up} = $opt->{1}; + my $done = PublicInbox::OnDestroy->new($$, \&up_fp_done, $self); + start_cmd($self, $cmd, $opt, $done); + } + $new = $self->{-ent}->{head}; + $cur = $self->{-local_manifest}->{$key}->{head} // "\0"; + if (defined($new) && $new ne $cur) { + # n.b. grokmirror writes raw contents to $dst/HEAD w/o locking + my $cmd = [ 'git', "--git-dir=$dst" ]; + if ($new =~ s/\Aref: //) { + push @$cmd, qw(symbolic-ref HEAD), $new; + } elsif ($new =~ /\A[a-f0-9]{40,}\z/) { + push @$cmd, qw(update-ref --no-deref HEAD), $new; + } else { + undef $cmd; + warn "W: $key: {head} => `$new' not understood\n"; + } + start_cmd($self, $cmd, { 2 => $self->{lei}->{2} }) if $cmd; + } + if (my $symlinks = $self->{-ent}->{symlinks}) { + my $top = File::Spec->rel2abs($self->{dst}); + for my $p (@$symlinks) { + my $ln = "$top/$p"; + $ln =~ tr!/!/!s; + my (undef, $dn, $bn) = File::Spec->splitpath($ln); + File::Path::mkpath($dn); + my $tgt = "$top/$key"; + $tgt = File::Spec->abs2rel($tgt, $dn); + if (lstat($ln)) { + if (-l _) { + next if readlink($ln) eq $tgt; + unlink($ln) or die "unlink($ln): $!"; + } else { + push @{$self->{chg}->{badlink}}, $p; + } + } + symlink($tgt, $ln) or die "symlink($tgt, $ln): $!"; + } + } + if (defined(my $t = $self->{-ent}->{modified})) { + my ($dn, $bn) = ("$dst/info/web", 'last-modified'); + my $orig = PublicInbox::Git::try_cat("$dn/$bn"); + $t = strftime('%F %T', gmtime($t))." +0000\n"; + File::Path::mkpath($dn); + atomic_write($dn, $bn, $t) if $orig ne $t; } + + $new = $self->{-ent}->{owner} // return; + $cur = $self->{-local_manifest}->{$key}->{owner} // "\0"; + return if $cur eq $new; + my $cmd = [ qw(git config -f), "$dst/config", 'gitweb.owner', $new ]; + start_cmd($self, $cmd, { 2 => $self->{lei}->{2} }); } sub v1_done { # called via OnDestroy my ($self) = @_; - return if $self->{dry_run} || !$LIVE; + return if $self->{dry_run} || !keep_going($self); _write_inbox_config($self); my $dst = $self->{cur_dst} // $self->{dst}; - if (defined(my $o = $self->{-ent} ? $self->{-ent}->{owner} : undef)) { - run_die([qw(git config -f), "$dst/config", 'gitweb.owner', $o]); - } + update_ent($self) if $self->{-ent}; my $o = "$dst/objects"; - if (open(my $fh, '<', "$o/info/alternates")) { - chomp(my @l = <$fh>); - for (@l) { $_ = File::Spec->abs2rel($_, $o)."\n" } - my $f = File::Temp->new(TEMPLATE => '.XXXX', DIR => "$o/info"); - print $f @l; - $f->flush or die "flush($f): $!"; - rename($f->filename, "$o/info/alternates") or - die "rename($f, $o/info/alternates): $!"; - $f->unlink_on_destroy(0); + if (open(my $fh, '<', my $fn = "$o/info/alternates")) {; + my $base = File::Spec->rel2abs($o); + my @l = <$fh>; + my $ft; + for (@l) { + next unless m!\A/!; + $_ = File::Spec->abs2rel($_, $base); + $ft //= File::Temp->new(TEMPLATE => '.XXXX', + DIR => "$o/info"); + } + if ($ft) { + print $ft @l or die "print($ft): $!"; + $ft->flush or die "flush($ft): $!"; + ft_rename($ft, $fn, 0666, $fh); + } } + eval { set_description($self) }; + warn $@ if $@; + return if ($self->{-is_epoch} || + $self->{lei}->{opt}->{'inbox-config'} ne 'always'); write_makefile($dst, 1); index_cloned_inbox($self, 1); } sub v2_done { # called via OnDestroy my ($self) = @_; - return if $self->{dry_run} || !