bug#37207: guix.gnu.org returns Last-Modified = Epoch

bug#37207: guix.gnu.org returns Last-Modified = Epoch

[Ludovic Courtès]

Hello Guix,

Since the use of the ‘static-web-site’ service, which puts web site
files in the store, nginx returns a ‘Last-Modified’ header that can
trick clients into caching things forever:

--8<---------------cut here---------------start------------->8---
$ wget --debug -O /dev/null   https://guix.gnu.org/packages.json 2>&1 | grep Last
Last-Modified: Thu, 01 Jan 1970 00:00:01 GMT
--8<---------------cut here---------------end--------------->8---

We should tell nginx to do not emit ‘Last-Modified’, or to take the
state from the /srv/guix.gnu.org symlink, if possible.

Ludo’.

bug#37207: guix.gnu.org Last Modified at epoch

[Gábor Boskovits]

[-- Attachment #1: Type: text/plain, Size: 262 bytes --]

Hello,

Supressing the last modified header is just an
add_header Last-Modified "";
away.

To get the info from the symlink seems to be much trickier, i would do with
either embedded perl or embedded lua. I am not sure if we should bother
with it, though. Wdyt?

[-- Attachment #2: Type: text/html, Size: 431 bytes --]

bug#37207: guix.gnu.org Last Modified at epoch

[Tobias Geerinckx-Rice via Bug reports for GNU Guix]

[-- Attachment #1: Type: text/plain, Size: 767 bytes --]

Gábor, Ludo',

Gábor Boskovits 写道：
> Supressing the last modified header is just an
> add_header Last-Modified "";
> away.

You'll also need:

# Don't honour client If-Modified-Since constraints.
if_modified_since off;
# Nginx's etags are hashes of file timestamp & file length.
etag off;

Turning these off will of course prevent all caching.  I don't 
know if that would add measurable load to guix.gnu.org (it would 
be more problematic if we used a CDN, but it might still make a 
difference).

Nix does something both interesting and icky — as always: patch[0] 
nginx to look up the realpath() instead, so clients can still 
cache using If-None-Match.

Kind regards,

T G-R

[0]: https://github.com/NixOS/nixpkgs/pull/48337

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 227 bytes --]

bug#37207: guix.gnu.org Last Modified at epoch

[Gábor Boskovits]

[-- Attachment #1: Type: text/plain, Size: 1556 bytes --]

Hello Tobias,

Tobias Geerinckx-Rice via Bug reports for GNU Guix <bug-guix@gnu.org> ezt
írta (időpont: 2019. aug. 28., Sze, 16:38):

> Gábor, Ludo',
>
> Gábor Boskovits 写道：
> > Supressing the last modified header is just an
> > add_header Last-Modified "";
> > away.
>
> You'll also need:
>
> # Don't honour client If-Modified-Since constraints.
> if_modified_since off;
> # Nginx's etags are hashes of file timestamp & file length.
> etag off;
>
>
You really have a point here.

Based on my reseach, I came up with the following:

we need
etag off;

we should create a file with the git last modification time of the files,
updated when there is a new commit in the repo => last-modified
we should create a file with some hash of the files, updated when there is
a new commit in the repo => etag
we could restrict these operations to the files modified since the last
checkout.

Retrieve these with embededd perl.
Wdyt?


> Turning these off will of course prevent all caching.  I don't
> know if that would add measurable load to guix.gnu.org (it would
> be more problematic if we used a CDN, but it might still make a
> difference).
>
> Nix does something both interesting and icky — as always: patch[0]
> nginx to look up the realpath() instead, so clients can still
> cache using If-None-Match.
>
> Kind regards,
>
> T G-R
>
> [0]: https://github.com/NixOS/nixpkgs/pull/48337
>

Best regards,
g_bor

-- 
OpenPGP Key Fingerprint: 7988:3B9F:7D6A:4DBF:3719:0367:2506:A96C:CF63:0B21

[-- Attachment #2: Type: text/html, Size: 2568 bytes --]

bug#37207: guix.gnu.org Last Modified at epoch

[Ludovic Courtès]

Hello,

Gábor Boskovits <boskovits@gmail.com> skribis:

> we should create a file with the git last modification time of the files,
> updated when there is a new commit in the repo => last-modified
> we should create a file with some hash of the files, updated when there is
> a new commit in the repo => etag
> we could restrict these operations to the files modified since the last
> checkout.
>
> Retrieve these with embededd perl.
> Wdyt?

