* [PATCH 15/23] searchview: use over.sqlite3 instead of Xapian docdata
2020-08-20 20:24 7% [PATCH 00/23] indexing: --skip-docdata + speedups Eric Wong
@ 2020-08-20 20:24 6% ` Eric Wong
0 siblings, 0 replies; 2+ results
From: Eric Wong @ 2020-08-20 20:24 UTC (permalink / raw)
To: meta
This is a step towards improving kernel page cache hit rates by
relying on over.sqlite3 for document data instead of Xapian.
Some micro-optimization to over->get_art was required to
maintain performance.
---
lib/PublicInbox/Over.pm | 18 +++++-------------
lib/PublicInbox/SearchView.pm | 10 +++++++---
2 files changed, 12 insertions(+), 16 deletions(-)
diff --git a/lib/PublicInbox/Over.pm b/lib/PublicInbox/Over.pm
index 81b9fca7..80e57e62 100644
--- a/lib/PublicInbox/Over.pm
+++ b/lib/PublicInbox/Over.pm
@@ -57,6 +57,7 @@ sub new {
sub disconnect {
my ($self) = @_;
if (my $dbh = delete $self->{dbh}) {
+ delete $self->{-get_art};
$self->{filename} = $dbh->sqlite_db_filename;
}
}
@@ -201,8 +202,8 @@ SELECT COUNT(num) FROM over WHERE num > 0
sub get_art {
my ($self, $num) = @_;
- my $dbh = $self->connect;
- my $sth = $dbh->prepare_cached(<<'', undef, 1);
+ # caching $sth ourselves is faster than prepare_cached
+ my $sth = $self->{-get_art} //= $self->connect->prepare(<<'');
SELECT num,ds,ts,ddd FROM over WHERE num = ? LIMIT 1
$sth->execute($num);
@@ -230,13 +231,7 @@ ORDER BY num ASC LIMIT 1
$sth->execute($$id, $$prev);
my $num = $sth->fetchrow_array or return;
$$prev = $num;
-
- $sth = $dbh->prepare_cached(<<"", undef, 1);
-SELECT num,ts,ds,ddd FROM over WHERE num = ? LIMIT 1
-
- $sth->execute($num);
- my $smsg = $sth->fetchrow_hashref or return;
- load_from_row($smsg);
+ get_art($self, $num);
}
# IMAP search, this is limited by callers to UID_SLICE size (50K)
@@ -278,10 +273,7 @@ sub check_inodes {
my $st = pack('dd', $st[0], $st[1]);
# don't actually reopen, just let {dbh} be recreated later
- if ($st ne ($self->{st} // $st)) {
- delete($self->{dbh});
- $self->{filename} = $f;
- }
+ disconnect($self) if $st ne ($self->{st} // $st);
} else {
warn "W: stat $f: $!\n";
}
diff --git a/lib/PublicInbox/SearchView.pm b/lib/PublicInbox/SearchView.pm
index 28d9ce5d..61534c25 100644
--- a/lib/PublicInbox/SearchView.pm
+++ b/lib/PublicInbox/SearchView.pm
@@ -13,6 +13,7 @@ use PublicInbox::WwwAtomStream;
use PublicInbox::WwwStream qw(html_oneshot);
use PublicInbox::SearchThread;
use PublicInbox::SearchQuery;
+use PublicInbox::Search qw(mdocid);
my %rmap_inc;
sub mbox_results {
@@ -90,19 +91,22 @@ sub mset_summary {
my $pfx = ' ' x $pad;
my $res = \($ctx->{-html_tip});
my $ibx = $ctx->{-inbox};
- my $srch = $ibx->search;
+ my $over = $ibx->over;
+ my $nshard = $ibx->search->{nshard} // 1;
my $obfs_ibx = $ibx->{obfuscate} ? $ibx : undef;
foreach my $m ($mset->items) {
my $rank = sprintf("%${pad}d", $m->get_rank + 1);
my $pct = get_pct($m);
- my $smsg = PublicInbox::Smsg::from_mitem($m, $srch);
+ my $num = mdocid($nshard, $m);
+ my $smsg = $over->get_art($num, 1);
unless ($smsg) {
eval {
- $m = "$m ".$m->get_docid . " expired\n";
+ $m = "$m $num expired\n";
$ctx->{env}->{'psgi.errors'}->print($m);
};
next;
}
+ PublicInbox::Smsg::psgi_cull($smsg);
my $s = ascii_html($smsg->{subject});
my $f = ascii_html($smsg->{from_name});
if ($obfs_ibx) {
^ permalink raw reply related [relevance 6%]
* [PATCH 00/23] indexing: --skip-docdata + speedups
@ 2020-08-20 20:24 7% Eric Wong
2020-08-20 20:24 6% ` [PATCH 15/23] searchview: use over.sqlite3 instead of Xapian docdata Eric Wong
0 siblings, 1 reply; 2+ results
From: Eric Wong @ 2020-08-20 20:24 UTC (permalink / raw)
To: meta
Some miscellaneous help and cleanup things, too.
