Stop waiting for method response when browser disconnects

[Mike Rylander]

When inside mod_perl, and the client disconnects, we should stop trying to receive results from a request. This will help prevent apache backend pile-ups when a cat sits on the Enter key at an OPAC computer.

Branch at: http://git.evergreen-ils.org/?p=working/OpenSRF.git;a=shortlog;h=refs/heads/user/miker/give-up-at-browser_stop

[Mike Rylander (mrylander)]

Here's the Evergreen half of this change:

http://git.evergreen-ils.org/?p=working/Evergreen.git;a=shortlog;h=refs/heads/user/miker/facilitate-browser_stop-short-circuit

[Galen Charlton (gmc)]

Using %ENV to pass the request object around won't work for Perl 5.18 and later, as stuff assigned to %ENV gets immediately stringified:

http://perldoc.perl.org/perlvar.html#%25ENV

It worked prior to 5.18 because the stringification used to happen only when passing the environment to child processes.

An alternative would be ensuring that +GlobalRequest is turned on for /eg, which in turn would allow use of Apache2::RequestUtil->request (although per bug 1579219 I at least would prefer that the check be wrapped well enough so that mod_perl remains an optional dependency for OpenSRF :) )

Removing pullrequest for now.

[Mike Rylander (mrylander)]

Ah! Thanks, modern perl.

In that case, I think it's simplest (both in code and in administrator effort) to use an OpenSRF.pm package-local variable wrapped in an API to accept a request-looking object, rather than requiring OpenSRF to include Apache2::RequestUtil. We just got rid of mod_perl as a dep ... :)

Thoughts on that alternative?

[Mike Rylander (mrylander)]

I've updated the two branches linked above to make that happen. Note that using an API means that Evergreen requires a supporting OpenSRF, instead of silently degrading to the old behavior as setting ENV (or another variable) would allow.

[Galen Charlton (gmc)]

I've pushed the (squashed) patches to OpenSRF for inclusion in 2.5-beta. Thanks, Mike! I'll continue testing the Evergreen side as a potential candidate bugfix for Evergreen 2.12-rc.

[Evergreen Bug Maintenance (bugmaster)]

Bumping to 3.2.2 so it is not lost there. It should be backported to Evergreen 3.1. also, but since this is an OpenSRF bug, we can apparently only target 1 Evergreen milestone in the Evergreen portion.

[Chris Sharp (chrissharp123)]

Seeing this error when doing a bib search after applying this patch over current Evergreen master on OpenSRF master:

2018-11-06 11:00:22 csharp-master-xenial gateway: [ERR :28889:Search.pm:466:15415199542888913] multiclass search error: Can't locate object method "aborted" via package "Apache2::Connection" at /usr/local/share/perl/5.22.1/OpenSRF/Transport/SlimJabber/XMPPReader.pm line 228.
2018-11-06 11:00:42 csharp-master-xenial gateway: [perl:error] [pid 28892] [client ::1:58072] egweb: Context Loader error: Can't locate object method "aborted" via package "Apache2::Connection" at /usr/local/share/perl/5.22.1/OpenSRF/Transport/SlimJabber/XMPPReader.pm line 228.\n, referer: https://csharp-master.gapines.org/eg/opac/results?query=sd&qtype=keyword&fi%3Asearch_format=&locg=1&detail_record_view=0&_adv=1&sort=

[Mike Rylander (mrylander)]

Chris,

Based on the error message, it looks like you're using a mod_perl from the 1.x series, which doesn't have an aborted() method in Apache2::Connection. Looks like that showed up with version 2.0.0. Can you confirm your Apache2 (perl module) and mod_perl versions?

Thanks!

[Chris Sharp (chrissharp123)]

Well, after installing your commits atop Evergreen master running on OpenSRF master on Ubuntu 16.04 (Apache 2.4.18/mod_perl version 2.0.9), I'm not seeing the errors. I don't know what was different back in November when I tested before, but I don't see a problem with this so far in my testing. I'll poke it some more and sign off as soon as I'm satisfied that everything looks fine.

[Blake GH (bmagic)]

I attempted to merge this onto 3.9.0. It seems that this patch is already merged? With the small change of

$ENV renamed to: OpenSRF

?

[Jason Boyer (jboyer)]

Hi Mike, I was testing this today and nothing I could do would trigger the "Upstream Apache client disconnected, aborting." log message, only the apache broken pipe message that normally appears when a client goes away. I tried searching for 'test' and just holding down Enter, along with searching for just 'a' the same way.

[Mike Rylander (mrylander)]

Thanks for testing, Jason!

I wonder if having nginx in the middle is impacting the ability of apache to see when the client went away. I'll look into that a bit.

[Bill Erickson (berick)]

Removing pullrequest tag pending resolution to previous comment.