user/dev discussion of public-inbox itself
 help / color / mirror / code / Atom feed
From: Eric Wong <e@yhbt.net>
To: meta@public-inbox.org
Subject: [PATCH 24/82] git: do our own read buffering for cat-file
Date: Wed, 10 Jun 2020 07:04:21 +0000	[thread overview]
Message-ID: <20200610070519.18252-25-e@yhbt.net> (raw)
In-Reply-To: <20200610070519.18252-1-e@yhbt.net>

To work with our event loop, we must perform read buffering
ourselves or risk starvation, as there doesn't appear to be
a way to check the amount of data buffered in userspace by
by the PerlIO layers without resorting to C or XS.

This lets us perform fewer syscalls at the expense of more Perl
ops.  As it stands, there seems to be a tiny performance
improvement, but more will be possible in the future.
---
 lib/PublicInbox/Git.pm | 82 +++++++++++++++++++++++++++---------------
 1 file changed, 54 insertions(+), 28 deletions(-)

diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index e1d5c386e7e..54c163e8c2f 100644
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -16,6 +16,8 @@ use PublicInbox::Spawn qw(popen_rd);
 use PublicInbox::Tmpfile;
 use base qw(Exporter);
 our @EXPORT_OK = qw(git_unquote git_quote);
+use Errno qw(EINTR);
+our $PIPE_BUFSIZ = 65536; # Linux default
 
 use constant MAX_INFLIGHT =>
 	(($^O eq 'linux' ? 4096 : POSIX::_POSIX_PIPE_BUF()) * 2)
@@ -121,32 +123,54 @@ sub _bidi_pipe {
 		fcntl($out_w, 1031, 4096);
 		fcntl($in_r, 1031, 4096) if $batch eq '--batch-check';
 	}
+	$self->{$batch} = \(my $rbuf = '');
 	$self->{$out} = $out_w;
 	$self->{$in} = $in_r;
 }
 
-sub read_cat_in_full ($$) {
-	my ($self, $len) = @_;
-	++$len; # for final "\n" added by git
-	read($self->{in}, my $buf, $len) == $len or fail($self, 'short read');
-	chop($buf) eq "\n" or fail($self, 'newline missing after blob');
-	\$buf;
+sub my_read ($$$) {
+	my ($fh, $rbuf, $len) = @_;
+	my $left = $len - length($$rbuf);
+	my $r;
+	while ($left > 0) {
+		$r = sysread($fh, $$rbuf, $PIPE_BUFSIZ, length($$rbuf));
+		if ($r) {
+			$left -= $r;
+		} else {
+			next if (!defined($r) && $! == EINTR);
+			return $r;
+		}
+	}
+	\substr($$rbuf, 0, $len, '');
+}
+
+sub my_readline ($$) {
+	my ($fh, $rbuf) = @_;
+	while (1) {
+		if ((my $n = index($$rbuf, "\n")) >= 0) {
+			return substr($$rbuf, 0, $n + 1, '');
+		}
+		my $r = sysread($fh, $$rbuf, $PIPE_BUFSIZ, length($$rbuf));
+		next if $r || (!defined($r) && $! == EINTR);
+		return defined($r) ? '' : undef; # EOF or error
+	}
 }
 
 sub _cat_async_step ($$) {
 	my ($self, $inflight) = @_;
 	die 'BUG: inflight empty or odd' if scalar(@$inflight) < 2;
 	my ($cb, $arg) = splice(@$inflight, 0, 2);
-	local $/ = "\n";
-	my $head = readline($self->{in});
+	my $head = my_readline($self->{in}, $self->{'--batch'});
 	$head =~ / missing$/ and return
 		eval { $cb->(undef, undef, undef, undef, $arg) };
 
 	$head =~ /^([0-9a-f]{40}) (\S+) ([0-9]+)$/ or
 		fail($self, "Unexpected result from async git cat-file: $head");
 	my ($oid_hex, $type, $size) = ($1, $2, $3 + 0);
-	my $bref = read_cat_in_full($self, $size);
-	eval { $cb->($bref, $oid_hex, $type, $size, $arg) };
+	my $ret = my_read($self->{in}, $self->{'--batch'}, $size + 1);
+	fail($self, defined($ret) ? 'read EOF' : "read: $!") if !$ret;
+	chop($$ret) eq "\n" or fail($self, 'newline missing after blob');
+	eval { $cb->($ret, $oid_hex, $type, $size, $arg) };
 	warn "E: $oid_hex $@\n" if $@;
 }
 