$LIVE; + return if $self->{dry_run} || !keep_going($self); my $dst = $self->{cur_dst} // $self->{dst}; + require PublicInbox::Lock; my $lk = bless { lock_path => "$dst/inbox.lock" }, 'PublicInbox::Lock'; my $lck = $lk->lock_for_scope($$); _write_inbox_config($self); @@ -385,40 +777,28 @@ sub v2_done { # called via OnDestroy my $mg = PublicInbox::MultiGit->new($dst, 'all.git', 'git'); $mg->fill_alternates; for my $i ($mg->git_epochs) { $mg->epoch_cfg_set($i) } - my $edst_owner = delete($self->{-owner}) // []; - while (@$edst_owner) { - my ($edst, $o) = splice(@$edst_owner); - run_die [qw(git config -f), "$edst/config", 'gitweb.owner', $o]; - } for my $edst (@{delete($self->{-read_only}) // []}) { my @st = stat($edst) or die "stat($edst): $!"; chmod($st[2] & 0555, $edst) or die "chmod(a-w, $edst): $!"; } write_makefile($dst, 2); undef $lck; # unlock + eval { set_description($self) }; + warn $@ if $@; index_cloned_inbox($self, 2); } -sub reap_live { - my $pid = waitpid(-1, 0) // die "waitpid(-1): $!"; - if (my $x = delete $LIVE->{$pid}) { - my $cb = shift @$x; - $cb->(@$x); - } else { - warn "reaped unknown PID=$pid ($?)\n"; - } -} - -sub clone_v2 ($$;$) { +sub clone_v2_prep ($$;$) { my ($self, $v2_epochs, $m) = @_; # $m => manifest.js.gz hashref my $lei = $self->{lei}; my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return; my $first_uri = (map { $_->[0] } values %$v2_epochs)[0]; - my $pfx = $curl->torsocks($lei, $first_uri) or return; + $self->{-torsocks} //= $curl->torsocks($lei, $first_uri) or return; my $dst = $self->{cur_dst} // $self->{dst}; my $want = parse_epochs($lei->{opt}->{epoch}, $v2_epochs); my $task = $m ? bless { %$self }, __PACKAGE__ : $self; - my (@src_edst, @skip); + my (@skip, $desc); + my $fini = PublicInbox::OnDestroy->new($$, \&v2_done, $task); for my $nr (sort { $a <=> $b } keys %$v2_epochs) { my ($uri, $key) = @{$v2_epochs->{$nr}}; my $src = $uri->as_string; @@ -428,35 +808,40 @@ failed to extract epoch number from $src $1 + 0 == $nr or die "BUG: <$uri> miskeyed $1 != $nr"; $edst .= "/git/$nr.git"; - $m->{$key} // die "BUG: `$key' not in manifest.js.gz"; + my $ent; + if ($m) { + $ent = $m->{$key} // + die("BUG: `$key' not in manifest.js.gz"); + if (defined(my $d = $ent->{description})) { + $d =~ s/ \[epoch [0-9]+\]\z//s; + $desc = $d; + } + } if (!$want || $want->{$nr}) { - push @src_edst, $src, $edst; - my $o = $m->{$key}->{owner}; - push(@{$task->{-owner}}, $edst, $o) if defined($o); + my $etask = bless { %$task, -key => $key }, __PACKAGE__; + $etask->{-ent} = $ent; # may have {reference} + $etask->{cur_src} = $src; + $etask->{cur_dst} = $edst; + $etask->{-is_epoch} = $fini; + my $ref = $ent->{reference} // ''; + push @{$TODO->{$ref}}, $etask; + $self->{any_want}->{$key} = 1; } else { # create a placeholder so users only need to chmod +w - init_placeholder($src, $edst, $m->{$key}); + init_placeholder($src, $edst, $ent); push @{$task->{-read_only}}, $edst; push @skip, $key; } } # filter out the epochs we skipped - $self->{-culled_manifest} = 1 if delete(@$m{@skip}); + $self->{chg}->{manifest} = 1 if $m && delete(@$m{@skip}); (!$self->{dry_run} && !-d $dst) and File::Path::mkpath($dst); - require PublicInbox::Lock; - my $fini = PublicInbox::OnDestroy->new($$, \&v2_done, $task); - $lei->{opt}->{'inbox-config'} =~ /\A(?:always|v2)\z/s and _get_txt_start($task, '_/text/config/raw', $fini); - _get_txt_start($task, 'description', $fini); - - my @cmd = clone_cmd($lei, my $opt = {}); - while (@src_edst && !