What would the config look like?  AFAICS our ‘nginx’ package doesn’t
embed Perl, and I think it’s better this way.  :-)  Can we do that with
pure nginx directives?

We create /srv/guix.gnu.org (as a symlink) with the correct mtime¹.  If
we can tell nginx to use it as the ‘Last-Modified’ date, that’s perfect.

Ludo’.

¹ https://git.savannah.gnu.org/cgit/guix/maintenance.git/tree/hydra/berlin.scm#n212

bug#37207: guix.gnu.org Last Modified at epoch

[Gábor Boskovits]

[-- Attachment #1: Type: text/plain, Size: 1619 bytes --]

Hello Ludo,

Ludovic Courtès <ludo@gnu.org> ezt írta (időpont: 2019. aug. 28., Sze,
22:32):

> Hello,
>
> Gábor Boskovits <boskovits@gmail.com> skribis:
>
> > we should create a file with the git last modification time of the files,
> > updated when there is a new commit in the repo => last-modified
> > we should create a file with some hash of the files, updated when there
> is
> > a new commit in the repo => etag
> > we could restrict these operations to the files modified since the last
> > checkout.
> >
> > Retrieve these with embededd perl.
> > Wdyt?
>
> What would the config look like?  AFAICS our ‘nginx’ package doesn’t
> embed Perl, and I think it’s better this way.  :-)  Can we do that with
> pure nginx directives?
>
> We create /srv/guix.gnu.org (as a symlink) with the correct mtime¹.  If
> we can tell nginx to use it as the ‘Last-Modified’ date, that’s perfect.
>
>
I was thinking about this. Yes, we can solve that with pure nginx. There is
an issue however.
It invalidates all cached entries on update, so files not modified will
also need to be downloaded again.

The easiest way to do that would be to simply generate an nginx config
snippet at a configurable location,
setting up the mtime and etags variable, and include that from the main
config.

If this would be ok, then I will have a look at implementing this.

Ludo’.
>
> ¹
> https://git.savannah.gnu.org/cgit/guix/maintenance.git/tree/hydra/berlin.scm#n212
>

Best regards,
g_bor

-- 
OpenPGP Key Fingerprint: 7988:3B9F:7D6A:4DBF:3719:0367:2506:A96C:CF63:0B21

[-- Attachment #2: Type: text/html, Size: 2623 bytes --]

bug#37207: guix.gnu.org Last Modified at epoch

[Ludovic Courtès]

Hi Gábor,

Gábor Boskovits <boskovits@gmail.com> skribis:

> Ludovic Courtès <ludo@gnu.org> ezt írta (időpont: 2019. aug. 28., Sze,
> 22:32):
>
>> Hello,
>>
>> Gábor Boskovits <boskovits@gmail.com> skribis:
>>
>> > we should create a file with the git last modification time of the files,
>> > updated when there is a new commit in the repo => last-modified
>> > we should create a file with some hash of the files, updated when there
>> is
>> > a new commit in the repo => etag
>> > we could restrict these operations to the files modified since the last
>> > checkout.
>> >
>> > Retrieve these with embededd perl.
>> > Wdyt?
>>
>> What would the config look like?  AFAICS our ‘nginx’ package doesn’t
>> embed Perl, and I think it’s better this way.  :-)  Can we do that with
>> pure nginx directives?
>>
>> We create /srv/guix.gnu.org (as a symlink) with the correct mtime¹.  If
>> we can tell nginx to use it as the ‘Last-Modified’ date, that’s perfect.
>>
>>
> I was thinking about this. Yes, we can solve that with pure nginx. There is
> an issue however.
> It invalidates all cached entries on update, so files not modified will
> also need to be downloaded again.
>
> The easiest way to do that would be to simply generate an nginx config
> snippet at a configurable location,
> setting up the mtime and etags variable, and include that from the main
> config.
>
> If this would be ok, then I will have a look at implementing this.

