* [PATCH 64/82] searchidx: v1 (re)-index uses git asynchronously
2020-06-10 7:03 6% [PATCH 00/82] public-inbox-imapd: read-only IMAP server Eric Wong
@ 2020-06-10 7:05 7% ` Eric Wong
0 siblings, 0 replies; 2+ results
From: Eric Wong @ 2020-06-10 7:05 UTC (permalink / raw)
To: meta
We can cleanup some of our v1 code slightly and let git do
I/O+decoding in parallel. This gives a slight 2-4%
re-indexing performance boost even on an SSD.
---
lib/PublicInbox/SearchIdx.pm | 76 +++++++++++++++++-------------------
1 file changed, 35 insertions(+), 41 deletions(-)
diff --git a/lib/PublicInbox/SearchIdx.pm b/lib/PublicInbox/SearchIdx.pm
index 3df7970ebf9..a790ac4076a 100644
--- a/lib/PublicInbox/SearchIdx.pm
+++ b/lib/PublicInbox/SearchIdx.pm
@@ -378,7 +378,7 @@ sub _msgmap_init ($) {
sub add_message {
# mime = PublicInbox::Eml or Email::MIME object
- my ($self, $mime, $smsg) = @_;
+ my ($self, $mime, $smsg, $sync) = @_;
my $hdr = $mime->header_obj;
my $mids = mids_for_index($hdr);
$smsg //= bless { blob => '' }, 'PublicInbox::Smsg'; # test-only compat
@@ -389,7 +389,7 @@ sub add_message {
};
# v1 and tests only:
- $smsg->populate($hdr, $self);
+ $smsg->populate($hdr, $sync);
$smsg->{bytes} //= length($mime->as_string);
eval {
@@ -549,24 +549,24 @@ sub unindex_mm {
$self->{mm}->mid_delete(mid_mime($mime));
}
-sub index_both {
- my ($self, $mime, $smsg) = @_;
- my $num = index_mm($self, $mime);
+sub index_both { # git->cat_async callback
+ my ($bref, $oid, $type, $size, $sync) = @_;
+ my ($nr, $max) = @$sync{qw(nr max)};
+ ++$$nr;
+ $$max -= $size;
+ my $smsg = bless { bytes => $size, blob => $oid }, 'PublicInbox::Smsg';
+ my $self = $sync->{sidx};
+ my $eml = PublicInbox::Eml->new($bref);
+ my $num = index_mm($self, $eml);
$smsg->{num} = $num;
- add_message($self, $mime, $smsg);
+ add_message($self, $eml, $smsg, $sync);
}
-sub unindex_both {
- my ($self, $mime) = @_;
- unindex_blob($self, $mime);
- unindex_mm($self, $mime);
-}
-
-sub do_cat_mail {
- my ($git, $blob, $sizeref) = @_;
- my $str = $git->cat_file($blob, $sizeref) or
- die "BUG: $blob not found in $git->{git_dir}";
- PublicInbox::Eml->new($str);
+sub unindex_both { # git->cat_async callback
+ my ($bref, $oid, $type, $size, $self) = @_;
+ my $eml = PublicInbox::Eml->new($bref);
+ unindex_blob($self, $eml);
+ unindex_mm($self, $eml);
}
# called by public-inbox-index
@@ -576,15 +576,6 @@ sub index_sync {
$self->{-inbox}->with_umask(sub { $self->_index_sync($opts) })
}
-sub batch_adjust ($$$$$) {
- my ($max, $bytes, $batch_cb, $latest, $nr) = @_;
- $$max -= $bytes;
- if ($$max <= 0) {
- $$max = $BATCH_BYTES;
- $batch_cb->($nr, $latest);
- }
-}
-
sub too_big ($$$) {
my ($self, $git, $oid) = @_;
my $max_size = $self->{index_max_size} or return;
@@ -597,24 +588,28 @@ sub too_big ($$$) {
# only for v1
sub read_log {
- my ($self, $log, $add_cb, $del_cb, $batch_cb) = @_;
+ my ($self, $log, $batch_cb) = @_;
my $hex = '[a-f0-9]';
my $h40 = $hex .'{40}';
my $addmsg = qr!^:000000 100644 \S+ ($h40) A\t${hex}{2}/${hex}{38}$!;
