how's memory usage on public-inbox-httpd?

how's memory usage on public-inbox-httpd?

[Eric Wong]

I haven't been around much, so not working on public-inbox means
fewer restarts :x

On my 2-core VM, I've been noticing public-inbox-httpd memory
spikes into the 500MB range, which is gross...  It seems caused
by slow clients and large threads/mbox downloads.  The PSGI code
only loads one email per-client in memory at-a-time when using
-httpd; but that adds up with many clients and larger messages.

I run two -httpd workers, one-per-core, but also varnish and an
experimental Ruby/C reverse-buffering proxy (yahns) for HTTPS.

The problem seems to be varnish isn't reading from -httpd fast
enough (and I lack CPU cores), but decreasing the niceness of
varnish seems to help with the problem...

Re: how's memory usage on public-inbox-httpd?

[Konstantin Ryabitsev]

Hello:

This is an old-ish discussion, but we finally had a chance to run the 
httpd daemon for a long time without restarting it to add more lists, 
and the memory usage on it is actually surprising:

$ ps -eF | grep public-inbox
publici+ 17741     1  0 52667 24836   8 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
publici+ 17744 17741  0 69739 90288   9 May24 ?        00:38:43 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
publici+ 18273     1  0 52599 23832   9 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log
publici+ 18275 18273  4 5016115 19713872 10 May24 ?    13:59:13 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log

You'll notice that process 18275 has been running since May 24 and takes 
up 19GB in RSS. This is a 16-core 64-GB system, so it's not necessarily 
super alarming, but seems large. :)

Is that normal, and if not, what can I do to help troubleshoot where 
it's all going?

-K


On Sat, Dec 01, 2018 at 07:44:29PM +0000, Eric Wong wrote:
>I haven't been around much, so not working on public-inbox means
>fewer restarts :x
>
>On my 2-core VM, I've been noticing public-inbox-httpd memory
>spikes into the 500MB range, which is gross...  It seems caused
>by slow clients and large threads/mbox downloads.  The PSGI code
>only loads one email per-client in memory at-a-time when using
>-httpd; but that adds up with many clients and larger messages.
>
>I run two -httpd workers, one-per-core, but also varnish and an
>experimental Ruby/C reverse-buffering proxy (yahns) for HTTPS.
>
>The problem seems to be varnish isn't reading from -httpd fast
>enough (and I lack CPU cores), but decreasing the niceness of
>varnish seems to help with the problem...

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> Hello:
> 
> This is an old-ish discussion, but we finally had a chance to run the httpd
> daemon for a long time without restarting it to add more lists, and the
> memory usage on it is actually surprising:

Thanks for getting back to this.

> $ ps -eF | grep public-inbox
> publici+ 17741     1  0 52667 24836   8 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
> publici+ 17744 17741  0 69739 90288   9 May24 ?        00:38:43 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
> publici+ 18273     1  0 52599 23832   9 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log
> publici+ 18275 18273  4 5016115 19713872 10 May24 ?    13:59:13 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log
> 
> You'll notice that process 18275 has been running since May 24 and takes up
> 19GB in RSS. This is a 16-core 64-GB system, so it's not necessarily super
> alarming, but seems large. :)

Yes, it's large and ugly :<  I don't even have 19GB and even
90MB RSS worries me.

Do you have commit 7d02b9e64455831d3bda20cd2e64e0c15dc07df5?
("view: stop storing all MIME objects on large threads")
That was most significant.

Also, it looks like you've yet to configure the wacky
coderepo+solver stuff, so that's not a culprit...

Otherwise it's probably a combination of several things...
httpd and nntpd both supports streaming, arbitrarily  large
endpoints (all.mbox.gz, and /T/, /t/, /t.mbox.gz threads with
thousands of messages, giant NNTP BODY/ARTICLE ranges).

All those endpoints should detect backpressure from a slow
client (varnish/nginx in your case) using the ->getline method.

gzip (for compressed mbox) also uses truckload of memory and I
would like to add options to control zlib window sizes to reduce
memory use (at the cost of less compression).  nginx has these
options, too, but they're not documented AFAIK.

For the case of varnish/nginx or whatever's in front of it not
keeping up...  the old design choice of Danga::Socket (now
inherited to PublicInbox::DS) made it buffer slow client data to
RAM, which doesn't make sense to me...  I prefer buffering to
the FS (similar to nginx/varnish) to avoid malloc fragmentation
and also to avoid delaying the extra kernel-to-user copy if
using sendfile.

By default, glibc malloc is really adverse to releasing memory
back to the OS, too.  It's fast in benchmarks that way; (until
the system starts swapping and slowdowns cascade to failure).

