From: Eric Wong <e@80x24.org>
To: meta@public-inbox.org
Subject: [PATCH 02/21] ipc: switch wq to use the event loop
Date: Sun, 31 Jan 2021 22:28:14 -1000 [thread overview]
Message-ID: <20210201082833.3293-3-e@80x24.org> (raw)
In-Reply-To: <20210201082833.3293-1-e@80x24.org>
This will let us to maximize the capability of our asynchronous
git API. This lets us avoid relying on EOF to notify lei2mail
workers; thus giving us the option of running fewer lei_xsearch
worker processes in parallel than local sources.
I tried using a synchronous git API; and even with libgit2 in
the same process to avoid the IPC cost failed to match the
throughput afforded by this change. This is because libgit2 is
built (at least on Debian) with the SHA-1 collision code enabled
and ubc_check stuff was dominating my profiles.
---
MANIFEST | 1 +
lib/PublicInbox/IPC.pm | 17 +++++++++++------
lib/PublicInbox/LeiToMail.pm | 26 +++++++++++---------------
lib/PublicInbox/WQWorker.pm | 34 ++++++++++++++++++++++++++++++++++
4 files changed, 57 insertions(+), 21 deletions(-)
create mode 100644 lib/PublicInbox/WQWorker.pm
diff --git a/MANIFEST b/MANIFEST
index 2077ab12..c10775e4 100644
--- a/MANIFEST
+++ b/MANIFEST
@@ -228,6 +228,7 @@ lib/PublicInbox/V2Writable.pm
lib/PublicInbox/View.pm
lib/PublicInbox/ViewDiff.pm
lib/PublicInbox/ViewVCS.pm
+lib/PublicInbox/WQWorker.pm
lib/PublicInbox/WWW.pm
lib/PublicInbox/WWW.pod
lib/PublicInbox/Watch.pm
diff --git a/lib/PublicInbox/IPC.pm b/lib/PublicInbox/IPC.pm
index d2ff038d..479c4377 100644
--- a/lib/PublicInbox/IPC.pm
+++ b/lib/PublicInbox/IPC.pm
@@ -14,6 +14,7 @@ use Carp qw(confess croak);
use PublicInbox::DS qw(dwaitpid);
use PublicInbox::Spawn;
use PublicInbox::OnDestroy;
+use PublicInbox::WQWorker;
use Socket qw(AF_UNIX MSG_EOR SOCK_STREAM);
use Errno qw(EMSGSIZE);
my $SEQPACKET = eval { Socket::SOCK_SEQPACKET() }; # portable enough?
@@ -151,6 +152,8 @@ sub wq_wait_old {
# for base class, override in sub classes
sub ipc_atfork_prepare {}
+sub wq_atexit_child {}
+
sub ipc_atfork_child {
my ($self) = @_;
my $io = delete($self->{-ipc_atfork_child_close}) or return;
@@ -251,10 +254,11 @@ sub ipc_sibling_atfork_child {
$pid == $$ and die "BUG: $$ ipc_atfork_child called on itself";
}
-sub _recv_and_run {
+sub recv_and_run {
my ($self, $s2, $len, $full_stream) = @_;
my @fds = $recv_cmd->($s2, my $buf, $len);
- my $n = length($buf // '') or return;
+ return if scalar(@fds) && !defined($fds[0]);
+ my $n = length($buf) or return 0;
my $nfd = 0;
for my $fd (@fds) {
if (open(my $cmdfh, '+<&=', $fd)) {
@@ -281,14 +285,15 @@ sub _recv_and_run {
sub wq_worker_loop ($) {
my ($self) = @_;
- my $len = $self->{wq_req_len} // (4096 * 33);
- my $s2 = $self->{-wq_s2} // die 'BUG: no -wq_s2';
- 1 while (_recv_and_run($self, $s2, $len));
+ my $wqw = PublicInbox::WQWorker->new($self);
+ PublicInbox::DS->SetPostLoopCallback(sub { $wqw->{sock} });
+ PublicInbox::DS->EventLoop;
+ PublicInbox::DS->Reset;
}
sub do_sock_stream { # via wq_do, for big requests
my ($self, $len) = @_;
- _recv_and_run($self, delete $self->{0}, $len, 1);
+ recv_and_run($self, delete $self->{0}, $len, 1);
}
sub wq_do { # always async
diff --git a/lib/PublicInbox/LeiToMail.pm b/lib/PublicInbox/LeiToMail.pm
index 244bfb67..1f6c2a3b 100644
--- a/lib/PublicInbox/LeiToMail.pm
+++ b/lib/PublicInbox/LeiToMail.pm
@@ -12,11 +12,16 @@ use PublicInbox::ProcessPipe;
use PublicInbox::Spawn qw(which spawn popen_rd);
use PublicInbox::LeiDedupe;
use PublicInbox::OnDestroy;
+use PublicInbox::Git;
+use PublicInbox::GitAsyncCat;
use Symbol qw(gensym);
use IO::Handle; # ->autoflush
use Fcntl qw(SEEK_SET SEEK_END O_CREAT O_EXCL O_WRONLY);
use Errno qw(EEXIST ESPIPE ENOENT);
-use PublicInbox::Git;
+
+# struggles with short-lived repos, Gcf2Client makes little sense with lei;
+# but we may use in-process libgit2 in the future.