my $delmsg = qr!^:100644 000000 ($h40) \S+ D\t${hex}{2}/${hex}{38}$!;
my $git = $self->{git};
my $latest;
- my $bytes;
my $max = $BATCH_BYTES;
local $/ = "\n";
my %D;
my $line;
my $newest;
my $nr = 0;
+ my $sync = { sidx => $self, nr => \$nr, max => \$max };
while (defined($line = <$log>)) {
if ($line =~ /$addmsg/o) {
my $blob = $1;
if (delete $D{$blob}) {
+ # make sure pending index writes are done
+ # before writing to ->mm
+ $git->cat_async_wait;
+
if (defined $self->{regen_down}) {
my $num = $self->{regen_down}--;
$self->{mm}->num_highwater($num);
@@ -622,12 +617,12 @@ sub read_log {
next;
}
next if too_big($self, $git, $blob);
- my $mime = do_cat_mail($git, $blob, \$bytes);
- my $smsg = bless {}, 'PublicInbox::Smsg';
- batch_adjust(\$max, $bytes, $batch_cb, $latest, ++$nr);
- $smsg->{blob} = $blob;
- $smsg->{bytes} = $bytes;
- $add_cb->($self, $mime, $smsg);
+ $git->cat_async($blob, \&index_both, { %$sync });
+ if ($max <= 0) {
+ $git->cat_async_wait;
+ $max = $BATCH_BYTES;
+ $batch_cb->($nr, $latest);
+ }
} elsif ($line =~ /$delmsg/o) {
my $blob = $1;
$D{$blob} = 1 unless too_big($self, $git, $blob);
@@ -635,18 +630,17 @@ sub read_log {
$latest = $1;
$newest ||= $latest;
} elsif ($line =~ /^author .*? ([0-9]+) [\-\+][0-9]+$/) {
- $self->{autime} = $1;
+ $sync->{autime} = $1;
} elsif ($line =~ /^committer .*? ([0-9]+) [\-\+][0-9]+$/) {
- $self->{cotime} = $1;
+ $sync->{cotime} = $1;
}
}
close($log) or die "git log failed: \$?=$?";
# get the leftovers
foreach my $blob (keys %D) {
- my $mime = do_cat_mail($git, $blob, \$bytes);
- $del_cb->($self, $mime);
+ $git->cat_async($blob, \&unindex_both, $self);
}
- delete @$self{qw(autime cotime)};
+ $git->cat_async_wait;
$batch_cb->($nr, $latest, $newest);
}
@@ -774,7 +768,7 @@ sub _index_sync {
} while (_last_x_commit($self, $mm) ne $last_commit);
my $dbh = $mm->{dbh} if $mm;
- my $cb = sub {
+ my $batch_cb = sub {
my ($nr, $commit, $newest) = @_;
if ($dbh) {
if ($newest) {
@@ -803,7 +797,7 @@ sub _index_sync {
};
$dbh->begin_work;
- read_log($self, $xlog, *index_both, *unindex_both, $cb);
+ read_log($self, $xlog, $batch_cb);
}
sub DESTROY {
^ permalink raw reply related [relevance 7%]
* [PATCH 00/82] public-inbox-imapd: read-only IMAP server
@ 2020-06-10 7:03 6% Eric Wong
2020-06-10 7:05 7% ` [PATCH 64/82] searchidx: v1 (re)-index uses git asynchronously Eric Wong
0 siblings, 1 reply; 2+ results
From: Eric Wong @ 2020-06-10 7:03 UTC (permalink / raw)
To: meta
So I finally wrote my first IMAP server! And I'm actually
fairly satisfied with how it's turning out to support a bunch
of other performance + scalability work I've wanted to do.
Some previous notes here:
https://public-inbox.org/meta/20200609113442.GA16856@dcvr/
I finally seem to have gotten it to play nicely with mutt header
caching, so it's fit for public consumption :)
imaps://news.public-inbox.org/INBOX.comp.mail.public-inbox.meta.0
You can use any username+password, and AUTH=ANONYMOUS also
works if your client does that.
