1255890 - Linux32 as tier 2 platform

Status: RESOLVED INVALID

(Infrastructure & Operations Graveyard :: CIDuty, task)

Description

[Kim Moir [:kmoir] ET]

In bug 1209932 we are looking at disabling linux32 testing 

Reasons
* Linux32 is a very small user base compared to Windows yet we allocate expensive resources to it.
* Most people don't use our Linux32 builds, but rather the ones from their distro. (Although it appears that there are larger numbers of linux32 users vs linux64 on some distros https://groups.google.com/forum/#!msg/mozilla.dev.planning/wBgLRXCTlaw/NvV-05jQBgAJ)
* We do not have reliable telemetry numbers for Linux32 from distros for various reasons 

There is some recent discussion on dev.planning

https://groups.google.com/forum/#!searchin/mozilla.dev.planning/revisiting$20linux32/mozilla.dev.planning/URCpj-kTjVU/OS8jd8QTAgAJ

If we disable linux32 testing, then it's we shouldn't ship linux32 as a release build.  

So I am looking for insight from relman on this issue. 

As an aside, Chrome is also moving in this direction
http://www.omgubuntu.co.uk/2016/01/google-chrome-linux-32-bit-discontinued

Comment 1

[Kim Moir [:kmoir] ET]

If relman has numbers on linux32 users from the various linux distributions that consume Firefox, this would be helpful in making the decision. The numbers provided on the dev-planning list above don't reflect all distributions.

Comment 2

[Lawrence Mandel [:lmandel] (use needinfo)]

I think this we need to include Firefox Product Management in this decision. I also think they have access to some of these numbers from work already in progress. ni Jeff, with whom we've been discussing other platforms recently.

Comment 3

[Chris Coulson]

(In reply to Kim Moir [:kmoir] from comment #1)
> If relman has numbers on linux32 users from the various linux distributions
> that consume Firefox, this would be helpful in making the decision. The
> numbers provided on the dev-planning list above don't reflect all
> distributions.

As of Firefox 45, Ubuntu Firefox builds are sending telemetry - will this help?

Comment 4

[Sylvestre Ledru [:Sylvestre]]

Debian is not a reference for desktop (mostly used for server) but usage of i386 is still high:
http://popcon.debian.org/stat/submission.png

I don't think Debian has statistics of firefox per archs.

Comment 5

[ISHIKAWA, Chiaki]

(In reply to Sylvestre Ledru [:sylvestre] from comment #4)
> Debian is not a reference for desktop (mostly used for server) but usage of
> i386 is still high:
> http://popcon.debian.org/stat/submission.png
> 
> I don't think Debian has statistics of firefox per archs.

As a Debian user of a long time, and an occasional contributor of patches to TB,
I have a comment.

I have been a user of Debian for more than a dozen years, and I can offer a reason why i386 is high in Debian.
Debian is VERY CONSERVATIVE and stick to stable version of the applications for a long time.
Thus, I have a version of i386 Debian installation for more than 10 years just because I love the stability due to its conservatism. For business use, it is important.
I think many installations with i386 version today will do so until Debian drops i386 (may never happen!).

OTOH, when one is faced with installing Debian on a new x86_64 hardware, one has a choice of installing either i386 x86_64 version and when one's machine is likely to have more than 4GB, naturally the choice is x86_64. 

As of now, I have three installation of Debian GNU/Linux.
One PC 64-bit inside VirtualBox (upgraded from 32-bit to 64-bit in 2015 after about a dozen years of usage of 32-bit native version, then imported to virtualbox a few years ago.).
ONE PC 32-bit native (have been using this more than 12 years, have lost track of when exactly. I think it is from 2001 originally. To be exact, the machine (CPUs et al) have been replaced a few times, but the OS has remained the same.).
One PC 64-bit inside VivrtualBox (it was installed as 64-bit initially about 6-7 years ago.)

And I am using 32-bit versions of Thunderbird directly from mozilla distribution. I think Debian has dropped producing its own version of TB sometime ago.
64-bit Debian GNU/Linux has features to support 32-bit applications like 32-bit TB.

I know I am only ONE sample: but when it comes to Debian GNU/Linux, I think the usage of i386 version can be explained by the stability (and "If it ain't broke, don't fix it" attitude of many. For business use, this is inevitable.), so unless the existing i386 hardware installations will begin to malfunction suddenly, the absolute # of i386 usage won't go down any time soon. And, I will be a little upset if 32-bit version of linux TB is officially dropped by mozilla.

Some may ask why I don't use Ubuntu. Hard to say. Usually, I don't trust new distributions until it has proven itself. Ubuntu has proved itself. But it is a derivative of Debian, and I have no killer app that runs only in Ubuntu, etc. So Debian is fine. (I forgot to explain, there was a time when I used Fedora for a few years on a PC, but it was a tad difficult to keep up with the new updates and breakages. Debian's conservatism helps in office settings.)