I'm also unsure about malloc fragmentation behavior at such
sizes and how it hurts locality.  So my preference is to avoid
putting big objects into heap and let the kernel/FS deal with
big buffers.

httpd/nntpd both try to avoid buffering at all with the
backpressure handling based on ->getline; but sometimes it's not
effective enough because some big chunks still end up in heap.

In any case, you can safely SIGQUIT the individual worker and
it'll restart gracefully w/o dropping active connections.

Also, are you only using the default of -W/--worker-process=1
on a 16-core machine?  Just checked public-inbox-httpd(8), the
-W switch is documented :)  You can use SIGTTIN/TTOU to
increase, decrease workers w/o restarting, too.

nntpd would have the same problem if people used it more;
but at the moment it doesn't do gzip.  I'm happy to see it's
at least gotten some traffic :)

> Is that normal, and if not, what can I do to help troubleshoot where it's
> all going?

There's definitely some problems with big threads, giant
messages and gzip overhead.  I was looking into a few big
threads earlier this year but forgot the Message-IDs :x

Do you have any stats on the number of simultaneous connections
public-inbox-httpd/nginx/varnish handles (and logging of that
info at peek)?  (perhaps running "ss -tan" periodically)(*)

Are you using the Plack::Middleware::Deflater endpoint in PSGI?
Removing it and doing gzip in varnish/nginx may be a little
faster since it can utilize multiple cores, but at higher IPC
cost.  I've gotten rid of the annoying warning for that
middleware install as a result...

But gzipped mboxes has the same problem; though; so adding
zlib window-size options would be necessary...

So I think supporting buffer-to-FS behavior in ::DS along
with gzip options should alleviate much of the memory use.  But
immediately you can increase worker process counts to distribute
the load between cores a bit...

I've also tried nicing down nginx/varnish so they're prioritized
by the kernel and don't bottleneck -httpd.  Makes sense to me in
theory but I was also making a lot of changes around the same
time to reduce httpd memory use.

Limiting HTTP endpoint response size isn't a real option to
protect the server; IMHO, because NNTP requires supporting giant
responses anyways.


(*) I did "raindrops" with Ruby+C back in the day but haven't really
    looked at it in ages, and I don't think the IPv6 counting was
    accurate <https://raindrops-demo.bogomips.org/>
    That's -httpd on :280

Re: how's memory usage on public-inbox-httpd?

[Konstantin Ryabitsev]

On Thu, Jun 06, 2019 at 08:37:52PM +0000, Eric Wong wrote:
>Do you have commit 7d02b9e64455831d3bda20cd2e64e0c15dc07df5?
>("view: stop storing all MIME objects on large threads")
>That was most significant.

Yes. We're running 743ac758 with a few cherry-picked patches on top of 
that (like epoch roll-over fix).

>Otherwise it's probably a combination of several things...
>httpd and nntpd both supports streaming, arbitrarily  large
>endpoints (all.mbox.gz, and /T/, /t/, /t.mbox.gz threads with
>thousands of messages, giant NNTP BODY/ARTICLE ranges).
>
>All those endpoints should detect backpressure from a slow
>client (varnish/nginx in your case) using the ->getline method.

Wouldn't that spike up and down? The size I'm seeing stays pretty 
constant without any significant changes across requests.

>Also, are you only using the default of -W/--worker-process=1
>on a 16-core machine?  Just checked public-inbox-httpd(8), the
>-W switch is documented :)  You can use SIGTTIN/TTOU to
>increase, decrease workers w/o restarting, too.

D'oh, yes... though it's not been a problem yet. :) I'm not sure I want 
to bump that up, though, if that means we're going to have multiple 
19GB-sized processes instead of one. :)

>Do you have any stats on the number of simultaneous connections
>public-inbox-httpd/nginx/varnish handles (and logging of that
>info at peek)?  (perhaps running "ss -tan" periodically)(*)

We don't collect that info, but I'm not sure it's the number of 
concurrent connections that's the culprit, as there is no fluctuation in 
RSS size based on the number of responses.

To answer the questions in your follow-up:

It would appear to be all in anon memory. Mem_usage [1] reports:

# ./Mem_usage 18275
Backed by file:
  Executable                r-x  16668
  Write/Exec (jump tables)  rwx  0
  RO data                   r--  106908
  Data                      rw-  232
  Unreadable                ---  94072
  Unknown                        0
Anonymous:
  Writable code (stack)     rwx  0
  Data (malloc, mmap)       rw-  19988892
  RO data                   r--  0
  Unreadable                ---  0
  Unknown                        12

I've been looking at lsof -p of that process and I see sqlite and xapian 
showing up and disappearing. The lkml ones are being accessed almost all 
the time, but even there I see them showing up with different FD 
entries, so they are being closed and reopened properly.