Document data is no longer read from Xapian by read-only
daemons; that data is redundant given over.sqlite3 always
exists. This should should improve page cache hit rates for
over.sqlite3 by a small bit. Being able to mass load a bunch of
rows from SQLite speeds up the default search summary view by
~10%, too.
--skip-docdata option to -init and -index can avoid writing
Xapian document data, saving ~1.5% in Xapian space overhead
(and associated I/O and page cache overheads). It breaks
rollbacks to old versions, though, so it won't be the default.
Eric Wong (23):
doc: note -compact and -xcpdb are rarely used
admin: progress shows the inbox being indexed
compact: support --help/-? and perform lazy loading
init: support --help and -?
init: support --newsgroup option
init: drop -N alias for --skip-artnum
search: v2: ensure shards are numerically sorted
xapcmd: simplify {reindex} parameter passing
www: reduce long-lived PublicInbox::Search references
search: improve comments around constants
search: export mdocid subroutine
searchquery: split off from searchview
search: make qparse_new an internal function
smsg: reduce utf8::decode call sites
searchview: use over.sqlite3 instead of Xapian docdata
searchview: speed up search summary by ~10%
searchview: convert nested and Atom display to over.sqlite3
extmsg: avoid using Xapian docdata
mbox: avoid Xapian docdata in search results
smsg: remove from_mitem
t/nntpd-v2: set PI_TEST_VERSION=2 properly
init+index: support --skip-docdata for Xapian
search: add mset_to_artnums method
Documentation/public-inbox-compact.pod | 5 +
Documentation/public-inbox-config.pod | 2 +-
Documentation/public-inbox-index.pod | 8 ++
Documentation/public-inbox-init.pod | 37 +++++++-
Documentation/public-inbox-xcpdb.pod | 3 +
MANIFEST | 1 +
lib/PublicInbox/Admin.pm | 15 ++-
lib/PublicInbox/ExtMsg.pm | 19 ++--
lib/PublicInbox/IMAP.pm | 5 +-
lib/PublicInbox/Inbox.pm | 11 ++-
lib/PublicInbox/Mbox.pm | 24 +++--
lib/PublicInbox/Over.pm | 28 +++---
lib/PublicInbox/Search.pm | 125 ++++++++++++++-----------
lib/PublicInbox/SearchIdx.pm | 35 +++++--
lib/PublicInbox/SearchIdxShard.pm | 2 +-
lib/PublicInbox/SearchQuery.pm | 53 +++++++++++
lib/PublicInbox/SearchView.pm | 90 ++++--------------
lib/PublicInbox/Smsg.pm | 10 +-
lib/PublicInbox/Xapcmd.pm | 20 ++--
script/public-inbox-compact | 39 ++++++--
script/public-inbox-convert | 3 +-
script/public-inbox-index | 7 +-
script/public-inbox-init | 101 +++++++++++++-------
t/imapd.t | 6 +-
t/inbox_idle.t | 2 +-
t/index-git-times.t | 11 ++-
t/init.t | 17 +++-
t/nntpd-v2.t | 2 +-
t/nntpd.t | 9 +-
t/search.t | 34 ++++---
30 files changed, 437 insertions(+), 287 deletions(-)
create mode 100644 lib/PublicInbox/SearchQuery.pm
^ 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 --
2020-08-20 20:24 7% [PATCH 00/23] indexing: --skip-docdata + speedups Eric Wong
2020-08-20 20:24 6% ` [PATCH 15/23] searchview: use over.sqlite3 instead of Xapian docdata 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).