scripts/osmo-ci-docker-rebuild: switch to trixie (details)
Commit
0ba4ce01b652c74832817ddd8eb893abc791c3a5
by Oliver Smith
build-old-tags: ignore failing due to api break
Ignore osmo-bsc and osmo-msc versions that are known to fail due to intentional API break. osmo-bsc 1.12.2 and osmo-msc 1.12.2 build are forward-compatible.
Note that previous releases don't need to be listed in tags_to_ignore() as this script only builds the last 3 release tags.
Store the apk after successful run of the android-apdu-proxy job. It can be found by clicking on a job at [1] and then clicking on "default" (since the job is behind a matrix, as all master jobs).
Commit
89a3b3f049c13c8d7e3ba17e305990fc9ea21f3d
by laforge
jobs/osmocom-release-android-apks: new job
Add a job similar to the release tarballs job, that iterates over the tags of the android app repositories (currently only android-apdu-proxy) and builds/uploads an apk for each of them.
Commit
b7e67c1df52050ebaa23437db8dc405dd223c446
by Oliver Smith
jobs/ttcn3-testsuites: adjust timer for bts
BTS testsuites tend to fail with timeouts on heavy loads. Group them in a timeslot where jenkins nodes are otherwise not busy according to stats in grafana to hopefully make the test results more stable. Once we can replace faketrx with a rewrite (OS#6672), this should not be necessary anymore.
Commit
0ba4dac82b83a8e6515c75e1b792014375595f37
by Oliver Smith
jobs/master,gerrit: use io_uring seccomp profile
Use the seccomp profile not only in ttcn3 testsuites, but also in master-builds and gerrit-verifications so we can test io_uring there too.
CI happened to work without this for libosmocore on build4 because of a specific docker version where io_uring was not yet (fully?) disabled in the default seccomp profile. It did not work without this patch on build5 where we currently have a newer docker version.
Commit
ef498e7efe3a5801bbfbf4b6e90a4e54bb3ee6e5
by Oliver Smith
jobs/{master,gerrit}: set ulimit filesize to 1 GiB
Set a max file size limit of 1 GiB via ulimit. This prevents jobs from filling up the disk by generating one huge log file.
I have chosen 1 GiB as limit, because jobs that clone linux kernel repos may have packed git artifacts that are several 100 MiB in size.
It would be even better to set the overall max disk space that can be used from one "docker run" call, but this is currently not supported for ext4: https://github.com/moby/moby/issues/29364
Commit
a71ccaa13598d4d5627b4e16e497f4d7d076b53e
by Oliver Smith
jobs: manually trigger "git gc" for linux.git
We already use shallow clones for linux.git, and git runs a garbage collector automatically to ensure no longer needed artifacts are removed. But this did not work anymore on build4 for build-kernel-net-next, and it increased up to 8.4 GiB:
8,4 GiB /build-kernel-net-next
After running "git gc" again:
3,3 GiB /build-kernel-net-next
The reason was that "git gc" did not exit successfully once and wrote a log file, then refused to run again. Run "git gc --auto" explicitly in foreground to let abort in such a case, so it does not go unnoticed in the future.
Commit
fa7a94dd4e75e8ead20d4cc5d1ed79a64a85f4ff
by Oliver Smith
scripts/docker-cleanup: add "docker system prune"
On build4 this freed up another 8.223GB when I just ran it:
$ docker system prune WARNING! This will remove: - all stopped containers - all networks not used by at least one container - all dangling images - all dangling build cache
Commit
7b4109ff906dd933d0db425bf2fb686740046646
by Oliver Smith
ansible/setup-jenkins-slave: add mingw job deps
Support running the legacy job Osmocom-OBS_MinGW_weekly_publish by installing its dependencies directly in the LXC. This is a workaround until we modernize the job, see the related issue.
Commit
e22ca0edcf020d1eca4cf20a729e50faa13cb9e7
by Oliver Smith
ansible: update build5-obsworker ssh key
I've reinstalled the build5-obsworker lxc after first upgrading to openSUSE 15.6 and then discovering that the lxc version is incompatible with obs-build: 15.6 doesn't have lxc 4 anymore, but a higher version is not supported by obs-build. I've added a big warning in the wiki so we don't run into this again until obs-build has been adjusted.