@@ -158,16 +182,18 @@ sub cat_async_wait ($) {
 	}
 }
 
+sub batch_prepare ($) {
+	_bidi_pipe($_[0], qw(--batch in out pid));
+}
+
 sub cat_file {
-	my ($self, $obj, $ref) = @_;
+	my ($self, $obj, $sizeref) = @_;
 	my ($retried, $head);
 	cat_async_wait($self);
 again:
 	batch_prepare($self);
 	print { $self->{out} } $obj, "\n" or fail($self, "write error: $!");
-
-	local $/ = "\n";
-	$head = readline($self->{in});
+	$head = my_readline($self->{in}, $self->{'--batch'});
 	if ($head =~ / missing$/) {
 		if (!$retried && alternates_changed($self)) {
 			$retried = 1;
@@ -179,19 +205,19 @@ again:
 	$head =~ /^[0-9a-f]{40} \S+ ([0-9]+)$/ or
 		fail($self, "Unexpected result from git cat-file: $head");
 
-	my $size = $1;
-	$$ref = $size if $ref;
-	read_cat_in_full($self, $size);
+	my $size = $1 + 0;
+	$$sizeref = $size if $sizeref;
+	my $ret = my_read($self->{in}, $self->{'--batch'}, $size + 1);
+	fail($self, defined($ret) ? 'read EOF' : "read: $!") if !$ret;
+	chop($$ret) eq "\n" or fail($self, 'newline missing after blob');
+	$ret;
 }
 
-sub batch_prepare ($) { _bidi_pipe($_[0], qw(--batch in out pid)) }
-
 sub check {
 	my ($self, $obj) = @_;
 	_bidi_pipe($self, qw(--batch-check in_c out_c pid_c err_c));
 	print { $self->{out_c} } $obj, "\n" or fail($self, "write error: $!");
-	local $/ = "\n";
-	chomp(my $line = readline($self->{in_c}));
+	chomp(my $line = my_readline($self->{in_c}, $self->{'--batch-check'}));
 	my ($hex, $type, $size) = split(' ', $line);
 
 	# Future versions of git.git may show 'ambiguous', but for now,
@@ -201,9 +227,9 @@ sub check {
 	return if $type eq 'missing' || $type eq 'ambiguous';
 
 	if ($hex eq 'dangling' || $hex eq 'notdir' || $hex eq 'loop') {
-		$size = $type + length("\n");
-		my $r = read($self->{in_c}, my $buf, $size);
-		defined($r) or fail($self, "read failed: $!");
+		my $ret = my_read($self->{in_c}, $self->{'--batch-check'},
+				$type + 1);
+		fail($self, defined($ret) ? 'read EOF' : "read: $!") if !$ret;
 		return;
 	}
 
@@ -211,9 +237,9 @@ sub check {
 }
 
 sub _destroy {
-	my ($self, $in, $out, $pid, $err) = @_;
+	my ($self, $batch, $in, $out, $pid, $err) = @_;
 	my $p = delete $self->{$pid} or return;
-	delete @$self{($in, $out)};
+	delete @$self{($batch, $in, $out)};
 	delete $self->{$err} if $err; # `err_c'
 
 	# PublicInbox::DS may not be loaded
@@ -251,8 +277,8 @@ sub qx {
 # returns true if there are pending "git cat-file" processes
 sub cleanup {
 	my ($self) = @_;
-	_destroy($self, qw(in out pid));
-	_destroy($self, qw(in_c out_c pid_c err_c));
+	_destroy($self, qw(--batch in out pid));
+	_destroy($self, qw(--batch-check in_c out_c pid_c err_c));
 	!!($self->{pid} || $self->{pid_c});
 }
 

  parent reply	other threads:[~2020-06-10  7:05 UTC|newest]