Hope this helps.

-K

.. [1] https://elinux.org/images/d/d3/Mem_usage

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> On Thu, Jun 06, 2019 at 08:37:52PM +0000, Eric Wong wrote:
> > Do you have commit 7d02b9e64455831d3bda20cd2e64e0c15dc07df5?
> > ("view: stop storing all MIME objects on large threads")
> > That was most significant.
> 
> Yes. We're running 743ac758 with a few cherry-picked patches on top of that
> (like epoch roll-over fix).
> 
> > Otherwise it's probably a combination of several things...
> > httpd and nntpd both supports streaming, arbitrarily  large
> > endpoints (all.mbox.gz, and /T/, /t/, /t.mbox.gz threads with
> > thousands of messages, giant NNTP BODY/ARTICLE ranges).
> > 
> > All those endpoints should detect backpressure from a slow
> > client (varnish/nginx in your case) using the ->getline method.
> 
> Wouldn't that spike up and down? The size I'm seeing stays pretty constant
> without any significant changes across requests.

Nope.  That's the thing with glibc malloc not wanting to trim
the heap for good benchmarks.

You could also try starting with MALLOC_MMAP_THRESHOLD_=131072
in env (or some smaller/larger number in bytes) to force it to
use mmap in more cases instead of sbrk.

> > Also, are you only using the default of -W/--worker-process=1
> > on a 16-core machine?  Just checked public-inbox-httpd(8), the
> > -W switch is documented :)  You can use SIGTTIN/TTOU to
> > increase, decrease workers w/o restarting, too.
> 
> D'oh, yes... though it's not been a problem yet. :) I'm not sure I want to
> bump that up, though, if that means we're going to have multiple 19GB-sized
> processes instead of one. :)

You'd probably end up with several smaller processes totalling
up to 19GB.

In any case, killing individual workers with QUIT/INT/TERM is
graceful and won't drop connections if memory use on one goes
awry.

> > Do you have any stats on the number of simultaneous connections
> > public-inbox-httpd/nginx/varnish handles (and logging of that
> > info at peek)?  (perhaps running "ss -tan" periodically)(*)
> 
> We don't collect that info, but I'm not sure it's the number of concurrent
> connections that's the culprit, as there is no fluctuation in RSS size based
> on the number of responses.

Without concurrent connections; I can't see that happening
unless there's a single message which is gigabytes in size.  I'm
already irked that Email::MIME requires slurping entire emails
into memory; but it should not be using more than one
Email::MIME object in memory-at-a-time for a single client.

Anything from varnish/nginx logs can't keep up for some reason?

Come to think of it, nginx proxy buffering might be redundant
and even harmful if varnish is already doing it.

Perhaps "proxy_buffering off" in nginx is worth trying...
I use yahns instead of nginx, which does lazy buffering (but
scary Ruby experimental server warning applies :x).

Last I checked: nginx is either buffer-in-full-before-first-byte
or no buffering at all (which is probably fine with varnish).

> To answer the questions in your follow-up:
> 
> It would appear to be all in anon memory. Mem_usage [1] reports:
> 
> # ./Mem_usage 18275
> Backed by file:
>  Executable                r-x  16668
>  Write/Exec (jump tables)  rwx  0
>  RO data                   r--  106908
>  Data                      rw-  232
>  Unreadable                ---  94072
>  Unknown                        0
> Anonymous:
>  Writable code (stack)     rwx  0
>  Data (malloc, mmap)       rw-  19988892
>  RO data                   r--  0
>  Unreadable                ---  0
>  Unknown                        12
> 
> I've been looking at lsof -p of that process and I see sqlite and xapian
> showing up and disappearing. The lkml ones are being accessed almost all the
> time, but even there I see them showing up with different FD entries, so
> they are being closed and reopened properly.

Yep, that's expected.  It's to better detect DB changes in case
of compact/copydatabase/xcpdb for Xapian.

Might not be necessary strictly necessary for SQLite, but maybe
somebody could be running VACUUM offline; then flock-ing
inbox.lock and rename-ing it into place or something (and
retrying/restarting the VACUUM if out-of-date, seq_lock style).

Re: how's memory usage on public-inbox-httpd?

