* [PATCH 11/26] xcpdb: implement using Perl bindings
2019-05-23 9:36 6% [PATCH 00/26] xcpdb: ease Xapian DB format migrations Eric Wong
@ 2019-05-23 9:36 7% ` Eric Wong
0 siblings, 0 replies; 2+ results
From: Eric Wong @ 2019-05-23 9:36 UTC (permalink / raw)
To: meta
By avoid copydatabase(1) entirely, we can make further changes
to avoid locking the entire inbox for a long operation and
switch to fine-grained locking.
---
lib/PublicInbox/Xapcmd.pm | 77 +++++++++++++++++++++++++++++++++++++--
script/public-inbox-xcpdb | 2 +-
2 files changed, 75 insertions(+), 4 deletions(-)
diff --git a/lib/PublicInbox/Xapcmd.pm b/lib/PublicInbox/Xapcmd.pm
index 81e2f10..ca74ea0 100644
--- a/lib/PublicInbox/Xapcmd.pm
+++ b/lib/PublicInbox/Xapcmd.pm
@@ -24,15 +24,36 @@ sub commit_changes ($$$) {
remove_tree("$old/old") or die "failed to remove $old/old: $!\n";
}
+sub xspawn {
+ my ($cmd, $env, $opt) = @_;
+ if (ref($cmd->[0]) eq 'CODE') {
+ my $cb = shift(@$cmd); # $cb = cpdb()
+ defined(my $pid = fork) or die "fork: $!";
+ return $pid if $pid > 0;
+ eval { $cb->($cmd, $env, $opt) };
+ die $@ if $@;
+ exit 0;
+ } else {
+ spawn($cmd, $env, $opt);
+ }
+}
+
sub run {
my ($ibx, $cmd, $env, $opt) = @_;
$opt ||= {};
my $dir = $ibx->{mainrepo} or die "no mainrepo in inbox\n";
- which($cmd->[0]) or die "$cmd->[0] not found in PATH\n";
+ my $exe = $cmd->[0];
+ my $pfx = $exe;
+ if (ref($exe) eq 'CODE') {
+ $pfx = 'CODE';
+ require Search::Xapian::WritableDatabase;
+ } else {
+ which($exe) or die "$exe not found in PATH\n";
+ }
$ibx->umask_prepare;
my $old = $ibx->search->xdir(1);
-d $old or die "$old does not exist\n";
- my $new = tempdir($cmd->[0].'-XXXXXXXX', DIR => $dir);
+ my $new = tempdir("$pfx-XXXXXXXX", DIR => $dir);
my $v = $ibx->{version} || 1;
my @cmds;
if ($v == 1) {
@@ -58,7 +79,7 @@ sub run {
while (@cmds) {
while (scalar(keys(%pids)) < $max && scalar(@cmds)) {
my $x = shift @cmds;
- $pids{spawn($x, $env, $opt)} = $x;
+ $pids{xspawn($x, $env, $opt)} = $x;
}
while (scalar keys %pids) {
@@ -71,4 +92,54 @@ sub run {
});
}
+sub cpdb_retryable ($$) {
+ my ($src, $err) = @_;
+ if (ref($err) eq 'Search::Xapian::DatabaseModifiedError') {
+ warn "$err, reopening and retrying\n";
+ $src->reopen;
+ return 1;
+ }
+ die $err if $err;
+ 0;
+}
+
+sub cpdb {
+ my ($args, $env, $opt) = @_;
+ my ($old, $new) = @$args;
+ my $src = Search::Xapian::Database->new($old);
+
+ # like copydatabase(1), be sure we don't overwrite anything in case
+ # of other bugs:
+ my $creat = Search::Xapian::DB_CREATE();
+ my $dst = Search::Xapian::WritableDatabase->new($new, $creat);
+ my ($it, $end);
+
+ do {
+ eval {
+ # update the only metadata key for v1:
+ my $lc = $src->get_metadata('last_commit');
+ $dst->set_metadata('last_commit', $lc) if $lc;
+
+ $it = $src->postlist_begin('');
+ $end = $src->postlist_end('');
+ };
+ } while (cpdb_retryable($src, $@));
+
+ do {
+ eval {
+ while ($it != $end) {
+ my $docid = $it->get_docid;
+ my $doc = $src->get_document($docid);
+ $dst->replace_document($docid, $doc);
+ $it->inc;
+ }
+
+ # unlike copydatabase(1), we don't copy spelling
+ # and synonym data (or other user metadata) since
+ # the Perl APIs don't expose iterators for them
+ # (and public-inbox does not use those features)
+ };
+ } while (cpdb_retryable($src, $@));
+}
+
1;
diff --git a/script/public-inbox-xcpdb b/script/public-inbox-xcpdb
index cbf9f55..d494991 100755
--- a/script/public-inbox-xcpdb
+++ b/script/public-inbox-xcpdb
@@ -8,7 +8,7 @@ use PublicInbox::Admin;
PublicInbox::Admin::require_or_die('-search');
my $usage = "Usage: public-inbox-xcpdb INBOX_DIR\n";
my @ibxs = PublicInbox::Admin::resolve_inboxes(\@ARGV) or die $usage;
-my $cmd = [qw(copydatabase --no-renumber)];
+my $cmd = [ \&PublicInbox::Xapcmd::cpdb ];
open my $null, '>', '/dev/null' or die "failed to open /dev/null: $!\n";
my $rdr = { 1 => fileno($null) };
foreach (@ibxs) {
--
EW
^ permalink raw reply related [relevance 7%]
* [PATCH 00/26] xcpdb: ease Xapian DB format migrations
@ 2019-05-23 9:36 6% Eric Wong
2019-05-23 9:36 7% ` [PATCH 11/26] xcpdb: implement using Perl bindings Eric Wong
0 siblings, 1 reply; 2+ results
From: Eric Wong @ 2019-05-23 9:36 UTC (permalink / raw)
To: meta
I've noticed performance problems in Xapian's old chert
backend which seem alleviated with the new glass backend;
particularly related to phrase searches.