I’m not sure I fully understand, but yes, if you could send a prototype
as a diff against maintenance.git, that’d be great!

Thank you,
Ludo’.

bug#37207: guix.gnu.org Last Modified at epoch

[Ludovic Courtès]

Hello!

Did one of you have chance to come up with a trick to emit the right
‘Last-Modified’?  We seemed to be close to having something.  :-)

Thanks,
Ludo’.

bug#37207: guix.gnu.org Last Modified at epoch

[Ludovic Courtès]

Hi Tobias,

Tobias Geerinckx-Rice <me@tobias.gr> skribis:

> Turning these off will of course prevent all caching.  I don't know if
> that would add measurable load to guix.gnu.org (it would be more
> problematic if we used a CDN, but it might still make a difference).
>
> Nix does something both interesting and icky — as always: patch[0]
> nginx to look up the realpath() instead, so clients can still cache
> using If-None-Match.

> [0]: https://github.com/NixOS/nixpkgs/pull/48337

(See
<https://raw.githubusercontent.com/NixOS/nixpkgs/9bc23f31d29138f09db6af52708a9b8b64deec64/pkgs/servers/http/nginx/nix-etag-1.15.4.patch>.)

I had overlooked this patch but it looks like the right approach
overall.  Calling ‘realpath’ each time seems a bit expensive as it
creates an ‘lstat’ storm, but I can’t think of a better solution.

I also found this post whose main interest is in showing how to write a
plugin to generate custom etags:

  https://mikewest.org/2008/11/generating-etags-for-static-content-using-nginx/

Thoughts?

Ludo’.

bug#37207: guix.gnu.org Last Modified at epoch

[Danny Milosavljevic]

[-- Attachment #1: Type: text/plain, Size: 601 bytes --]

Hi Gabor,

On Wed, 28 Aug 2019 12:40:37 +0200
Gábor Boskovits <boskovits@gmail.com> wrote:

> Supressing the last modified header is just an
> add_header Last-Modified "";
> away.
> 
> To get the info from the symlink seems to be much trickier, i would do with
> either embedded perl or embedded lua. I am not sure if we should bother
> with it, though. Wdyt?

Since we already emit ETag, I don't think we need to bother with Last-Modified.

Why is the ETag so short, though?

>wget --debug -O /dev/null   https://guix.gnu.org/packages.json 2>&1 | grep -i etag
>ETag: "1-2f38b1"


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

bug#37207: guix.gnu.org Last Modified at epoch

[Gábor Boskovits]

[-- Attachment #1: Type: text/plain, Size: 1000 bytes --]

Hello Danny,

Danny Milosavljevic <dannym@scratchpost.org> ezt írta (időpont: 2019. aug.
28., Sze, 17:05):

> Hi Gabor,
>
> On Wed, 28 Aug 2019 12:40:37 +0200
> Gábor Boskovits <boskovits@gmail.com> wrote:
>
> > Supressing the last modified header is just an
> > add_header Last-Modified "";
> > away.
> >
> > To get the info from the symlink seems to be much trickier, i would do
> with
> > either embedded perl or embedded lua. I am not sure if we should bother
> > with it, though. Wdyt?
>
> Since we already emit ETag, I don't think we need to bother with
> Last-Modified.
>
> Why is the ETag so short, though?
>
>
The ETag we emit is also bad. Nginx calculates this from mtime and
content-lenght,
so in our case it's just content length.