[Konstantin Ryabitsev]

On Thu, Jun 06, 2019 at 10:10:09PM +0000, Eric Wong wrote:
>> > All those endpoints should detect backpressure from a slow
>> > client (varnish/nginx in your case) using the ->getline method.
>>
>> Wouldn't that spike up and down? The size I'm seeing stays pretty constant
>> without any significant changes across requests.
>
>Nope.  That's the thing with glibc malloc not wanting to trim
>the heap for good benchmarks.
>
>You could also try starting with MALLOC_MMAP_THRESHOLD_=131072
>in env (or some smaller/larger number in bytes) to force it to
>use mmap in more cases instead of sbrk.

I've restarted the process and I'm running mmap -x $PID | tail -1 on it 
once a minute. I'll try to collect this data for a while and see if I 
can notice significant increases and correlate that with access logs.  
From the first few minutes of running I see:

Thu Jun  6 22:06:03 UTC 2019
total kB          298160  102744   96836
Thu Jun  6 22:07:03 UTC 2019
total kB          355884  154968  147664
Thu Jun  6 22:08:03 UTC 2019
total kB          355884  154980  147664
Thu Jun  6 22:09:03 UTC 2019
total kB          359976  156788  148336
Thu Jun  6 22:10:03 UTC 2019
total kB          359976  156788  148336
Thu Jun  6 22:11:03 UTC 2019
total kB          359976  156788  148336
Thu Jun  6 22:12:03 UTC 2019
total kB          365464  166612  158160
Thu Jun  6 22:13:03 UTC 2019
total kB          366884  167908  159456
Thu Jun  6 22:14:03 UTC 2019
total kB          366884  167908  159456
Thu Jun  6 22:15:03 UTC 2019
total kB          366884  167908  159456

>Without concurrent connections; I can't see that happening
>unless there's a single message which is gigabytes in size.  I'm
>already irked that Email::MIME requires slurping entire emails
>into memory; but it should not be using more than one
>Email::MIME object in memory-at-a-time for a single client.
>
>Anything from varnish/nginx logs can't keep up for some reason?

Speaking of logs, I did notice that even though we're passing -1 
/var/log/public-inbox/httpd.out.log, that file stays empty. There's 
nttpd.out.log there, which is non-empty, so that's curious:

# ls -ahl
total 2.6M
drwx------.  2 publicinbox publicinbox  177 Jun  6 22:05 .
drwxr-xr-x. 21 root        root        4.0K Jun  2 03:12 ..
-rw-r--r--.  1 publicinbox publicinbox    0 Jun  6 22:05 httpd.out.log
-rw-r--r--.  1 publicinbox publicinbox 422K Jun  6 22:04 nntpd.out.log
-rw-r--r--.  1 publicinbox publicinbox 771K May 12 01:02 nntpd.out.log-20190512.gz
-rw-r--r--.  1 publicinbox publicinbox 271K May 19 03:45 nntpd.out.log-20190519.gz
-rw-r--r--.  1 publicinbox publicinbox  86K May 25 22:23 nntpd.out.log-20190526.gz
-rw-r--r--.  1 publicinbox publicinbox 1.1M Jun  2 00:52 nntpd.out.log-20190602

Could it be that stdout is not being written out and is just perpetually 
buffered? That could explain the ever-growing size.

-K

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> On Thu, Jun 06, 2019 at 10:10:09PM +0000, Eric Wong wrote:
> > > > All those endpoints should detect backpressure from a slow
> > > > client (varnish/nginx in your case) using the ->getline method.
> > > 
> > > Wouldn't that spike up and down? The size I'm seeing stays pretty constant
> > > without any significant changes across requests.
> > 
> > Nope.  That's the thing with glibc malloc not wanting to trim
> > the heap for good benchmarks.
> > 
> > You could also try starting with MALLOC_MMAP_THRESHOLD_=131072
> > in env (or some smaller/larger number in bytes) to force it to
> > use mmap in more cases instead of sbrk.
> 
> I've restarted the process and I'm running mmap -x $PID | tail -1 on it once
> a minute. I'll try to collect this data for a while and see if I can notice
> significant increases and correlate that with access logs.  From the first
> few minutes of running I see:
> 
> Thu Jun  6 22:06:03 UTC 2019
> total kB          298160  102744   96836
> Thu Jun  6 22:07:03 UTC 2019
> total kB          355884  154968  147664
> Thu Jun  6 22:08:03 UTC 2019
> total kB          355884  154980  147664
> Thu Jun  6 22:09:03 UTC 2019
> total kB          359976  156788  148336
> Thu Jun  6 22:10:03 UTC 2019
> total kB          359976  156788  148336
> Thu Jun  6 22:11:03 UTC 2019
> total kB          359976  156788  148336
> Thu Jun  6 22:12:03 UTC 2019
> total kB          365464  166612  158160
> Thu Jun  6 22:13:03 UTC 2019
> total kB          366884  167908  159456
> Thu Jun  6 22:14:03 UTC 2019
> total kB          366884  167908  159456
> Thu Jun  6 22:15:03 UTC 2019
> total kB          366884  167908  159456

Would also be good to correlate that to open sockets, too.
(168M is probably normal for 64-bit, I'm still on 32-bit and its <100M).
I'm not happy with that memory use, even; but it's better than
gigabytes.

