* [PATCH 68/95] lei_mirror: support resuming multi-repo clones
2022-11-28 5:30 7% [PATCH 00/95] clone: multi-inbox/repo support Eric Wong
@ 2022-11-28 5:32 5% ` Eric Wong
0 siblings, 0 replies; 2+ results
From: Eric Wong @ 2022-11-28 5:32 UTC (permalink / raw)
To: meta
This is actually a combination of clone and fetch, and I don't
think `public-inbox-fetch' will be used to update multiple git
repos (inbox or not).
Our use of `git update-ref --stdin -z' was broken for
incremental updates, but now fixed to properly NUL-terminate
commands.
---
lib/PublicInbox/LeiMirror.pm | 34 +++++++++++++++++++++++++++-------
1 file changed, 27 insertions(+), 7 deletions(-)
diff --git a/lib/PublicInbox/LeiMirror.pm b/lib/PublicInbox/LeiMirror.pm
index 8b276336..f2111dd2 100644
--- a/lib/PublicInbox/LeiMirror.pm
+++ b/lib/PublicInbox/LeiMirror.pm
@@ -256,7 +256,7 @@ sub run_reap {
$ret;
}
-sub start_clone {
+sub start_cmd {
my ($self, $cmd, $opt, $fini) = @_;
do_reap($self);
$self->{lei}->qerr("# @$cmd");
@@ -279,9 +279,9 @@ sub fetch_args ($$) {
}
sub upr { # feed `git update-ref --stdin -z' verbosely
- my ($lei, $w, $op, $ref, $oid) = @_;
- $lei->qerr("# $op $ref $oid") if $lei->{opt}->{verbose};
- print $w "$op $ref\0$oid\0" or die "print(w): $!";
+ my ($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 {
@@ -306,7 +306,8 @@ sub fgrp_update {
my $new = delete $src{$ref};
my $old = $dst{$ref};
if (defined $new) {
- upr($lei, $w, 'update', $ref, $new) if $new ne $old;
+ $new eq $old or
+ upr($lei, $w, 'update', $ref, $new, $old);
} else {
upr($lei, $w, 'delete', $ref, $old);
}
@@ -482,6 +483,23 @@ EOM
bless { %$self, -osdir => $dir, -remote => $rn }, __PACKAGE__;
}
+sub resume_fetch {
+ my ($self, $uri, $fini) = @_;
+ my $dst = $self->{cur_dst} // $self->{dst};
+ my @git = ('git', "--git-dir=$dst");
+ my $opt = +{ map { $_ => $self->{lei}->{$_} } (0..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), $rn ];
+ start_cmd($self, $cmd, $opt, $fini);
+}
+
sub clone_v1 {
my ($self, $nohang) = @_;
my $lei = $self->{lei};
@@ -495,7 +513,9 @@ sub clone_v1 {
if (my $fgrp = forkgroup_prep($self, $uri)) {
$fgrp->{-fini} = $fini;
push @{$self->{fgrp_todo}->{$fgrp->{-osdir}}}, $fgrp;
- } else { # normal fetch
+ } elsif (-d $dst) {
+ resume_fetch($self, $uri, $fini);
+ } else { # normal clone
my $cmd = [ @{$self->{-torsocks}},
clone_cmd($lei, my $opt = {}), "$uri", $dst ];
if (defined($self->{-ent})) {
@@ -505,7 +525,7 @@ sub clone_v1 {
"$self->{dst}$ref";
}
}
- start_clone($self, $cmd, $opt, $fini);
+ start_cmd($self, $cmd, $opt, $fini);
}
if (!$self->{-is_epoch} && $lei->{opt}->{'inbox-config'} =~
/\A(?:always|v1)\z/s) {
^ permalink raw reply related [relevance 5%]
* [PATCH 00/95] clone: multi-inbox/repo support...
@ 2022-11-28 5:30 7% Eric Wong
2022-11-28 5:32 5% ` [PATCH 68/95] lei_mirror: support resuming multi-repo clones Eric Wong
0 siblings, 1 reply; 2+ results
From: Eric Wong @ 2022-11-28 5:30 UTC (permalink / raw)
To: meta
A large patchset, and not done, yet :P It's only tested live,
but it seems to work reasonably well against live hosts...