Commit
31db96c312b953afc26ae4b16b266e367a0fdcef
by Oliver Smith
lint/lint_diff: run clang-format if config was found
The osmo-asf4-dfu and osmo-ccid-firmware now have clang-format configs. Adjust the lint script to run clang-format if such a config is found in the repository for which the lint script is running.
Commit
3dffb34587dc0bba9ce46c7ef0d7e07c8167a8d4
by Oliver Smith
lint/lint_diff: allow skipping with OSMO_LINT=0
When this script is set as pre-commit hook, it can be useful to skip the linter with this env var to make a commit that intentionally does not pass the linter (e.g. to test if it will be caught in gerrit + jenkins).
There is also "git commit --no-verify", but with that the hook that applies the Change-Id does not run.
Commit
9a18ebaa78fc255a0aea42186a2743829d69162e
by Oliver Smith
obs: gerrit_binpkgs: set --allow-unknown-package
The gerrit_binpkgs.sh script can be used to quickly test building of debian packages in a container even for packages that we don't have in the OBS osmocom repositories. For example I'm currently using it to test-build kamailio packages. Use the --allow-unknown-package argument so build_srcpkg.py doesn't complain if this gets called with a package that isn't a known Osmocom package.
Commit
e28454ea4b120a7ab96e133589a7b7b3ee8c0846
by Oliver Smith
*.py: ignore: E722 Do not use bare `except`
Let the ruff linter not complain about the existing cases of "Do not use bare `except`" so we get the linter to pass without changing the existing logic here. With this we don't disable the check globally, so we can consider making stricter excepts in new code.
Commit
c8ec5d58507718165b863c36b51e29e9afd1a4a1
by Oliver Smith
scripts/verify_log_statements: fix undefined name
Append it to "errors_found" instead of undefined variable "ignored". This makes it work similarly to further above:
# Skip 'LOG("bla" fmt )' strings that typically appear as #defines. if fmt_re.match(quoted): if debug: errors_found.append(error_found(f, log.start(), 'Skipping define', log.group(0))) continue
Fix the following bug:
scripts/verify_log_statements.py:92:11: F821 Undefined name `ignored` | 90 | # hopefully eval broke because of some '## args' macro def 91 | if debug: 92 | ignored.append(error_found(f, log.start(), 'Ignoring', log.group(0))) | ^^^^^^^ F821 93 | continue |
Commit
b97a04f431503a5aa0a0f07a68910153c13721be
by Oliver Smith
jobs/osmocom-build-tags…: set seccomp profile
Set the seccomp profile to make io_uring work in docker, so the regression test in libosmocore using io_uring does not fail. This happened to work previously on the build4 jenkins node because of an older docker version.
Commit
01602b16c59838be23259762a5c1d99a0240e764
by Oliver Smith
OBS: cleanup: new script
Add a script for cleaning old sources uploaded to OBS. This was needed as we were running out of space on the OBS server, due to source tarballs being generated in nightly jobs (e.g. Osmocom_OBS_nightly and never getting deleted). This script is now running automatically every day.
The script does the following: * Iterate over all OBS projects and packages inside the projects. * Delete all source files matching *.dsc, *.tar.xz in old revisions from disk, which are not in the most recent revision anymore. * Create an empty file instead (as otherwise OBS becomes unstable) * Write a marker file with the revisions already iterated through (per project+package) and skip those on the next run.
Since osmo-pcap.git Change-Id I72e8a6ceb4fb1eb70372e13bb139ead0e2bc0860, osmo-pcap-client uses osmo_stream which in turn uses osmo_io (which can u io_uring backend). Hence, add io-uring backend to the test matrix for this project.
Commit
06b815c32d7a639e716f657720795baff2bb4c9c
by Oliver Smith
jobs: free up runners from 01:00 - 03:00
When looking at grafana, the build hosts (build4 and build5) are not busy during 00:00 and 03:00. Binary packages seem to be finished at 00:00 already. Shorten the time for "Jobs that don't need binary packages" by two hours to make room for running ttcn3-bts-test with less load in a follow-up patch.
Commit
e5493058eb750818c1460346c92bd7700255d2a1
by Oliver Smith
jobs: run ttcn3-bts-test-* between 01:00 - 06:00
The automatic respawns of osmo-bts-trx have been intentionally removed in the related docker-playground patch, so we can better notice when osmo-bts-trx crashes instead of just restarting it and moving on with the testsuite.