$lei->{child_error}) { - my $cmd = [ @$pfx, @cmd, splice(@src_edst, 0, 2) ]; - start_clone($self, $cmd, $opt, $fini); - } + defined($desc) ? ($task->{'txt.description'} = $desc) : + _get_txt_start($task, 'description', $fini); } sub decode_manifest ($$$) { @@ -471,6 +856,19 @@ sub decode_manifest ($$$) { $m; } +sub load_current_manifest ($) { + my ($self) = @_; + my $fn = $self->{-manifest} // return; + if (open(my $fh, '<', $fn)) { + decode_manifest($fh, $fn, $fn); + } elsif ($!{ENOENT}) { # non-fatal, we can just do it slowly + warn "open($fn): $!\n" if !$self->{-initial_clone}; + undef; + } else { + die "open($fn): $!\n"; + } +} + sub multi_inbox ($$$) { my ($self, $path, $m) = @_; my $incl = $self->{lei}->{opt}->{include}; @@ -488,21 +886,21 @@ sub multi_inbox ($$$) { my $n = scalar(keys %$v2) + scalar(@v1); my @orig = defined($incl // $excl) ? (keys %$v2, @v1) : (); if (defined $incl) { - my $re = '(?:'.join('|', map { - $self->{lei}->glob2re($_) // qr/\A\Q$_\E\z/ - } @$incl).')'; + my $re = '(?:'.join('\\z|', map { + $self->{lei}->glob2re($_) // qr/\A\Q$_\E/ + } @$incl).'\\z)'; my @gone = delete @$v2{grep(!/$re/, keys %$v2)}; - delete @$m{map { @$_ } @gone} and $self->{-culled_manifest} = 1; - delete @$m{grep(!/$re/, @v1)} and $self->{-culled_manifest} = 1; + delete @$m{map { @$_ } @gone} and $self->{chg}->{manifest} = 1; + delete @$m{grep(!/$re/, @v1)} and $self->{chg}->{manifest} = 1; @v1 = grep(/$re/, @v1); } if (defined $excl) { - my $re = '(?:'.join('|', map { - $self->{lei}->glob2re($_) // qr/\A\Q$_\E\z/ - } @$excl).')'; + my $re = '(?:'.join('\\z|', map { + $self->{lei}->glob2re($_) // qr/\A\Q$_\E/ + } @$excl).'\\z)'; my @gone = delete @$v2{grep(/$re/, keys %$v2)}; - delete @$m{map { @$_ } @gone} and $self->{-culled_manifest} = 1; - delete @$m{grep(/$re/, @v1)} and $self->{-culled_manifest} = 1; + delete @$m{map { @$_ } @gone} and $self->{chg}->{manifest} = 1; + delete @$m{grep(/$re/, @v1)} and $self->{chg}->{manifest} = 1; @v1 = grep(!/$re/, @v1); } my $ret; # { v1 => [ ... ], v2 => { "/$inbox_name" => [ epochs ] }} @@ -529,27 +927,64 @@ sub multi_inbox ($$$) { } sub clone_all { - my ($self, $todo, $m) = @_; - # handle no-dependency repos, first - for (@{delete($todo->{''}) // []}) { - clone_v1($_, 1); - return if $self->{lei}->{child_error}; + my ($self, $m) = @_; + my $todo = $TODO; + $TODO = \'BUG on further use'; + my $end = PublicInbox::OnDestroy->new($$, \&fgrp_fetch_all, $self); + { + my $nodep = delete $todo->{''}; + + # do not download unwanted deps + my $any_want = delete $self->{any_want}; + my @unwanted = grep { !$any_want->{$_} } keys %$todo; + my @nodep = delete(@$todo{@unwanted}); + push(@$nodep, @$_) for @nodep; + + # handle no-dependency repos, first + for (@$nodep) { + clone_v1($_, $end); + return if !keep_going($self); + } } # resolve references, deepest, first: while (scalar keys %$todo) { for my $x (keys %$todo) { + my ($nr, $nxt); # resolve multi-level references - while (defined($m->{$x}->{reference})) { - $x = $m->{$x}->{reference}; + while ($m && defined($nxt = $m->{$x}->{reference})) { + exists($todo->{$nxt}) or last; + die < 1000; +E: dependency loop detected (`$x' => `$nxt') +EOM + $x = $nxt; } my $y = delete $todo->{$x} // next; # already done for (@$y) { - clone_v1($_, 1); - return if $self->{lei}->{child_error}; + clone_v1($_, $end); + return if !keep_going($self); } last; # restart %$todo iteration } } + + # $end->DESTROY will call fgrp_fetch_all once all references + # in $LIVE are gone, and do_reap will eventually drain $LIVE + $end = undef; + do_reap($self, 1); +} + +sub dump_manifest ($$) { + my ($m, $ft) = @_; + # write the smaller manifest if epochs were skipped so + # users won't have to delete manifest if they +w an + # epoch they no longer want to skip + my $json = PublicInbox::Config->json->encode($m); + my $mtime = (stat($ft))[9]; + seek($ft, SEEK_SET, 0) or die "seek($ft): $!"; + truncate($ft, 0) or die "truncate($ft): $!"; + gzip(\$json => $ft) or die "gzip($ft): $GzipError"; + $ft->flush or die "flush($ft): $!"; + utime($mtime, $mtime, "$ft") or die "utime(..., $ft): $!"; } # FIXME: this gets confused by single inbox instance w/ global manifest.js.gz @@ -558,28 +993,41 @@ sub try_manifest { my $uri = URI->new($self->{src}); my $lei = $self->{lei}; my $curl = $self->{curl} //= PublicInbox::LeiCurl->new($lei) or return; + $self->{-torsocks} //= $curl->torsocks($lei, $uri) or return; my $path = $uri->path; chop($path) eq '/' or die "BUG: $uri not canonicalized"; $uri->path($path . '/manifest.js.gz'); - my $ft = File::Temp->new(TEMPLATE => '.manifest-XXXX', - UNLINK => 1, TMPDIR => 1, SUFFIX => '.tmp'); - my $fn = $ft->filename; - my $cmd = $curl->for_uri($lei, $uri, '-R', '-o', $fn); - my %opt = map { $_ => $lei->{$_} } (0..2); + my $manifest = $self->{-manifest} // "$self->{dst}/manifest.js.gz"; + my %opt = (UNLINK => 1, SUFFIX => '.tmp', TMPDIR => 1); + if (!$self->{dry_run} && $manifest =~ m!\A(.+?)/[^/]+\z! and -d $1) { + $opt{DIR} = $1; # allows fast rename(2) w/o EXDEV + delete $opt{TMPDIR}; + } + my $ft = File::Temp->new(TEMPLATE => '.manifest-XXXX', %opt); + my $cmd = $curl->for_uri($lei, $uri, qw(-R -o), $ft->filename); + push(@$cmd, '-z', $manifest) if -f $manifest; + my $mf_url = "$uri"; + %opt = map { $_ => $lei->{$_} } (0..2); my $cerr = run_reap($lei, $cmd, \%opt); - local $LIVE; if ($cerr) { return try_scrape($self) if ($cerr >> 8) == 22; # 404 missing return $lei->child_error($cerr, "@$cmd failed"); } - my $m = eval { decode_manifest($ft, $fn, $uri) }; + + # bail out if curl -z/--timecond hit 304 Not Modified, $ft will be empty + return $lei->qerr("# $manifest unchanged") if -f $manifest && !-s $ft; + + my $m = eval { decode_manifest($ft, $ft, $uri) }; if ($@) { warn $@; return try_scrape($self); } + local $self->{chg} = {}; + local $self->{-local_manifest} = load_current_manifest($self); my ($path_pfx, $n, $multi) = multi_inbox($self, \$path, $m); return $lei->child_error(1, $multi) if !ref($multi); - if (my $v2 = delete $multi->{v2}) { + my $v2 = delete $multi->{v2}; + if ($v2) { for my $name (sort keys %$v2) { my $epochs = delete $v2->{$name}; my %v2_epochs = map { @@ -600,22 +1048,21 @@ sub try_manifest { index($self->{cur_dst}, "\n") >= 0 and die <{cur_dst}' must not contain newline EOM - clone_v2($self, \%v2_epochs, $m); - return if $self->{lei}->{child_error}; + clone_v2_prep($self, \%v2_epochs, $m); + return if !keep_going($self); } } if (my $v1 = delete $multi->{v1}) { my $p = $path_pfx.$path; chop($p) if substr($p, -1, 1) eq '/'; $uri->path($p); - my $todo = {}; - my %want = map { $_ => 1 } @$v1; for my $name (@$v1) { my $task = bless { %$self }, __PACKAGE__; $task->{-ent} = $m->{$name} // die("BUG: no `$name' in manifest"); $task->{cur_src} = "$uri"; $task->{cur_dst} = $task->{dst}; + $task->{-key} = $name; if ($n > 1) { $task->{cur_dst} .= $name; $task->{cur_src} .