Behavior changes to public-inbox-clone are NOT final; but
public-inbox-fetch|PublicInbox::Fetch will probably become
thin wrappers around LeiMirror.
--include=/--exclude= support now exists with glob support
--keep-going and --dry-run support added, too, since it's
make(1) influenced (more below)
It supports coderepos, too, using --inbox-config=never (default: always);
--project-list=, --manifest=, --objstore=, and --prune.
key differences from grok-pull (grokmirror) for coderepos:
* uses relative paths on the FS (dumb HTTP untested, but dumb
HTTP is a goal for memory-constrained hosts). This means
I can relocate coderepos freely within my FS or do sneakernet
transfers across machines without having to `perl -ipe s/x/y/'
on hundreds of info/alternates and config files.
* CLI-only, no extra config files (may generate a Makefile, like
individual inbox clones)
* objstore repos fetches from remotes directly
(does not need, use, nor benefit from hardlinks at all)
* no sleep states
It is not a full replacement for grokmirror
* reliant on default `git gc' behavior for repack. This is OK
since it's only one-way relationships between objstore and
non-objstore repos.
* no fsck support (probably will be in generated Makefile)
* doesn't generate forkgroups nor manifest.js.gz
(I may do this for coderepo Xapian indexing)
It relies on parallel git-fetch for objstores, so `-j $NUM'
calculations may end up being ($NUM * $NUM) in the worst case.
Not sure how to best approach this...
Maybe `-j $M,$N' similar to `lei q -j$M,$N` is a solution...
Design note:
This is an exercise in building make(1)-like parallelism using
->DESTROY callbacks for prerequisites; so it's a newish paradigm
for me. It forced me to fix a reference cycle, already.
TODO: repo|symlink pruning, --exit-code, retry/refetch, manpage updates
Eric Wong (95):
clone: support multi-inbox clone
clone: support --include and --exclude with multi-clone
clone: parallelize v2 epoch clones
lei_mirror: async config retrieval for v2 w/ manifest
lei_mirror: rely on DESTROY to index v2 inbox
lei_mirror: rely on global process reaper
clone: support parallel v1 clones
lei_mirror: default to single job by default
lei_mirror: move directory creation to v2-only path
lei_mirror: retrieve description text asynchronously, too
switch inotify/kevent stuff to v5.12
manifest: update module blurb + v5.12
lei_mirror: simplify _get_txt_start callers
lei_mirror: elide description retrieval for v1|coderepo
lei_mirror: add a hint for skipped epoch permissions
lei_mirror: consolidate clone process management
lei_mirror: load File::Path unconditionally
lei_mirror: load most modules up-front
lei_mirror: set gitweb.owner from manifest
clone: support --dry-run / -n flag
lei_mirror: initialize placeholders with "head" from manifest
lei_mirror: support {reference} for v1 manifest clones
lei_mirror: reduce noise on interrupted clones
clone: support --inbox-config option
lei_mirror: retrieve v2 description properly
lei_mirror: reduce scope of v2 lock
lei_mirror: allow --epoch on mixed v1/v2 clones
lei_mirror: fix infinite loop in dependency resolution
lei_mirror: defend against infinite loops
lei_mirror: do not fetch descriptions if using manifest
lei_mirror: require PublicInbox::Lock at use
lei_mirror: fix glob semantics to match end-of-path
lei_mirror: differentiate -entv vs -ent
lei_mirror: support manifest {references} for v2 epochs
lei_mirror: simplify v2 code paths
clone: support --inbox-version
lei_mirror: require Perl v5.12+
lei_mirror: ensure curl exits 22 on HTTP 404 responses
lei_mirror: cleanup File::Temp OO usage
lei_mirror: add `index' target to generated Makefile
lei_mirror: do not write Makefile for --inbox-config=never
