user/dev discussion of public-inbox itself
 help / color / mirror / code / Atom feed
Search results ordered by [date|relevance]  view[summary|nested|Atom feed]
thread overview below | download mbox.gz: |
* [PATCH 20/21] lei: avoid ETOOMANYREFS, cleanup imports
  2021-02-01  8:28  7% [PATCH 00/21] lei2mail worker segfault finally fixed Eric Wong
@ 2021-02-01  8:28  6% ` Eric Wong
  0 siblings, 0 replies; 2+ results
From: Eric Wong @ 2021-02-01  8:28 UTC (permalink / raw)
  To: meta

As with PublicInbox::IPC, we'll attempt to bump RLIMIT_NOFILE
and transparently workaround ETOOMANYREFS.  If that fails,
we'll give the user a hint to bump RLIMIT_NOFILE since
ETOOMANYREFS is an uncommon error which users may be unfamiliar
with.

Found while stress testing for segfaults.
---
 script/lei | 29 ++++++++++++++++++++---------
 1 file changed, 20 insertions(+), 9 deletions(-)

diff --git a/script/lei b/script/lei
index f92dd302..58f0dbe9 100755
--- a/script/lei
+++ b/script/lei
@@ -4,10 +4,9 @@
 use strict;
 use v5.10.1;
 use Socket qw(AF_UNIX SOCK_SEQPACKET MSG_EOR pack_sockaddr_un);
-use Errno qw(EINTR ECONNRESET);
 use PublicInbox::CmdIPC4;
 my $narg = 5;
-my ($sock, $pwd);
+my $sock;
 my $recv_cmd = PublicInbox::CmdIPC4->can('recv_cmd4');
 my $send_cmd = PublicInbox::CmdIPC4->can('send_cmd4') // do {
 	require PublicInbox::Spawn; # takes ~50ms even if built *sigh*
@@ -73,20 +72,32 @@ connect($path): $! (after attempted daemon start)
 Falling back to (slow) one-shot mode
 
 	}
-	1;
-}) { # (Socket::MsgHdr|Inline::C), $sock, $pwd are all available:
+	# (Socket::MsgHdr|Inline::C), $sock are all available:
 	open my $dh, '<', '.' or die "open(.) $!";
 	my $buf = join("\0", scalar(@ARGV), @ARGV);
 	while (my ($k, $v) = each %ENV) { $buf .= "\0$k=$v" }
 	$buf .= "\0\0";
-	$send_cmd->($sock, [ 0, 1, 2, fileno($dh) ], $buf, MSG_EOR) or
-		die "sendmsg: $!";
+	my $n = $send_cmd->($sock, [0, 1, 2, fileno($dh)], $buf, MSG_EOR);
+	if (!$n && $!{ETOOMANYREFS} && eval { require BSD::Resource }) {
+		my $NOFILE = BSD::Resource::RLIMIT_NOFILE();
+		my ($s, $h) = BSD::Resource::getrlimit($NOFILE);
+		if ($s < $h && BSD::Resource::setrlimit($NOFILE, $h, $h)) {
+			$n = $send_cmd->($sock, [0, 1, 2, fileno($dh)],
+					$buf, MSG_EOR);
+		}
+	}
+	if (!$n) {
+		die "sendmsg: $! (check RLIMIT_NOFILE)\n" if $!{ETOOMANYREFS};
+		die "sendmsg: $!\n";
+	}
+	1;
+}) { # connected and request sent to lei-daemon, wait for responses or EOF
 	my $x_it_code = 0;
 	while (1) {
-		my (@fds) = $recv_cmd->($sock, $buf, 4096 * 33);
+		my (@fds) = $recv_cmd->($sock, my $buf, 4096 * 33);
 		if (scalar(@fds) == 1 && !defined($fds[0])) {
-			last if $! == ECONNRESET;
-			next if $! == EINTR;
+			next if $!{EINTR};
+			last if $!{ECONNRESET};
 			die "recvmsg: $!";
 		}
 		last if $buf eq '';

^ permalink raw reply related	[relevance 6%]

* [PATCH 00/21] lei2mail worker segfault finally fixed
@ 2021-02-01  8:28  7% Eric Wong
  2021-02-01  8:28  6% ` [PATCH 20/21] lei: avoid ETOOMANYREFS, cleanup imports Eric Wong
  0 siblings, 1 reply; 2+ results
From: Eric Wong @ 2021-02-01  8:28 UTC (permalink / raw)
  To: meta

This lei2mail segfault turned out to be an old Perl 5 quirk
which plagued many before me.  It was not consistently
reproducible, and random changes seemed to make it happen more
or less frequently.  There were several times when I thought I
fixed it (and maybe this is still one of them!) only to have it
pop up again.

Still, I found many other little bugs and improvements worth
doing along the way.  Hope things go more smoothly in the
future...

Anyways, [PATCH 18/21] is the fix (and I'll followup with more
on how I found the fix).  19/21 is purely defensive
future-proofing.

Eric Wong (21):
  lei: more consistent dedupe and ovv_buf init
  ipc: switch wq to use the event loop
  lei: remove per-child SIG{__WARN__}
  lei: remove SIGPIPE handler
  ipc: more helpful ETOOMANYREFS error messages
  lei: remove syslog dependency
  sharedkv: release {dbh} before rmtree
  lei: keep $lei around until workers are reaped
  lei_dedupe: use Digest::SHA
  lei_xsearch: load PublicInbox::Smsg
  lei: deep clone {ovv} for l2m workers
  sharedkv: lock and explicitly disconnect {dbh}
  lei: increase initial timeout
  sharedkv: use lock_for_scope_fast
  lei_to_mail: reduce spew on Maildir removal
  sharedkv: do not set cache_size by default
  import: reap git-config(1) synchronously
  ds: guard against stack-not-refcounted quirk of Perl 5
  ds: next_tick: avoid $_ in top-level loop iterator
  lei: avoid ETOOMANYREFS, cleanup imports
  doc: note optional BSD::Resource use

 Documentation/public-inbox-config.pod |  2 +-
 INSTALL                               |  6 ++
 MANIFEST                              |  2 +
 lib/PublicInbox/DS.pm                 | 12 ++--
 lib/PublicInbox/IPC.pm                | 43 +++++++-----
 lib/PublicInbox/Import.pm             |  1 +
 lib/PublicInbox/LEI.pm                | 95 +++++++++++++++------------
 lib/PublicInbox/LeiDedupe.pm          |  6 +-
 lib/PublicInbox/LeiExternal.pm        |  3 +-
 lib/PublicInbox/LeiOverview.pm        | 51 +++++++-------
 lib/PublicInbox/LeiToMail.pm          | 84 +++++++++++------------
 lib/PublicInbox/LeiXSearch.pm         | 36 +++++-----
 lib/PublicInbox/Lock.pm               | 17 +++++
 lib/PublicInbox/SharedKV.pm           | 33 +++++++---
 lib/PublicInbox/WQWorker.pm           | 34 ++++++++++
 script/lei                            | 28 +++++---
 t/lei_to_mail.t                       | 31 +++++----
 xt/stress-sharedkv.t                  | 50 ++++++++++++++
 18 files changed, 342 insertions(+), 192 deletions(-)
 create mode 100644 lib/PublicInbox/WQWorker.pm
 create mode 100644 xt/stress-sharedkv.t

^ 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 --
2021-02-01  8:28  7% [PATCH 00/21] lei2mail worker segfault finally fixed Eric Wong
2021-02-01  8:28  6% ` [PATCH 20/21] lei: avoid ETOOMANYREFS, cleanup imports 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).