The result is that now it is even more important when ttcn3-bts-test jobs are running, because when the builders are under high load then faketrx cannot send the time quickly enough and it also causes osmo-bts-trx to shutdown.
In a discussion with Vadim we've decided to schedule the ttcn3-bts-tests so that they don't only run at a time where no other jobs are running. But also that they don't run at the same time on the same builder anymore to give them the least unrelated load possible. Other solutions are also being investigated, but this should make the job in its current form more stable already.
Move misc ttcn3 jobs to run from 06:00 to 08:00 and run ttcn3-bts-tests between 01:00 and 06:00. Pin them to specific runners to make sure we don't have both running on the same node.
Commit
88bc9b1e07c1f31c410780288658cbe0c4076737
by Oliver Smith
ansible: build-hosts: add testenv-coredump-helper
The Osmocom jenkins nodes run inside LXCs. When we get a coredump it appears on the host. Add a helper script to the hosts so the jenkins jobs can fetch the coredumps in case an Osmocom program crashes while running a ttcn3 testsuite.
The helper script has the following safety features to ensure jenkins can't just fetch any coredump:
* Only fetch coredumps within the last 3 seconds and only if the executable matches osmo-* or open5gs-* * Only listen on the lxc IP
Commit
56e77437c64f252765302ee9eb5739da21e9af2d
by Oliver Smith
jobs/master: pass osmo-ci to docker
Let jobs such as the upcoming one for osmo-rp2040-firmware use the known_hosts file from osmo-ci when uploading binaries. In other jobs we've copied the known_hosts into the contrib/jenkins.sh file, but let's try to have it in a central place at least for new jobs.
Fix that installing openjdk-21 as builddep of eclipse-titan fails with: Setting up openjdk-21-jre-headless:amd64 (21.0.7+6~us1-0ubuntu1~24.04) ... update-alternatives: using /usr/lib/jvm/java-21-openjdk-amd64/bin/java to provide /usr/bin/java (java) in auto mode update-alternatives: error: error creating symbolic link '/usr/share/man/man1/java.1.gz.dpkg-tmp': No such file or directory
Commit
9db271c7b325fd57e84adaaedba8644b2db03d23
by Oliver Smith
OBS: gerrit_binpkgs: add FEED parameter
Allow switching to another feed easily. I'm using this to build packages for ubuntu quickly, for which we don't have a master feed (only for debian 10, 11, 12 currently).
Commit
3b348d5effcd8b6cfde611b00473dd37545cebbb
by Oliver Smith
OBS: remove pdf draft watermark for latest pkgs
Add the line ASCIIDOCSTYLE = $(BUILDDIR)\/custom-dblatex.sty before the line include $(OSMO_GSM_MANUALS_DIR)/build/Makefile.asciidoc.inc when building source packages for the latest feed, so there is no "DRAFT" watermark.
Commit
9f5b48a2978e652671a51b6511a068e05798cffc
by Oliver Smith
scripts/manuals: use new method for removing DRAFT
osmo-gsm-manuals now has two "ASCIIDOCSTYLE" files: * custom-dblatex.sty (without draft) * custom-dblatex-draft.sty (new default)
This change was necessary to make it feasible to build the manuals inside the debian binary packages without the watermark for the osmocom:latest feed (see next patch).
Adjust the method for removing the "DRAFT" watermark: instead of patching custom-dblatex.sty, switch from custom-dblatex-draft.sty to custom-dblatex.sty.
Commit
f7e3533392ef98ba46ed1e1b4ac3a60363de3bd2
by laforge
jobs/ttcn3-testsuites-testenv: no rm on error
Do not delete the cache dir if testenv returns with error, so we can inspect it. Keep the linux kernel too, it doesn't make much of a size difference and makes the code slightly simpler.
Commit
d0c3e3bf9bb540cd82f71b2a47c109b3889c401a
by Oliver Smith
OBS: update_obs_eclipse_titan: update
* Set the branch as argument instead of hardcoding it. * Don't use --version-append, the ~osmocom part will be added to debian/changelog instead when pushing new versions. * Add usage description.
Commit
43c09c91818d9c62e60896c3680b1bea05a48604
by Oliver Smith
OBS: add the new osmocom-keyring package
The public key for OBS packages needs to be extended every two years. When the previous key expires, it is not possible to update packages anymore until users replace the old key with the new, extended key.