> >wget --debug -O /dev/null   https://guix.gnu.org/packages.json 2>&1 |
> grep -i etag
> >ETag: "1-2f38b1"
>
>
Best regards,
g_bor

-- 
OpenPGP Key Fingerprint: 7988:3B9F:7D6A:4DBF:3719:0367:2506:A96C:CF63:0B21

[-- Attachment #2: Type: text/html, Size: 1985 bytes --]

bug#37207: nginx serving files from the store returns Last-Modified = Epoch

[Vincent Legoll]

This bug prevents repology [1] to show
the latest versions of packages in guix,
as it relies on 'Last-Modified' for:
https://guix.gnu.org/packages.json
changing in a meaningful way...

[1] https://repology.org/

-- 
Vincent Legoll

bug#37207: nginx serving files from the store returns Last-Modified = Epoch

[Gábor Boskovits]

Hello Vincent,

Vincent Legoll <vincent.legoll@gmail.com> ezt írta (időpont: 2020.
márc. 27., P, 0:07):
>
> This bug prevents repology [1] to show
> the latest versions of packages in guix,
> as it relies on 'Last-Modified' for:
> https://guix.gnu.org/packages.json
> changing in a meaningful way...
>

Does it also use etags, or just last-modified?

I ask this because we already have bug similar to this, and it would
be interesting to know if
it would be enough to have a meaningful etags generation, or we still have to
deal with last-modified.

> [1] https://repology.org/
>
> --
> Vincent Legoll
>
>
>

Best regards,
g_bor
-- 
OpenPGP Key Fingerprint: 7988:3B9F:7D6A:4DBF:3719:0367:2506:A96C:CF63:0B21

bug#37207: nginx serving files from the store returns Last-Modified = Epoch

[Vincent Legoll]

Hello,

On Sun, Mar 29, 2020 at 11:50 AM Gábor Boskovits <boskovits@gmail.com> wrote:
> Does it also use etags, or just last-modified?

From the email exchange I had with the maintainer of the site,
I think it only uses last-modified.

> I ask this because we already have bug similar to this, and it would
> be interesting to know if
> it would be enough to have a meaningful etags generation, or we
> still have to deal with last-modified.

Is etags easier for us to handle ?

-- 
Vincent Legoll

bug#37207: Repology

[Vincent Legoll]

It also paint us a a fairly outdated distro,
despite our efforts to keep the pace and
update to latest versions of packages.

We may even get into the top ten, which
may give us a bit of attention and attract
some distrohoppers^Wusers.

-- 
Vincent Legoll

bug#37207: guix.gnu.org returns Last-Modified = Epoch

[Christopher Baines]

[-- Attachment #1: Type: text/plain, Size: 1329 bytes --]


Ludovic Courtès <ludo@gnu.org> writes:

> Since the use of the ‘static-web-site’ service, which puts web site
> files in the store, nginx returns a ‘Last-Modified’ header that can
> trick clients into caching things forever:
>
> --8<---------------cut here---------------start------------->8---
> $ wget --debug -O /dev/null   https://guix.gnu.org/packages.json 2>&1 | grep Last
> Last-Modified: Thu, 01 Jan 1970 00:00:01 GMT
> --8<---------------cut here---------------end--------------->8---
>
> We should tell nginx to do not emit ‘Last-Modified’, or to take the
> state from the /srv/guix.gnu.org symlink, if possible.

I ended up looking at this again in relation to Repology [1].

1: https://github.com/repology/repology-updater/issues/218#issuecomment-525905704

Going back to that comment, given that the Last-Modified header (and the
ETag) is wrong, it's probably sensible to remove them. That might even
fix the issue with Repology fetching the packages.json file.

Alternatively (or in addition), we could run a really simple Guile web
server that just serves the packages.json file with the right
Last-Modified value, and have NGinx proxy requests to that server. This
would be pretty easy to setup I believe, and would allow providing a
correct value.

Thoughts?

Chris

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 962 bytes --]

bug#37207: guix.gnu.org returns Last-Modified = Epoch

[Ludovic Courtès]

Howdy!

