ARM builders connectivity issues cause the build to fail

Note: Potentially related in some way to https://bugs.launchpad.net/launchpad/+bug/2023961?

I don't think they are related, because this one affects the Launchpad
builders themselves and are only observable for snap builds, as only
those access the internet; In LP:2023961 the problem described is
uploading a orig tarball to PPA, the failure of which prevents any build
whatsoever from starting, and even when in the past I didn't have that
issue and managed to get builds there, the problem described in this
LP:2024181 never affected any builder because such builds do not access
the internet.

A few notes on this issue:
I have been trying to reproduce it for a while now, and don't seem to be able to. I
tried to stress our testing environments, and tried a few things in production, without success.

Our current best guess is that it might have been related to the load of the builders. I don't see particular peaks in the dates where the builds mentioned in this ticket failed, except for the 1st of June where there was a firewall issue. But it's hard to discard that was the reason.

Ideally we would set up performance tests for the builders.

Thank you very much for taking a closer look into this issue, Ines.

 > Our current best guess is that it might have been related to the load
of the builders.

 From my side I've been observing this happening more or less uniformly
(i.e., the chance of failure per ARM build doesn't noticeably change)
for a long time.

 > I have been trying to reproduce it for a while now, and don't seem to
be able to. I tried to stress our testing environments, and tried a few
things in production, without success.

To be clear, I never observe this when downloading from Ubuntu archives.
Did your test involve requests to third-parties? Maybe it's actually
Google servers that are misbehaving, not the builders (and then I should
direct my report elsewhere)? Amin (in the Mozilla team) reports similar
observations for Firefox builds, but for him the rate of failure is much
smaller, so I'm unsure.

If you try testing with this one, can you reproduce the failure?

--->
wget
https://commondatastorage.googleapis.com/chromium-browser-official/chromium-114.0.5735.45.tar.xz
<---

I did try specifically with that chromium artifact after seeing all the "Connection Broken" errors came during that download.

I also tried building your artifact in our test environment a few times, and it never failed that or other downloads.

This is an odd one.

We still haven't completely ruled out that it might be related to builders load, but it might also be the case that Google servers were misbehaving.

Let us know if you keep getting that error.

I got an error again[1]. This is a new error message; I'm adding it to the bug description.

[1] https://launchpad.net/~chromium-team/+snap/chromium-snap-from-source-dev/+build/2150102/+files/buildlog_snap_ubuntu_jammy_arm64_chromium-snap-from-source-dev_BUILDING.txt.gz

Actually I got some more but this could be a general instability as they fail to pull from the archive and I also got failure on x86 at about the same time.

--[x86]-->
Cloning into 'chromium'...
[26/Jun/2023:07:03:10 +0000] "CONNECT git.launchpad.net:443 HTTP/1.1" 407 1913 "-" "git/2.34.1"
fatal: unable to access 'https://git.launchpad.net/~nteodosio/chromium-browser/': Received HTTP code 407 from proxy after CONNECT
<---

--[arm64]-->
Cannot install all requested build packages: chrpath, cmake, default-jre-headless, elfutils, g++, gcc, git, gperf, gzip, libasound2-dev, libcap-dev, libcups2-dev, libcurl4-openssl-dev, libevdev-dev, libffi-dev, libicu-dev, libkrb5-dev, libnss3-dev, libpam0g-dev, libpci-dev, libpipewire-0.3-dev, libssl-dev, libsystemd-dev, libva-dev, libxshmfence-dev, libxss-dev, lsb-release, make, mesa-common-dev, ninja-build, pkg-config, python3-pkg-resources, python3-xcbgen, qtbase5-dev, quilt, sed, subversion, tar, wget, xcb-proto, yasm
<---

[1] https://launchpadlibrarian.net/674077435/buildlog_snap_ubuntu_jammy_amd64_unstage-linters_BUILDING.txt.gz
[2] https://launchpadlibrarian.net/674077632/buildlog_snap_ubuntu_jammy_arm64_chromium-snap-from-source-dev_BUILDING.txt.gz

> more rarely they are restarted without any intervention — the logs are wiped anew, but the build time isn't reset when this happens.

As an example of this I just got https://launchpad.net/~chromium-team/+snap/chromium-snap-from-source-dev/+build/2150101, which took 318 min.

This one took 45 hours. Certainly it is also a case of the build being silently reset, because there is absolutely no way the run captured in the log would take that long.

https://launchpad.net/~chromium-team/+snap/chromium-snap-from-source-dev/+build/2150102

By the way let me know if there are enough data points so I stop posting them here.

Hi Nathan!

Thanks for keeping us up to date with information.

The error that happened the most consistently when you first reported these was

--->
Downloading 'chromium-114.0.5735.90.tar.xz' 0%
Sorry, an error occurred in Snapcraft:
("Connection broken: ConnectionResetError(104, 'Connection reset by peer')", ConnectionResetError(104, 'Connection reset by peer'))
<---

I recall seeing it in 4 different logs. Has this one particularly happened again?

I suspect that there is more than one issue going on here, and it's hard to separate what is what currently.

Regarding the job that took 45 hours, I would expect that any job that takes that long would fail with connection errors because the proxy token only lasts about 6 hours, meaning any connections after that token expire would fail...
But also, we had a few issues (now fixed) in the past 3-4 days that could potentially lead to builder being busy retrying builds, so I wonder if generally busy builders ===> your builder's connection issues. Those issues have been fixed now, so would be interesting to know if there is indeed a correlation here.

Hi Inês!

 > I recall seeing it in 4 different logs. Has this one particularly
 > happened again?

This particular one I don't think so.

 > Regarding the job that took 45 hours, I would expect that any job
that takes that long would fail with connection errors because the proxy
token only lasts about 6 hours, meaning any connections after that token
expire would fail...

Although Chromium builds never take less than 6 h — on ARM it takes at
least 1 day —, as far as I can tell no connection is attempted after
this step is completed:

:: + python3 tools/clang/scripts/build.py --skip-checkout --bootstrap
--disable-asserts --pgo --without-android --without-fuchsia
--use-system-cmake --with-ml-inliner-model=

Which is rather early in a successful build[1].

So I believe you are right and there is more than one issue going on.

[1]
https://launchpadlibrarian.net/674561689/buildlog_snap_ubuntu_jammy_arm64_chromium-snap-from-source-stable_BUILDING.txt.gz

Still getting several of

--->
:: + craftctl default
None
Full execution log: '/root/.local/state/snapcraft/log/snapcraft-20230720-103153.207821.log'
Build failed
<---

accompanied by the mysterious silent resets.

https://launchpadlibrarian.net/678116449/buildlog_snap_ubuntu_jammy_arm64_chromium-snap-from-source-beta_BUILDING.txt.gz
https://launchpad.net/~chromium-team/+snap/chromium-snap-from-source-dev/+build/2173457/+files/buildlog_snap_ubuntu_jammy_armhf_chromium-snap-from-source-dev_BUILDING.txt.gz