= $name; @@ -625,25 +1072,35 @@ E: `$task->{cur_dst}' must not contain newline EOM $task->{cur_src} .= '/'; my $dep = $task->{-ent}->{reference} // ''; - $dep = '' if !$want{$dep}; - push @{$todo->{$dep}}, $task; + push @{$TODO->{$dep}}, $task; # for clone_all + $self->{any_want}->{$name} = 1; } - clone_all($self, $todo, $m); } - reap_live() while keys(%$LIVE); - return if $self->{lei}->{child_error} || $self->{dry_run}; + delete local $lei->{opt}->{epoch} if defined($v2); + clone_all($self, $m); + return if $self->{dry_run} || !keep_going($self); - if (delete $self->{-culled_manifest}) { # set by clone_v2/-I/--exclude - # write the smaller manifest if epochs were skipped so - # users won't have to delete manifest if they +w an - # epoch they no longer want to skip - my $json = PublicInbox::Config->json->encode($m); - my $mtime = (stat($fn))[9]; - gzip(\$json => $fn) or die "gzip: $GzipError"; - utime($mtime, $mtime, $fn) or die "utime(..., $fn): $!"; + # set by clone_v2_prep/-I/--exclude + my $mis = delete $self->{chg}->{fp_mismatch}; + if ($mis) { + my $t = (stat($ft))[9]; + $t = strftime('%F %k:%M:%S %z', localtime($t)); + warn <{lei}->{opt}->{prune}; +W: The above fingerprints may never match without --prune +EOM } - ft_rename($ft, "$self->{dst}/manifest.js.gz", 0666); - open my $x, '>', "$self->{dst}/mirror.done"; # for _wq_done_wait + dump_manifest($m => $ft) if delete($self->{chg}->{manifest}) || $mis; + my $bad = delete $self->{chg}->{badlink}; + warn(<{lei}; + $self->{dry_run} = 1 if $lei->{opt}->{'dry-run'}; umask($lei->{client_umask}) if defined $lei->{client_umask}; + $self->{-initial_clone} = 1 if !-d $self->{dst}; eval { my $ic = $lei->{opt}->{'inbox-config'} //= 'always'; $ic =~ /\A(?:v1|v2|always|never)\z/s or die <<""; --inbox-config must be one of `always', `v2', `v1', or `never' - my $iv = $lei->{opt}->{'inbox-version'}; - if (defined $iv) { - local $LIVE; - return clone_v1($self) if $iv == 1; - return try_scrape($self) if $iv == 2; - die "bad --inbox-version=$iv\n"; + # we support --objstore= and --manifest= with '' (empty string) + for my $default (qw(objstore manifest.js.gz)) { + my ($k) = (split(/\./, $default))[0]; + my $v = $lei->{opt}->{$k} // next; + $v = $default if $v eq ''; + $v = "$self->{dst}/$v" if $v !~ m!\A\.{0,2}/!; + $self->{"-$k"} = $v; } - return start_clone_url($self) if $self->{src} =~ m!://!; - die "TODO: cloning local directories not supported, yet"; + local $LIVE = {}; + local $TODO = {}; + local $FGRP_TODO = {}; + my $iv = $lei->{opt}->{'inbox-version'} // + return start_clone_url($self); + return clone_v1($self) if $iv == 1; + die "bad --inbox-version=$iv\n" if $iv != 2; + die <{src} !~ m!://!; +cloning local v2 inboxes not supported +EOM + try_scrape($self, 1); }; $lei->fail($@) if $@; } @@ -719,6 +1188,7 @@ help : @echo Rarely needed targets: @echo ' make reindex - may be needed for new features/bugfixes' @echo ' make compact - rewrite Xapian storage to save space' + @echo ' make index - initial index after clone fetch : public-inbox-fetch @@ -735,12 +1205,14 @@ update : echo 'public-inbox index not initialized'; \ echo 'see public-inbox-index(1) man page'; \ fi +index : + public-inbox-index reindex : public-inbox-index --reindex compact : public-inbox-compact -.PHONY : help fetch update reindex compact +.PHONY : help fetch update index reindex compact EOM close $fh or die "close($f): $!"; } else {