> > Without concurrent connections; I can't see that happening
> > unless there's a single message which is gigabytes in size.  I'm
> > already irked that Email::MIME requires slurping entire emails
> > into memory; but it should not be using more than one
> > Email::MIME object in memory-at-a-time for a single client.
> > 
> > Anything from varnish/nginx logs can't keep up for some reason?
> 
> Speaking of logs, I did notice that even though we're passing -1
> /var/log/public-inbox/httpd.out.log, that file stays empty. There's
> nttpd.out.log there, which is non-empty, so that's curious:
> 
> # ls -ahl
> total 2.6M
> drwx------.  2 publicinbox publicinbox  177 Jun  6 22:05 .
> drwxr-xr-x. 21 root        root        4.0K Jun  2 03:12 ..
> -rw-r--r--.  1 publicinbox publicinbox    0 Jun  6 22:05 httpd.out.log
> -rw-r--r--.  1 publicinbox publicinbox 422K Jun  6 22:04 nntpd.out.log
> -rw-r--r--.  1 publicinbox publicinbox 771K May 12 01:02 nntpd.out.log-20190512.gz
> -rw-r--r--.  1 publicinbox publicinbox 271K May 19 03:45 nntpd.out.log-20190519.gz
> -rw-r--r--.  1 publicinbox publicinbox  86K May 25 22:23 nntpd.out.log-20190526.gz
> -rw-r--r--.  1 publicinbox publicinbox 1.1M Jun  2 00:52 nntpd.out.log-20190602
> 
> Could it be that stdout is not being written out and is just perpetually
> buffered? That could explain the ever-growing size.

There's no HTTP access logging by default.
AccessLog::Timed is commented out in examples/public-inbox.psgi;
and the example uses syswrite, even.

Also, PublicInbox::Daemon definitely enables autoflush on STDOUT.

[RFC] optionally support glibc malloc_info via SIGCONT

[Eric Wong]

Eric Wong <e@80x24.org> wrote:
> Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> > On Thu, Jun 06, 2019 at 10:10:09PM +0000, Eric Wong wrote:
> > > > > All those endpoints should detect backpressure from a slow
> > > > > client (varnish/nginx in your case) using the ->getline method.
> > > > 
> > > > Wouldn't that spike up and down? The size I'm seeing stays pretty constant
> > > > without any significant changes across requests.
> > > 
> > > Nope.  That's the thing with glibc malloc not wanting to trim
> > > the heap for good benchmarks.

glibc's malloc_info(3) function shows a bunch of info about
malloc internal structures which can't be examined easily
with other tools.  Maybe this can help track down what glibc
malloc is doing and can tell us how much free memory glibc
is holding onto w/o giving back to the kernel.

------------8<-------------
Subject: [RFC] optionally support glibc malloc_info via SIGCONT

If run with PERL_INLINE_DIRECTORY for Inline::C support
along with INBOX_DEBUG=malloc_info, we can allow users
to opt-in to compiling extra code to support the glibc
malloc_info(3) function.

We'll also add SIGCONT handler to dump the malloc_info(3)
output to stderr on our daemons.
---
 MANIFEST                 |  1 +
 lib/PublicInbox/Spawn.pm | 17 +++++++++++++++++
 t/malloc_info.t          | 25 +++++++++++++++++++++++++
 3 files changed, 43 insertions(+)
 create mode 100644 t/malloc_info.t

diff --git a/MANIFEST b/MANIFEST
index 5085bff..4a7f7ef 100644
--- a/MANIFEST
+++ b/MANIFEST
@@ -210,6 +210,7 @@ t/indexlevels-mirror.t
 t/init.t
 t/linkify.t
 t/main-bin/spamc
+t/malloc_info.t
 t/mda.t
 t/mda_filter_rubylang.t
 t/mid.t
diff --git a/lib/PublicInbox/Spawn.pm b/lib/PublicInbox/Spawn.pm
index 66b916d..9210f11 100644
--- a/lib/PublicInbox/Spawn.pm
+++ b/lib/PublicInbox/Spawn.pm
@@ -149,6 +149,23 @@ int pi_fork_exec(int in, int out, int err,
 }
 VFORK_SPAWN
 