+$PublicInbox::GitAsyncCat::GCF2C = 0;
my %kw2char = ( # Maildir characters
draft => 'D',
@@ -467,27 +472,18 @@ sub write_mail { # via ->wq_do
$self->write_cb($lei);
};
my $git = $self->{"$$\0$git_dir"} //= PublicInbox::Git->new($git_dir);
- $git->cat_async($smsg->{blob}, \&git_to_mail, [$wcb, $smsg, $not_done]);
+ git_async_cat($git, $smsg->{blob}, \&git_to_mail,
+ [$wcb, $smsg, $not_done]);
}
-# We rely on OnDestroy to run this before ->DESTROY, since ->DESTROY
-# ordering is unstable at worker exit and may cause segfaults
-sub reap_gits {
+sub wq_atexit_child {
my ($self) = @_;
delete $self->{wcb};
for my $git (delete @$self{grep(/\A$$\0/, keys %$self)}) {
$git->async_wait_all;
}
-}
-
-sub DESTROY { delete $_[0]->{wcb} }
-
-sub ipc_atfork_child { # runs after IPC::wq_worker_loop
- my ($self) = @_;
- $self->SUPER::ipc_atfork_child;
- # reap_gits needs to run before $self->DESTROY,
- # IPC.pm will ensure that.
- PublicInbox::OnDestroy->new($$, \&reap_gits, $self);
+ $SIG{__WARN__} = 'DEFAULT';
+ $SIG{PIPE} = 'DEFAULT';
}
1;
diff --git a/lib/PublicInbox/WQWorker.pm b/lib/PublicInbox/WQWorker.pm
new file mode 100644
index 00000000..25a5e4fb
--- /dev/null
+++ b/lib/PublicInbox/WQWorker.pm
@@ -0,0 +1,34 @@
+# Copyright (C) 2021 all contributors <meta@public-inbox.org>
+# License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
+
+# for PublicInbox::IPC wq_* (work queue) workers
+package PublicInbox::WQWorker;
+use strict;
+use v5.10.1;
+use parent qw(PublicInbox::DS);
+use PublicInbox::Syscall qw(EPOLLIN EPOLLEXCLUSIVE EPOLLET);
+use Errno qw(EAGAIN ECONNRESET);
+use IO::Handle (); # blocking
+
+sub new {
+ my (undef, $wq) = @_;
+ my $s2 = $wq->{-wq_s2} // die 'BUG: no -wq_s2';
+ $s2->blocking(0);
+ my $self = bless { sock => $s2, wq => $wq }, __PACKAGE__;
+ $self->SUPER::new($s2, EPOLLEXCLUSIVE|EPOLLIN|EPOLLET);
+ $self;
+}
+
+sub event_step {
+ my ($self) = @_;
+ my $n;
+ do {
+ $n = $self->{wq}->recv_and_run($self->{sock}, 4096 * 33);
+ } while ($n);
+ return if !defined($n) && $! == EAGAIN; # likely
+ warn "wq worker error: $!\n" if !defined($n) && $! != ECONNRESET;
+ $self->{wq}->wq_atexit_child;
+ $self->close; # PublicInbox::DS::close
+}
+
+1;
next prev parent reply other threads:[~2021-02-01 8:28 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-02-01 8:28 [PATCH 00/21] lei2mail worker segfault finally fixed Eric Wong
2021-02-01 8:28 ` [PATCH 01/21] lei: more consistent dedupe and ovv_buf init Eric Wong
2021-02-01 8:28 ` Eric Wong [this message]
2021-02-01 8:28 ` [PATCH 03/21] lei: remove per-child SIG{__WARN__} Eric Wong
2021-02-01 8:28 ` [PATCH 04/21] lei: remove SIGPIPE handler Eric Wong
2021-02-01 8:28 ` [PATCH 05/21] ipc: more helpful ETOOMANYREFS error messages Eric Wong
2021-02-01 8:28 ` [PATCH 06/21] lei: remove syslog dependency Eric Wong
2021-02-01 8:28 ` [PATCH 07/21] sharedkv: release {dbh} before rmtree Eric Wong
2021-02-01 8:28 ` [PATCH 08/21] lei: keep $lei around until workers are reaped Eric Wong
2021-02-01 8:28 ` [PATCH 09/21] lei_dedupe: use Digest::SHA Eric Wong
2021-02-01 8:28 ` [PATCH 10/21] lei_xsearch: load PublicInbox::Smsg Eric Wong
2021-02-01 8:28 ` [PATCH 11/21] lei: deep clone {ovv} for l2m workers Eric Wong
2021-02-01 8:28 ` [PATCH 12/21] sharedkv: lock and explicitly disconnect {dbh} Eric Wong
2021-02-01 8:28 ` [PATCH 13/21] lei: increase initial timeout Eric Wong
2021-02-01 8:28 ` [PATCH 14/21] sharedkv: use lock_for_scope_fast Eric Wong
2021-02-01 8:28 ` [PATCH 15/21] lei_to_mail: reduce spew on Maildir removal Eric Wong
2021-02-01 8:28 ` [PATCH 16/21] sharedkv: do not set cache_size by default Eric Wong
2021-02-01 8:28 ` [PATCH 17/21] import: reap git-config(1) synchronously Eric Wong
2021-02-01 8:28 ` [PATCH 18/21] ds: guard against stack-not-refcounted quirk of Perl 5 Eric Wong
2021-02-01 9:07 ` Perl debug patches used to track down source of segfault Eric Wong
2021-02-01 8:28 ` [PATCH 19/21] ds: next_tick: avoid $_ in top-level loop iterator Eric Wong
2021-02-01 8:28 ` [PATCH 20/21] lei: avoid ETOOMANYREFS, cleanup imports Eric Wong
2021-02-01 8:28 ` [PATCH 21/21] doc: note optional BSD::Resource use 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=20210201082833.3293-3-e@80x24.org \
--to=e@80x24.org \
--cc=meta@public-inbox.org \
/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
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).