I think if mozilla announces the dropping of 32-bit linux version one year advance, that would be hugely appreciated.

TIA

Comment 6

[Jeff Griffiths (:canuckistani) (:⚡︎)]

(In reply to Lawrence Mandel [:lmandel] (use needinfo) from comment #2)
> I think this we need to include Firefox Product Management in this decision.
> I also think they have access to some of these numbers from work already in
> progress. ni Jeff, with whom we've been discussing other platforms recently.

Based on the insight provided in the original email thread, I have no objections to making Linux 32 as a tier 2 platform, with the following caveats:

* it was reported that we still test windows 7 / 32bit
* we should change our contribution documentation to reflect that Linux 64 is preferred when contributing to Firefox

Comment 7

[Kim Moir [:kmoir] ET]

Lawrence given that we have agreement from Product Management on making Linux32 a tier 2 platform, is the removal of Linux32 something that should ride the trains so we can continue to release the platform until it is deprecated or is it something that should be removed from the release process entirely on a given date?

Comment 8

[Lawrence Mandel [:lmandel] (use needinfo)]

I think we can drop support soon. Let's not make any changes this cycle (46). I don't see a problem in doing so in the next cycle (47) and am happy to move the builds to tier 2 for m-c and m-a as soon as we can update the download pages to avoid breaking links on mozilla.org.

Sylvestre - We're going to need to update the download pages and may want to modify the system requirements (although we need to decide whether we want to explicitly state no support for Linux32 there).

https://www.mozilla.org/en-US/firefox/all
https://www.mozilla.org/en-US/firefox/beta/all/
https://www.mozilla.org/en-US/firefox/developer/all/
https://nightly.mozilla.org/

Comment 9

[Sylvestre Ledru [:Sylvestre]]

Hmm, I didn't understand that we were going to remove it from the download page. Do we know how many downloads do we still have on these pages?

Comment 10

[Mike Hommey [:glandium]]

(In reply to Kim Moir [:kmoir] from comment #0)
> If we disable linux32 testing, then it's we shouldn't ship linux32 as a
> release build.

My understanding was that we were assuming that testing linux64 would be good enough to assess the linux32 builds, not that we would stop shipping linux32.

Comment 11

[Mike Hommey [:glandium]]

Also, if we do go with not shipping linux32, do we have an upgrade story for the people using those builds?

Comment 12

[Kim Moir [:kmoir] ET]

glandium: We don't ship release builds unless there is testing associated with it. With respect to the update story, that is a good question.  I would assume that users would be encouraged to update to linux64 where possible, let me find out more specific information,

Comment 13

[Phil Ringnalda (:philor)]

We do, actually: we stopped testing the 32-bit half of the Mac universal builds years ago.

Comment 14

[Sylvestre Ledru [:Sylvestre]]

> We do, actually: we stopped testing the 32-bit half of the Mac universal
> builds years ago.
Stop testing is one thing, stop building and releasing another.

Comment 15

[Kim Moir [:kmoir] ET]

Sylvestre: I don't know where we would get the number of downloads for linux32 from mozilla directly vs distros.  Metrics?

I looked into what other browser vendors did when they deprecated linux32. The distros stopped shipping the linux32 version, stopped shipping updates and advised users to upgrade to linux64. Simple message to the mailing list.  The responses to the mailing list were not so positive but that is standard practice when deprecating platforms in open source projects.

Comment 16

[Chris Coulson]

(In reply to Kim Moir [:kmoir] from comment #15)
> Sylvestre: I don't know where we would get the number of downloads for
> linux32 from mozilla directly vs distros.  Metrics?
> 
> I looked into what other browser vendors did when they deprecated linux32.
> The distros stopped shipping the linux32 version, stopped shipping updates
> and advised users to upgrade to linux64. Simple message to the mailing list.
> The responses to the mailing list were not so positive but that is standard
> practice when deprecating platforms in open source projects.

I can't see us (Ubuntu) dropping support for x86/linux32 just yet, given that Ubuntu 16.04 is about to be released with it as a supported platform and a support length of 5 years.

Which other browser vendors are you referring to? The only other major browser that's commonly packaged by distros is Chromium, and despite Google dropping 32-bit builds for Chrome 49 it seems that Debian and Ubuntu at least are still providing 32-bit builds of Chromium (and our Chromium maintainer plans to continue doing so).

Ubuntu will have to continue to ship 32-bit builds of Firefox for the foreseeable future. My only concern is how long it will be before we're unable to ship updates, given how quickly other tier 3 platforms rot, Ubuntu is the only distro not shipping the ESR, and it doesn't have a Firefox maintainer. I guess that's something for us to resolve though.

Comment 17

[Chris Cooper [:coop] (he/him)]

I think this is a non-starter for now. We can revisit in the future.