It doesn't support UTF-7 (mailbox names) or advertise UTF-8
in CAPABILITIES, yet; I still have RFCs to read :P
And there's a bunch of new things which could use some
testing from non-mutt/mbsync/offlineimap users.
Maybe you'll find some client-side bugs like I did :P
v1 reindexing also gets a little bit of parallelism :)
Anyways, I'll probably be porting some of the scalability
and slow-storage work to older parts of the code before
fiddling with more IMAP extensions.
Eric Wong (82):
doc: add some IMAP standards
nntpd: restrict allowed newsgroup names
preliminary imap server implementation
inboxidle: new class to detect inbox changes
imap: support IDLE
msgmap: split ->max into its own method
imap: delay InboxIdle start, support refresh
imap: implement STATUS command
imap: use Text::ParseWords::parse_line to handle quoted words
imap: support LIST command
t/imapd: support FakeInotify and KQNotify
imap: support fetch for BODYSTRUCTURE and BODY
eml: each_part: single part $idx is 1
imap: allow fetch of partial of BODY[...] and headers
imap: always include `resp-text' in responses
imap: split out unit tests and benchmarks
imap: fix multi-message partial header fetches
imap: simplify partial fetch structure
imap: support sequence number FETCH
imap: do not include ".PEEK" in responses
imap: support the CLOSE command
imap: speed up HEADER.FIELDS[.NOT] range fetches
git: async: flatten the inflight array
git: do our own read buffering for cat-file
imap: use git-cat-file asynchronously
git: idle rbuf for async
imap: support LSUB command
imap: FETCH: support comma-delimited ranges
add imapd compression test
testcommon: tcp_(server|connect): BAIL_OUT on failure
*deflate: drop invalid comment about rbuf
imap: fix pipelining with async git
git: cat_async: provide requested OID + "missing" on missing blobs
git: move async_cat reference to PublicInbox::Git
git: async: automatic retry on alternates change
imapclient: wrapper for Mail::IMAPClient
xt: add imapd-validate and imapd-mbsync-oimap
imap: support out-of-bounds ranges
xt/perf-imap-list: time refresh_inboxlist
imap: case-insensitive mailbox name comparisons
imap: break giant inboxes into sub-inboxes of 50K messages
imap: start introducing iterative config reloading
imap: require ".$UID_MIN-$UID_END" suffix
imapd: ensure LIST is sorted alphabetically, for now
imap: omit $UID_END from mailbox name, use index
t/config.t: always compare against git bool behavior
xt/*: show some tunable parameters
imap: STATUS and LIST are case-insensitive, too
imap: EXAMINE/STATUS: return correct counts
imap: avoid uninitialized warnings on incomplete commands
imap: start parsing out queries for SQLite and Xapian
imap: SEARCH: clamp results to the 50K UID range
imap: allow UID range search on timestamps
over: get_art: use dbh->prepare_cached
search: index byte size of a message for IMAP search
search: index UID for IMAP search, too
imap: remove dummies from sequence number FETCH
imap: compile UID FETCH to opcodes
imap: UID FETCH: optimize for smsg-only case
imap: UID FETCH: optimize (UID FLAGS) harder
imap: IDLE: avoid extraneous wakeups, keep-alive
imap: 30 minute auto-logout timer
imap: split ->logged_in attribute into a separate class
searchidx: v1 (re)-index uses git asynchronously
index: account for CRLF conversion when storing bytes
imap: rely on smsg->{bytes} for RFC822.SIZE