Thread overview: 84+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-06-10  7:03 [PATCH 00/82] public-inbox-imapd: read-only IMAP server Eric Wong
2020-06-10  7:03 ` [PATCH 01/82] doc: add some IMAP standards Eric Wong
2020-06-10  7:03 ` [PATCH 02/82] nntpd: restrict allowed newsgroup names Eric Wong
2020-06-10  7:04 ` [PATCH 03/82] preliminary imap server implementation Eric Wong
2020-06-10  7:04 ` [PATCH 04/82] inboxidle: new class to detect inbox changes Eric Wong
2020-06-10  7:04 ` [PATCH 05/82] imap: support IDLE Eric Wong
2020-06-10  7:04 ` [PATCH 06/82] msgmap: split ->max into its own method Eric Wong
2020-06-10  7:04 ` [PATCH 07/82] imap: delay InboxIdle start, support refresh Eric Wong
2020-06-10  7:04 ` [PATCH 08/82] imap: implement STATUS command Eric Wong
2020-06-10  7:04 ` [PATCH 09/82] imap: use Text::ParseWords::parse_line to handle quoted words Eric Wong
2020-06-10  7:04 ` [PATCH 10/82] imap: support LIST command Eric Wong
2020-06-10  7:04 ` [PATCH 11/82] t/imapd: support FakeInotify and KQNotify Eric Wong
2020-06-10  7:04 ` [PATCH 12/82] imap: support fetch for BODYSTRUCTURE and BODY Eric Wong
2020-06-10  7:04 ` [PATCH 13/82] eml: each_part: single part $idx is 1 Eric Wong
2020-06-10  7:04 ` [PATCH 14/82] imap: allow fetch of partial of BODY[...] and headers Eric Wong
2020-06-10  7:04 ` [PATCH 15/82] imap: always include `resp-text' in responses Eric Wong
2020-06-10  7:04 ` [PATCH 16/82] imap: split out unit tests and benchmarks Eric Wong
2020-06-10  7:04 ` [PATCH 17/82] imap: fix multi-message partial header fetches Eric Wong
2020-06-10  7:04 ` [PATCH 18/82] imap: simplify partial fetch structure Eric Wong
2020-06-10  7:04 ` [PATCH 19/82] imap: support sequence number FETCH Eric Wong
2020-06-10  7:04 ` [PATCH 20/82] imap: do not include ".PEEK" in responses Eric Wong
2020-06-10  7:04 ` [PATCH 21/82] imap: support the CLOSE command Eric Wong
2020-06-10  7:04 ` [PATCH 22/82] imap: speed up HEADER.FIELDS[.NOT] range fetches Eric Wong
2020-06-10  7:04 ` [PATCH 23/82] git: async: flatten the inflight array Eric Wong
2020-06-10  7:04 ` Eric Wong [this message]
2020-06-10  7:04 ` [PATCH 25/82] imap: use git-cat-file asynchronously Eric Wong
2020-06-10  7:04 ` [PATCH 26/82] git: idle rbuf for async Eric Wong
2020-06-10  7:04 ` [PATCH 27/82] imap: support LSUB command Eric Wong
2020-06-10  7:04 ` [PATCH 28/82] imap: FETCH: support comma-delimited ranges Eric Wong
2020-06-10  7:04 ` [PATCH 29/82] add imapd compression test Eric Wong
2020-06-10  7:04 ` [PATCH 30/82] testcommon: tcp_(server|connect): BAIL_OUT on failure Eric Wong
2020-06-10  7:04 ` [PATCH 31/82] *deflate: drop invalid comment about rbuf Eric Wong
2020-06-10  7:04 ` [PATCH 32/82] imap: fix pipelining with async git Eric Wong
2020-06-10  7:04 ` [PATCH 33/82] git: cat_async: provide requested OID + "missing" on missing blobs Eric Wong
2020-06-10  7:04 ` [PATCH 34/82] git: move async_cat reference to PublicInbox::Git Eric Wong
2020-06-10  7:04 ` [PATCH 35/82] git: async: automatic retry on alternates change Eric Wong
2020-06-10  7:04 ` [PATCH 36/82] imapclient: wrapper for Mail::IMAPClient Eric Wong
2020-06-10  7:04 ` [PATCH 37/82] xt: add imapd-validate and imapd-mbsync-oimap Eric Wong
2020-06-10  7:04 ` [PATCH 38/82] imap: support out-of-bounds ranges Eric Wong
2020-06-10  7:04 ` [PATCH 39/82] xt/perf-imap-list: time refresh_inboxlist Eric Wong
2020-06-10  7:04 ` [PATCH 40/82] imap: case-insensitive mailbox name comparisons Eric Wong
2020-06-10  7:04 ` [PATCH 41/82] imap: break giant inboxes into sub-inboxes of 50K messages Eric Wong
2020-06-10  7:04 ` [PATCH 42/82] imap: start doing iterative config reloading Eric Wong
2020-06-10  7:04 ` [PATCH 43/82] imap: require ".$UID_MIN-$UID_END" suffix Eric Wong
2020-06-10  7:04 ` [PATCH 44/82] imapd: ensure LIST is sorted alphabetically, for now Eric Wong