Unfortunately, the tool distributed with Xapian for updating DB
formats, copydatabase(1), is extremely slow and blocking updates
for hours at a time to perform the migration is not acceptable.
(That's right, "copydatabase" is NOT a Postgres command!)
So, I've written "public-inbox-xcpdb" and gotten it to perform
the bulk copy operation without holding inbox.lock and have it
deal gracefully with Xapian DB modifications. xcpdb is still
slow, but I've (finally!) implemented partial reindexing to
allow it to minimize the lock time and not stall -mda or -watch
processes while it is working.
There's a bunch of cleanups along the way, too; and it should
make future changes to repartition the Xapian DB on existing v2
inboxes easier.
Eric Wong (26):
t/convert-compact: skip on missing xapian-compact(1)
v1writable: retire in favor of InboxWritable
doc: document the reason for --no-renumber
search: reenable phrase search on non-chert Xapian
xapcmd: new module for wrapping Xapian commands
admin: hoist out resolve_inboxes for -compact and -index
xapcmd: support spawn options
xcpdb: new tool which wraps Xapian's copydatabase(1)
xapcmd: do not cleanup on errors
admin: move index_inbox over
xcpdb: implement using Perl bindings
xapcmd: xcpdb supports compaction
v2writable: hoist out log_range sub for readability
xcpdb: use fine-grained locking
xcpdb: implement progress reporting
xcpdb: cleanup error handling and diagnosis
xapcmd: avoid EXDEV when finalizing changes
doc: xcpdb: update to reflect the current state
xapcmd: use "print STDERR" for progress reporting
xcpdb: show re-indexing progress
xcpdb: remove temporary directories on aborts
compact: reuse infrastructure from xcpdb
xcpdb|compact: support some xapian-compact switches
xapcmd: cleanup on interrupted xcpdb "--compact"
xcpdb|compact: support --jobs/-j flag like gmake(1)
xapcmd: do not reset %SIG until last Xtmpdir is done
Documentation/include.mk | 6 +-
Documentation/public-inbox-v1-format.pod | 4 +
Documentation/public-inbox-v2-format.pod | 4 +
Documentation/public-inbox-xcpdb.pod | 57 ++++
MANIFEST | 4 +-
lib/PublicInbox/Admin.pm | 66 ++++
lib/PublicInbox/InboxWritable.pm | 35 ++-
lib/PublicInbox/Search.pm | 48 +--
lib/PublicInbox/SearchIdx.pm | 34 ++-
lib/PublicInbox/V1Writable.pm | 34 ---
lib/PublicInbox/V2Writable.pm | 109 ++++---
lib/PublicInbox/Xapcmd.pm | 370 +++++++++++++++++++++++
script/public-inbox-compact | 102 +------
script/public-inbox-index | 102 +------
script/public-inbox-init | 13 +-
script/public-inbox-xcpdb | 19 ++
t/cgi.t | 4 +-
t/convert-compact.t | 4 +
t/indexlevels-mirror.t | 27 +-
t/init.t | 4 +-
t/nntpd.t | 15 +-
t/search.t | 1 +
t/v2mirror.t | 1 +
23 files changed, 740 insertions(+), 323 deletions(-)
create mode 100644 Documentation/public-inbox-xcpdb.pod
delete mode 100644 lib/PublicInbox/V1Writable.pm
create mode 100644 lib/PublicInbox/Xapcmd.pm
create mode 100755 script/public-inbox-xcpdb
--
EW
^ permalink raw reply [relevance 6%]
Results 1-2 of 2 | reverse | options above
-- pct% links below jump to the message on this page, permalinks otherwise --
2019-05-23 9:36 6% [PATCH 00/26] xcpdb: ease Xapian DB format migrations Eric Wong
2019-05-23 9:36 7% ` [PATCH 11/26] xcpdb: implement using Perl bindings 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).