imap: UID FETCH requires at least one data item
imap: LIST shows "INBOX" in all caps
imap: support 8000 octet lines
imap: reinstate some message sequence number support
imap: cleanup ->{uid_base} usage
imap: FETCH: more granular CRLF conversion
imap: further speed up HEADER.FIELDS FETCH requests
imap: FETCH: try to make fake MSNs sequentially
imap: STATUS/EXAMINE: rely on SQLite overview
imap: UID SEARCH: support multiple ranges
imap: wire up Xapian search, msn SEARCH and multiple ranges
imap: misc cleanups and notes
imapd: don't bother sorting LIST output
imap: drop non-UID SEARCH for now
over: uid_range: remove LIMIT
imap: FETCH: proper MSN => UID mapping for requests
Documentation/public-inbox-imapd.pod | 91 ++
Documentation/standards.perl | 10 +
MANIFEST | 18 +
lib/PublicInbox/Config.pm | 18 +
lib/PublicInbox/Daemon.pm | 24 +-
lib/PublicInbox/DummyInbox.pm | 22 +
lib/PublicInbox/Eml.pm | 9 +-
lib/PublicInbox/FakeInotify.pm | 59 ++
lib/PublicInbox/Git.pm | 163 +--
lib/PublicInbox/GitAsyncCat.pm | 51 +
lib/PublicInbox/IMAP.pm | 1397 ++++++++++++++++++++++++++
lib/PublicInbox/IMAPClient.pm | 119 +++
lib/PublicInbox/IMAPD.pm | 114 +++
lib/PublicInbox/IMAPdeflate.pm | 126 +++
lib/PublicInbox/Import.pm | 2 +-
lib/PublicInbox/In2Tie.pm | 17 +
lib/PublicInbox/Inbox.pm | 33 +-
lib/PublicInbox/InboxIdle.pm | 79 ++
lib/PublicInbox/KQNotify.pm | 66 ++
lib/PublicInbox/Lock.pm | 7 +
lib/PublicInbox/MsgIter.pm | 2 +-
lib/PublicInbox/Msgmap.pm | 20 +-
lib/PublicInbox/NNTPD.pm | 12 +-
lib/PublicInbox/NNTPdeflate.pm | 1 -
lib/PublicInbox/Over.pm | 50 +-
lib/PublicInbox/Search.pm | 32 +-
lib/PublicInbox/SearchIdx.pm | 89 +-
lib/PublicInbox/SearchIdxShard.pm | 11 +-
lib/PublicInbox/Smsg.pm | 8 +-
lib/PublicInbox/TestCommon.pm | 7 +-
lib/PublicInbox/V2Writable.pm | 10 +-
script/public-inbox-imapd | 14 +
t/config.t | 15 +-
t/eml.t | 2 +-
t/git.t | 40 +-
t/imap.t | 133 +++
t/imapd-tls.t | 204 ++++
t/imapd.t | 398 ++++++++
t/import.t | 5 +-
t/inbox_idle.t | 72 ++
t/nntpd.t | 5 +-
t/over.t | 3 +
t/search.t | 19 +
xt/cmp-msgstr.t | 1 -
xt/cmp-msgview.t | 1 -
xt/eml_check_limits.t | 6 +-
xt/git_async_cmp.t | 2 +-
xt/imapd-mbsync-oimap.t | 132 +++
xt/imapd-validate.t | 177 ++++
xt/mem-msgview.t | 1 +
xt/msgtime_cmp.t | 1 -
xt/perf-msgview.t | 1 -
52 files changed, 3718 insertions(+), 181 deletions(-)
create mode 100644 Documentation/public-inbox-imapd.pod
create mode 100644 lib/PublicInbox/DummyInbox.pm
create mode 100644 lib/PublicInbox/FakeInotify.pm
create mode 100644 lib/PublicInbox/GitAsyncCat.pm
create mode 100644 lib/PublicInbox/IMAP.pm
create mode 100644 lib/PublicInbox/IMAPClient.pm
create mode 100644 lib/PublicInbox/IMAPD.pm
create mode 100644 lib/PublicInbox/IMAPdeflate.pm
create mode 100644 lib/PublicInbox/In2Tie.pm
create mode 100644 lib/PublicInbox/InboxIdle.pm
create mode 100644 lib/PublicInbox/KQNotify.pm
create mode 100644 script/public-inbox-imapd
create mode 100644 t/imap.t
create mode 100644 t/imapd-tls.t
create mode 100644 t/imapd.t
create mode 100644 t/inbox_idle.t
create mode 100644 xt/imapd-mbsync-oimap.t
create mode 100644 xt/imapd-validate.t
^ 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 --
2020-06-10 7:03 6% [PATCH 00/82] public-inbox-imapd: read-only IMAP server Eric Wong
2020-06-10 7:05 7% ` [PATCH 64/82] searchidx: v1 (re)-index uses git asynchronously 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).