2020-06-10  7:04 ` [PATCH 45/82] imap: omit $UID_END from mailbox name, use index Eric Wong
2020-06-10  7:04 ` [PATCH 46/82] t/config.t: always compare against git bool behavior Eric Wong
2020-06-10  7:04 ` [PATCH 47/82] xt/*: show some tunable parameters Eric Wong
2020-06-10  7:04 ` [PATCH 48/82] imap: STATUS and LIST are case-insensitive, too Eric Wong
2020-06-10  7:04 ` [PATCH 49/82] imap: EXAMINE/STATUS: return correct counts Eric Wong
2020-06-10  7:04 ` [PATCH 50/82] imap: avoid uninitialized warnings on incomplete commands Eric Wong
2020-06-10  7:04 ` [PATCH 51/82] imap: start parsing out queries for SQLite and Xapian Eric Wong
2020-06-10  7:04 ` [PATCH 52/82] imap: SEARCH: clamp results to the 50K UID range Eric Wong
2020-06-10  7:04 ` [PATCH 53/82] imap: allow UID range search on timestamps Eric Wong
2020-06-10  7:04 ` [PATCH 54/82] over: get_art: use dbh->prepare_cached Eric Wong
2020-06-10  7:04 ` [PATCH 55/82] search: index byte size of a message for IMAP search Eric Wong
2020-06-10  7:04 ` [PATCH 56/82] search: index UID for IMAP search, too Eric Wong
2020-06-10  7:04 ` [PATCH 57/82] imap: remove dummies from sequence number FETCH Eric Wong
2020-06-10  7:04 ` [PATCH 58/82] imap: compile UID FETCH to opcodes Eric Wong
2020-06-10  7:04 ` [PATCH 59/82] imap: UID FETCH: optimize for smsg-only case Eric Wong
2020-06-10  7:04 ` [PATCH 60/82] imap: UID FETCH: optimize (UID FLAGS) harder Eric Wong
2020-06-10  7:04 ` [PATCH 61/82] imap: IDLE: avoid extraneous wakeups, keep-alive Eric Wong
2020-06-10  7:04 ` [PATCH 62/82] imap: 30 minute auto-logout timer Eric Wong
2020-06-10  7:05 ` [PATCH 63/82] imap: split ->logged_in attribute into a separate class Eric Wong
2020-06-10  7:05 ` [PATCH 64/82] searchidx: v1 (re)-index uses git asynchronously Eric Wong
2020-06-10  7:05 ` [PATCH 65/82] index: account for CRLF conversion when storing bytes Eric Wong
2020-06-10  7:05 ` [PATCH 66/82] imap: rely on smsg->{bytes} for RFC822.SIZE Eric Wong
2020-06-10  7:05 ` [PATCH 67/82] imap: UID FETCH requires at least one data item Eric Wong
2020-06-10  7:05 ` [PATCH 68/82] imap: LIST shows "INBOX" in all caps Eric Wong
2020-06-10  7:05 ` [PATCH 69/82] imap: support 8000 octet lines Eric Wong
2020-06-10  7:05 ` [PATCH 70/82] imap: reinstate some message sequence number support Eric Wong
2020-06-10  7:05 ` [PATCH 71/82] imap: cleanup ->{uid_base} usage Eric Wong
2020-06-10  7:05 ` [PATCH 72/82] imap: FETCH: more granular CRLF conversion Eric Wong
2020-06-10  7:05 ` [PATCH 73/82] imap: further speed up HEADER.FIELDS FETCH requests Eric Wong
2020-06-10  7:05 ` [PATCH 74/82] imap: FETCH: try to make fake MSNs sequentially Eric Wong
2020-06-10  7:05 ` [PATCH 75/82] imap: STATUS/EXAMINE: rely on SQLite overview Eric Wong
2020-06-10  7:05 ` [PATCH 76/82] imap: UID SEARCH: support multiple ranges Eric Wong
2020-06-10  7:05 ` [PATCH 77/82] imap: wire up Xapian, MSN SEARCH and multi sequence-sets Eric Wong
2020-06-10  7:05 ` [PATCH 78/82] imap: misc cleanups and notes Eric Wong
2020-06-10  7:05 ` [PATCH 79/82] imapd: don't bother sorting LIST output Eric Wong
2020-06-10  7:05 ` [PATCH 80/82] imap: remove non-UID SEARCH for now Eric Wong
2020-06-10  7:05 ` [PATCH 81/82] over: uid_range: remove LIMIT Eric Wong
2020-06-10  7:05 ` [PATCH 82/82] imap: FETCH: proper MSN => UID mapping for requests Eric Wong
2020-06-12 23:49 ` [PATCH 83/82] imap: introduce memory-efficient uo2m mapping Eric Wong

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: https://public-inbox.org/README

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200610070519.18252-25-e@yhbt.net \
    --to=e@yhbt.net \
    --cc=meta@public-inbox.org \
    --subject='Re: [PATCH 24/82] git: do our own read buffering for cat-file' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Code repositories for project(s) associated with this 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).