+# TODO: we may support other mallocs through this parameter
+if (($ENV{INBOX_DEBUG} // '') =~ /\bmalloc_info\b/) {
+	$vfork_spawn .= <<MALLOC_DEBUG;
+#include <malloc.h>
+
+int inbox_malloc_info(int options)
+{
+	int rc = malloc_info(options, stderr);
+
+	return rc == 0 ? TRUE : FALSE;
+}
+MALLOC_DEBUG
+
+	# dump malloc info to stderr on SIGCONT
+	$SIG{CONT} = sub { inbox_malloc_info(0) };
+}
+
 my $inline_dir = $ENV{PERL_INLINE_DIRECTORY};
 $vfork_spawn = undef unless defined $inline_dir && -d $inline_dir && -w _;
 if (defined $vfork_spawn) {
diff --git a/t/malloc_info.t b/t/malloc_info.t
new file mode 100644
index 0000000..352ec5c
--- /dev/null
+++ b/t/malloc_info.t
@@ -0,0 +1,25 @@
+# Copyright (C) 2019 all contributors <meta@public-inbox.org>
+# License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
+use strict;
+use warnings;
+use Test::More;
+use PublicInbox::Spawn ();
+
+if (!PublicInbox::Spawn->can('inbox_malloc_info')) {
+	plan skip_all => 'inbox_malloc_info not enabled';
+}
+
+open my $olderr, '>&', \*STDERR or die "dup stderr: $!";
+open my $tmp, '+>', undef or die "tmpfile: $!";
+open STDERR, '>&', $tmp or die "redirect stderr to \$tmp: $!";
+my @x = map { '0' x (1024 * 1024) } (1..128);
+my $cb = $SIG{CONT};
+$cb->();
+@x = ('hello');
+PublicInbox::Spawn::inbox_malloc_info(0);
+open STDERR, '>&', $olderr or die "restore stderr: $!";
+sysseek($tmp, 0, 0) == 0 or die "sysseek: $!";
+my @info = <$tmp>;
+like($info[0], qr/</, 'output looks like XML');
+
+done_testing;
-- 
EW

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Eric Wong <e@80x24.org> wrote:
> Without concurrent connections; I can't see that happening
> unless there's a single message which is gigabytes in size.  I'm
> already irked that Email::MIME requires slurping entire emails
> into memory; but it should not be using more than one
> Email::MIME object in memory-at-a-time for a single client.

Giant multipart messages do a lot of damage.  Maybe concurrent
clients hitting the same endpoints will do more damage.

Largest I see in LKML is 7204747 bytes (which is frightening).
That bloats to 21626795 bytes when parsed by Email::MIME.

I thought it was bad enough that all Perl mail modules seem to
require slurping 7M into memory...

-------8<--------
use strict; use warnings;
require Email::MIME;
use bytes ();
use Devel::Size qw(total_size);
my $in = do { local $/; <STDIN> };
print 'string: ', total_size($in), ' actual: ', bytes::length($in), "\n";
print 'MIME: ', total_size(Email::MIME->new(\$in)), "\n";
-------8<--------

That shows (on amd64):

	string: 7204819 actual: 7204747
	MIME: 21626795

Maybe you have bigger messages outside of LKML.
This prints all objects >1MB in a git dir:

	git cat-file --buffer --batch-check --batch-all-objects \
		--unordered | awk '$3 > 1048576 { print }'

And I also remember you're supporting non-vger lists where HTML
mail is allowed, so that can't be good for memory use at all :<

Streaming MIME handling has been on the TODO for a while,
at least...

|* streaming Email::MIME replacement: currently we generate many
|  allocations/strings for headers we never look at and slurp
|  entire message bodies into memory.
|  (this is pie-in-the-sky territory...)

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Eric Wong <e@80x24.org> wrote:
> Maybe you have bigger messages outside of LKML.
> This prints all objects >1MB in a git dir:
> 
> 	git cat-file --buffer --batch-check --batch-all-objects \
> 		--unordered | awk '$3 > 1048576 { print }'
> 
> And I also remember you're supporting non-vger lists where HTML
> mail is allowed, so that can't be good for memory use at all :<

Btw, have you had a chance to run the above to scan your repos?

The 19GB RSS would interact poorly with more frequent fork+exec;
such as the change to do unconditional expiry of long-lived
cat-file --batch* processes:

	https://public-inbox.org/meta/20190601033706.18113-2-e@80x24.org/


If you run Inline::C + PERL_INLINE_DIRECTORY to use vfork
instead of fork, then RSS won't affect process spawning
speeds.

But yeah, 19GB RSS still sucks and I'd like to get to the bottom
of it.