lei_mirror: hoist out dump_manifest sub
lei_mirror: avoid convoluted lazy_cb usage
lei_mirror: simplify clone_v2_prep
lei_mirror: support --objstore and forkgroups
lei_mirror: cleanup process reaping logic
lei_mirror: ensure git <1.8.5 fallback can use torsocks
clone: flesh out --objstore behavior and document
lei_mirror: always pack refs for coderepos
lei_mirror: set description for non-inboxes, too
lei_mirror: force --no-tags when fetching forkgroups
lei_mirror: preserve permissions of existing alternates file
lei_mirror: do not show ref updates w/o --verbose
lei_mirror: drop git <1.8.5 support
lei_mirror: make basename more descriptive
lei_mirror: fix --dry-run for forkgroups
lei_mirror: forkgroups use `git fetch --multiple'
clone: move --dry-run handling to lei_mirror
clone: drop unnecessary requires
clone: use v5.12
clone: require `--objstore=' for default location
lei_mirror: shorten remote names
fetch: use v5.12
fetch: eliminate File::Temp->filename var
lei_mirror: properly pack-refs in non-forkgroup repos
lei_mirror: show child error error code
on_destroy: support ->cancel callback
lei_mirror: support resuming multi-repo clones
lei_mirror: check fingerprints before fetching
clone: support loading manifest.js.gz from destination
lei_mirror: delay configuring forkgroups
clone: canonicalize destination path from CLI
clone|fetch: support passing --prune(-tags) to `git fetch'
lei_mirror: avoid needless FD passing
clone: support --keep-going/-k like make(1)
lei_mirror: don't warn on missing manifest on initial clone
lei_mirror: respect `./' and `../' prefixes for CLI args
lei_mirror: --manifest= affects destination, too
lei_mirror: update fingerprints when writing local manifest.js.gz
lei_mirror: remove janky mirror.done stamp file
lei_mirror: simplify most process spawning
lei_mirror: run v1_done earlier on forkgroup done
lei_mirror: simplify forkgroup-related subs
lei_mirror: shorten scope mirror objects
lei_mirror: set {head} from manifest
lei_mirror: support {symlinks} from manifest
lei_mirror: eliminate circular references
lei_mirror: use curl -z/--timecond if manifest exists
lei_mirror: avoid redundant curl `-f' use
lei_mirror: omit trailing slash for git remote.*.url
lei_mirror: set info/web/last-modified from manifest
lei_mirror: don't clobber inbox.config.example if it exists
lei_mirror: break out of fgrp fetch iteration early
clone: support --project-list= for cgit
lei_mirror: handle forkgroup changes
Documentation/lei-add-external.pod | 4 +-
Documentation/public-inbox-clone.pod | 76 ++
Documentation/public-inbox-fetch.pod | 6 +
lib/PublicInbox/DSKQXS.pm | 5 +-
lib/PublicInbox/DirIdle.pm | 4 +-
lib/PublicInbox/FakeInotify.pm | 13 +-
lib/PublicInbox/Fetch.pm | 50 +-
lib/PublicInbox/In2Tie.pm | 4 +-
lib/PublicInbox/InboxIdle.pm | 2 +-
lib/PublicInbox/KQNotify.pm | 12 +-
lib/PublicInbox/LEI.pm | 5 +-
lib/PublicInbox/LeiMirror.pm | 1104 +++++++++++++++++++++-----
lib/PublicInbox/ManifestJsGz.pm | 8 +-
lib/PublicInbox/OnDestroy.pm | 5 +-
lib/PublicInbox/TestCommon.pm | 1 +
script/public-inbox-clone | 23 +-
script/public-inbox-fetch | 4 +-
t/on_destroy.t | 8 +-
t/www_listing.t | 71 +-
19 files changed, 1148 insertions(+), 257 deletions(-)
^ permalink raw reply [relevance 7%]
Results 1-2 of 2 | reverse | options above
-- pct% links below jump to the message on this page, permalinks otherwise --
2022-11-28 5:30 7% [PATCH 00/95] clone: multi-inbox/repo support Eric Wong
2022-11-28 5:32 5% ` [PATCH 68/95] lei_mirror: support resuming multi-repo clones Eric Wong
Code repositories for project(s) associated with this public inbox
https://80x24.org/public-inbox.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).