Christopher Baines <mail@cbaines.net> skribis:

> Ludovic Courtès <ludo@gnu.org> writes:
>
>> Since the use of the ‘static-web-site’ service, which puts web site
>> files in the store, nginx returns a ‘Last-Modified’ header that can
>> trick clients into caching things forever:
>>
>> --8<---------------cut here---------------start------------->8---
>> $ wget --debug -O /dev/null   https://guix.gnu.org/packages.json 2>&1 | grep Last
>> Last-Modified: Thu, 01 Jan 1970 00:00:01 GMT
>> --8<---------------cut here---------------end--------------->8---
>>
>> We should tell nginx to do not emit ‘Last-Modified’, or to take the
>> state from the /srv/guix.gnu.org symlink, if possible.
>
> I ended up looking at this again in relation to Repology [1].
>
> 1: https://github.com/repology/repology-updater/issues/218#issuecomment-525905704
>
> Going back to that comment, given that the Last-Modified header (and the
> ETag) is wrong, it's probably sensible to remove them. That might even
> fix the issue with Repology fetching the packages.json file.
>
> Alternatively (or in addition), we could run a really simple Guile web
> server that just serves the packages.json file with the right
> Last-Modified value, and have NGinx proxy requests to that server. This
> would be pretty easy to setup I believe, and would allow providing a
> correct value.
>
> Thoughts?

I think it wouldn’t really help because the Last-Modified issue is
pervasive.  It shows for instance when accessing the web site: one often
has to force the browser to reload pages to get the latest version.

So I’m all for one of the solutions that were proposed earlier.

WDYT?

Ludo’.

bug#37207: guix.gnu.org returns Last-Modified = Epoch

[Christopher Baines]

[-- Attachment #1: Type: text/plain, Size: 2918 bytes --]


Ludovic Courtès <ludo@gnu.org> writes:

> Howdy!
>
> Christopher Baines <mail@cbaines.net> skribis:
>
>> Ludovic Courtès <ludo@gnu.org> writes:
>>
>>> Since the use of the ‘static-web-site’ service, which puts web site
>>> files in the store, nginx returns a ‘Last-Modified’ header that can
>>> trick clients into caching things forever:
>>>
>>> --8<---------------cut here---------------start------------->8---
>>> $ wget --debug -O /dev/null   https://guix.gnu.org/packages.json 2>&1 | grep Last
>>> Last-Modified: Thu, 01 Jan 1970 00:00:01 GMT
>>> --8<---------------cut here---------------end--------------->8---
>>>
>>> We should tell nginx to do not emit ‘Last-Modified’, or to take the
>>> state from the /srv/guix.gnu.org symlink, if possible.
>>
>> I ended up looking at this again in relation to Repology [1].
>>
>> 1: https://github.com/repology/repology-updater/issues/218#issuecomment-525905704
>>
>> Going back to that comment, given that the Last-Modified header (and the
>> ETag) is wrong, it's probably sensible to remove them. That might even
>> fix the issue with Repology fetching the packages.json file.
>>
>> Alternatively (or in addition), we could run a really simple Guile web
>> server that just serves the packages.json file with the right
>> Last-Modified value, and have NGinx proxy requests to that server. This
>> would be pretty easy to setup I believe, and would allow providing a
>> correct value.
>>
>> Thoughts?
>
> I think it wouldn’t really help because the Last-Modified issue is
> pervasive.  It shows for instance when accessing the web site: one often
> has to force the browser to reload pages to get the latest version.
>
> So I’m all for one of the solutions that were proposed earlier.
>
> WDYT?

So I think removing the Last-Modified header from the responses will fix
the issue with the Repology fetcher (as it will stop thinking it's
already fetch the file, since it was last modified in 1970), instead it
will just always process the file.

Removing the Last-Modified header, and maybe the ETag as well from
responses should avoid the issue with web browsers using a cached
version of the page when they probably shouldn't.