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> Is that normal, and if not, what can I do to help troubleshoot where it's
> all going?

Oh, another thing is to check if any of that is from mmap-ed
files or if its all anon heap memory.  The SQLite I'm using
from Debian doesn't seem to mmap files (I've never had to look
too hard at tuning SQLite) and I don't think libxapian uses mmap
at all either (just pread).  But LKML's sqlite files are huge.

I know git(1) will mmap entire packs into memory.  If/when
public-inbox start supporting Git::Raw (libgit2) or even
Cache::FastMmap, then yeah; process sizes will look a lot
bigger.

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> Hello:
> 
> This is an old-ish discussion, but we finally had a chance to run the httpd
> daemon for a long time without restarting it to add more lists, and the
> memory usage on it is actually surprising:
> 
> $ ps -eF | grep public-inbox
> publici+ 17741     1  0 52667 24836   8 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
> publici+ 17744 17741  0 69739 90288   9 May24 ?        00:38:43 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
> publici+ 18273     1  0 52599 23832   9 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log
> publici+ 18275 18273  4 5016115 19713872 10 May24 ?    13:59:13 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log
> 
> You'll notice that process 18275 has been running since May 24 and takes up
> 19GB in RSS. This is a 16-core 64-GB system, so it's not necessarily super
> alarming, but seems large. :)
> 
> Is that normal, and if not, what can I do to help troubleshoot where it's
> all going?

Btw, has this gotten better since the Perl 5.16.3 workarounds?

My 32-bit instance which sees the most HTTP traffic hasn't
exceeded 80M per-process in a while.

Re: how's memory usage on public-inbox-httpd?

[Konstantin Ryabitsev]

On Wed, Oct 16, 2019 at 10:10:45PM +0000, Eric Wong wrote:
>> This is an old-ish discussion, but we finally had a chance to run the 
>> httpd
>> daemon for a long time without restarting it to add more lists, and the
>> memory usage on it is actually surprising:
>>
>> $ ps -eF | grep public-inbox
>> publici+ 17741     1  0 52667 24836   8 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
>> publici+ 17744 17741  0 69739 90288   9 May24 ?        00:38:43 /usr/bin/perl -w /usr/local/bin/public-inbox-nntpd -1 /var/log/public-inbox/nntpd.out.log
>> publici+ 18273     1  0 52599 23832   9 May24 ?        00:00:00 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log
>> publici+ 18275 18273  4 5016115 19713872 10 May24 ?    13:59:13 /usr/bin/perl -w /usr/local/bin/public-inbox-httpd -1 /var/log/public-inbox/httpd.out.log
>>
>> You'll notice that process 18275 has been running since May 24 and takes up
>> 19GB in RSS. This is a 16-core 64-GB system, so it's not necessarily super
>> alarming, but seems large. :)
>>
>> Is that normal, and if not, what can I do to help troubleshoot where it's
>> all going?
>
>Btw, has this gotten better since the Perl 5.16.3 workarounds?
>
>My 32-bit instance which sees the most HTTP traffic hasn't
>exceeded 80M per-process in a while.

It's been definitely dramatically better. We keep adding lists to lore, 
so I haven't really been able to watch memory usage after a long period 
of daemon uptime, but it's never really gone very much above 1GB. In 
fact, we're downgrading lore to a smaller instance in the near future 
since we don't need to worry about running out of RAM any more.

Best,
-K

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> On Wed, Oct 16, 2019 at 10:10:45PM +0000, Eric Wong wrote:
> > Btw, has this gotten better since the Perl 5.16.3 workarounds?
> > 
> > My 32-bit instance which sees the most HTTP traffic hasn't
> > exceeded 80M per-process in a while.
> 
> It's been definitely dramatically better. We keep adding lists to lore, so I
> haven't really been able to watch memory usage after a long period of daemon
> uptime, but it's never really gone very much above 1GB. In fact, we're
> downgrading lore to a smaller instance in the near future since we don't
> need to worry about running out of RAM any more.

Cool, but 1GB is still an order of magnitude worse that what
I'd expect :<   I remember Email::MIME had huge explosions with
some 30MB+ spam messages:
  https://public-inbox.org/meta/20190609083918.gfr2kurah7f2hysx@dcvr/
  (maybe gmime can help)

Depending on your storage speed/latency, more RAM can still help
significantly with Xapian.  The NVME stuff has amazing numbers,
but my mobos are too old and I'm still stuck on SATA 2.

Is nntpd better?  That only uses Email::Simple and not MIME;
so less explosions.

Re: how's memory usage on public-inbox-httpd?