I realise what I described with using a Guile web server to serve the
packages.json file wouldn't help with other pages (unless they're served
as well, which is a possibility), but that was just an optimisation over
removing the header entirely, as having the Last-Modified header, with a
correct value would help the Repology fetcher cache the file.

Does that make sense? It still seems to me that a small change to the
NGinx config (I think these lines somewhere in the config would do it
[1]) would help with the Repology fetcher issue, and the issue you
describe with web browsers.

1:

add_header Last-Modified "";
if_modified_since off;
etag off;

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 962 bytes --]

bug#37207: guix.gnu.org returns Last-Modified = Epoch

[Ludovic Courtès]

Hi,

Christopher Baines <mail@cbaines.net> skribis:

> So I think removing the Last-Modified header from the responses will fix
> the issue with the Repology fetcher (as it will stop thinking it's
> already fetch the file, since it was last modified in 1970), instead it
> will just always process the file.
>
> Removing the Last-Modified header, and maybe the ETag as well from
> responses should avoid the issue with web browsers using a cached
> version of the page when they probably shouldn't.

It would prevent client-side caching altogether.  So perhaps we can do
that as a stopgap (and it has the advantage of requiring only a tiny
config change).

Eventually, it’d be nice to have something better, like one of the Etag
patches discussed upthread.

Thanks,
Ludo’.

bug#37207: guix.gnu.org returns Last-Modified = Epoch

[Christopher Baines]

[-- Attachment #1: Type: text/plain, Size: 805 bytes --]


Ludovic Courtès <ludo@gnu.org> writes:

> Hi,
>
> Christopher Baines <mail@cbaines.net> skribis:
>
>> So I think removing the Last-Modified header from the responses will fix
>> the issue with the Repology fetcher (as it will stop thinking it's
>> already fetch the file, since it was last modified in 1970), instead it
>> will just always process the file.
>>
>> Removing the Last-Modified header, and maybe the ETag as well from
>> responses should avoid the issue with web browsers using a cached
>> version of the page when they probably shouldn't.
>
> It would prevent client-side caching altogether.  So perhaps we can do
> that as a stopgap (and it has the advantage of requiring only a tiny
> config change).

Great, I've finally got around to sending a patch for this now.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 962 bytes --]

bug#37207: [PATCH] nginx: berlin: Work around Last-Modified issues for guix.gnu.org.

[Christopher Baines]

* hydra/nginx/berlin.scm (%berlin-servers): Add some config to the
nginx-server-configurations for guix.gnu.org.
---
 hydra/nginx/berlin.scm | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/hydra/nginx/berlin.scm b/hydra/nginx/berlin.scm
index 303fd35..8c90eb1 100644
--- a/hydra/nginx/berlin.scm
+++ b/hydra/nginx/berlin.scm
@@ -514,6 +514,13 @@ PUBLISH-URL."
     (locations guix.gnu.org-locations)
     (raw-content
      (list
+      ;; TODO This works around NGinx using the epoch for the
+      ;; Last-Modified date, as well as the etag.
+      ;; See http://issues.guix.info/issue/37207
+      "add_header Last-Modified \"\";"
+      "if_modified_since off;"
+      "etag off;"
+
       "access_log /var/log/nginx/guix-info.access.log;")))
 
    (nginx-server-configuration
@@ -634,6 +641,13 @@ PUBLISH-URL."
      (append
       %tls-settings
       (list
+       ;; TODO This works around NGinx using the epoch for the
+       ;; Last-Modified date, as well as the etag.
+       ;; See http://issues.guix.info/issue/37207
+       "add_header Last-Modified \"\";"
+       "if_modified_since off;"
+       "etag off;"
+
        "access_log /var/log/nginx/guix-gnu-org.https.access.log;"))))
 
    (nginx-server-configuration
-- 
2.26.2

bug#37207: nginx serving files from the store returns Last-Modified = Epoch

[anadon via web]

 Any movement on this?