[Konstantin Ryabitsev]

On Sat, Oct 19, 2019 at 12:11:44AM +0000, Eric Wong wrote:
>> It's been definitely dramatically better. We keep adding lists to 
>> lore, so I
>> haven't really been able to watch memory usage after a long period of daemon
>> uptime, but it's never really gone very much above 1GB. In fact, we're
>> downgrading lore to a smaller instance in the near future since we don't
>> need to worry about running out of RAM any more.
>
>Cool, but 1GB is still an order of magnitude worse that what
>I'd expect :<   I remember Email::MIME had huge explosions with
>some 30MB+ spam messages:
>  https://public-inbox.org/meta/20190609083918.gfr2kurah7f2hysx@dcvr/
>  (maybe gmime can help)
>
>Depending on your storage speed/latency, more RAM can still help
>significantly with Xapian.  The NVME stuff has amazing numbers,
>but my mobos are too old and I'm still stuck on SATA 2.

My goal is to rework lore.kernel.org significantly. Currently, it's a 
single system hosted at AWS that both receives mail and serves the 
archives, but I would actually like to split it into two:

- the archiver that just generates git repositories but serves no 
  traffic (probably running directly on mail.kernel.org).
- several front-ends that replicate repositories from the archiver and 
  provide http/nntp access, probably reusing mirrors.edge.kernel.org 
  nodes that run from us-west, us-east, eu-west and ap-east.

That should provide both redundancy and better geographic availability 
of the service. This requires some testing first to ensure that 
grokmirror hooks and reindexing works reliably for replicated repo 
collections.

>Is nntpd better?  That only uses Email::Simple and not MIME;
>so less explosions.

The number of people using nntp is several orders of magnitude lower, so 
I'm not sure it's a good metric for anything.

Best,
-K

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Konstantin Ryabitsev <konstantin@linuxfoundation.org> wrote:
> On Sat, Oct 19, 2019 at 12:11:44AM +0000, Eric Wong wrote:
> >> It's been definitely dramatically better. We keep adding lists to 
> >> lore, so I
> >> haven't really been able to watch memory usage after a long period of daemon
> >> uptime, but it's never really gone very much above 1GB. In fact, we're
> >> downgrading lore to a smaller instance in the near future since we don't
> >> need to worry about running out of RAM any more.
> >
> >Cool, but 1GB is still an order of magnitude worse that what
> >I'd expect :<   I remember Email::MIME had huge explosions with
> >some 30MB+ spam messages:
> >  https://public-inbox.org/meta/20190609083918.gfr2kurah7f2hysx@dcvr/
> >  (maybe gmime can help)
> >
> >Depending on your storage speed/latency, more RAM can still help
> >significantly with Xapian.  The NVME stuff has amazing numbers,
> >but my mobos are too old and I'm still stuck on SATA 2.
> 
> My goal is to rework lore.kernel.org significantly. Currently, it's a 
> single system hosted at AWS that both receives mail and serves the 
> archives, but I would actually like to split it into two:
> 
> - the archiver that just generates git repositories but serves no 
>   traffic (probably running directly on mail.kernel.org).
> - several front-ends that replicate repositories from the archiver and 
>   provide http/nntp access, probably reusing mirrors.edge.kernel.org 
>   nodes that run from us-west, us-east, eu-west and ap-east.
> 
> That should provide both redundancy and better geographic availability 
> of the service. This requires some testing first to ensure that 
> grokmirror hooks and reindexing works reliably for replicated repo 
> collections.

Yeah, I noticed some mirror indexing bugs at:
https://public-inbox.org/meta/20191016211415.GA6084@dcvr/

But patches 4/3 and 5/3 seem to be doing everything right and
I expect the series to be merged soon.

I'm also planning to dogfood a mirror of lore off some of my
.onions, soon.

> >Is nntpd better?  That only uses Email::Simple and not MIME;
> >so less explosions.
> 
> The number of people using nntp is several orders of magnitude lower, so 
> I'm not sure it's a good metric for anything.

Hopefully nntp usage goes up over time.

Re: how's memory usage on public-inbox-httpd?

[Eric Wong]

Eric Wong <e@80x24.org> wrote:
> Cool, but 1GB is still an order of magnitude worse that what
> I'd expect :<   I remember Email::MIME had huge explosions with
> some 30MB+ spam messages:
>   https://public-inbox.org/meta/20190609083918.gfr2kurah7f2hysx@dcvr/
>   (maybe gmime can help)

Fwiw, I'm working on porting a scripting-language-aware malloc
tracker I wrote for another scripting language over to Perl/XS
which can track down which line of Perl called a particular
malloc() statement.

...And reading perlguts/perlxstut manpages again :x