To make this less inconvenient for users, introduce a new osmocom-keyring package that holds the current key (valid for 2 years), which will get updated every year with an extended key (valid for another 2 years, I'll add a jenkins job to remind us). This means as long as users upgrade their packages once a year, they will automatically receive the new key.
Commit
baf71c6784a5ad9b4508396d5e467555f2e5e520
by Oliver Smith
OBS: let meta package depend on osmocom-keyring
Let the meta package (osmocom-latest, osmocom-nightly etc.) depend on osmocom-keyring. This means that after the user has enabled any of the official Osmocom OBS repositories and installs a package from there, the osmocom-keyring package will get installed as well, which results in having an up-to-date /usr/share/osmocom-keyring/osmocom.asc.
Commit
8b28d63e980d051e5d39893549fe985218bae28f
by Oliver Smith
jobs/reminder-extend-obs-key: new job
Add a script that downloads the current OBS pubkey, checks its expiration date and fails if it is in less than a year. The idea is to extend it for two years every year (so users have enough time to update their osmocom-keyring package).
Commit
af02d1fd3d6c840df49cc7dce2f8e59cf15c4f58
by Oliver Smith
jobs: make references to master branch consistent
Use "master" instead of: - "refs/remotes/origin/master" - "origin/master" - "*/master"
This improves consistency and makes clear what should be used when adding new jobs. It also fixes the problem with */master that it may actually not only use the master branch, but also e.g. osmith/master which can lead to unintentionally building and publishing artifacts from the wrong branch. If we want to run a job on a different branch intentionally, such a job should have an extra argument for branch that can be changed when starting the job.
Commit
3d1e68dd2303c19130e13a192d6d6bc72a6d4841
by Oliver Smith
repo-install-test: ignore eclipse-titan-optdir
The osmocom:latest repository now has eclipse-titan-optdir-9.0.0 and eclipse-titan-optdir-11.1.0 packages, which allow installing multiple titan versions side-by-side. The related -dbgsym packages can conflict with the main eclipse-titan-dbgsym package. Ignore these packages in the repo-install test.
Fixes: Unpacking eclipse-titan-optdir-9.0.0-dbgsym (9.0.0~osmocom1.2263.c685) ... dpkg: error processing archive /tmp/apt-dpkg-install-4XXdZ1/205-eclipse-titan-optdir-9.0.0-dbgsym_9.0.0~osmocom1.2263.c685_amd64.deb (--unpack): trying to overwrite '/usr/lib/debug/.build-id/0a/072e12f92a7b564adec3a0896c18aa6c942391.debug', which is also in package eclipse-titan-dbgsym 9.0.0.3.0bbfc~osmocom
Commit
1262888d49057d212a16a29506e8a30b79202510
by Oliver Smith
jobs/registry…: fix syntax error
Use $REGISTRY instead of ${REGISTRY}, so jenkins-job-builder 6.3.0 doesn't interpret it as JJB variable. It needs to be interpreted as shell variable here. This probably worked with older JJB versions.
Commit
80a9abae512b4f28c7531c27bb91ea8757bb1e72
by Oliver Smith
contrib/obs-mirror/obs-mirror-include: update
Update the daily backup/mirror configuration.
* Add Debian 13 * Add Raspbian 12 (not sure why it was not there yet) * Add nightly/CentOS_7 (we build a few packages for it and we also have the mirror configured for latest) * Remove CentOS 8 (dropped 1 year ago: https://osmocom.org/news/255) * Remove Debian 10, Raspbian 10 (https://osmocom.org/news/308)
Commit
f071c697887247030b8ec43c1a9dc2d873984d35
by Oliver Smith
OBS: support building debian trixie (13) pkgs
Translate "debian:13" to "debian:trixie", as for some reason no docker tag is getting pushed for "debian:13". With this we can keep using the usual short naming scheme in pipeline_binpkgs lines in gerrit-verifications.yml (see next patch).
Commit
5c0b9984b437759fea55fb9101ccde49acb27d25
by Oliver Smith
jobs/gerrit: binpkgs: update debian versions
Change the default from "debian:10 debian:12" to "debian:11 debian:13", as Debian 13 has just been released and we are about to remove Debian 10 binary packages.
Commit
6d19e82eedcd4852cce3b43482ab54f1df195ef2
by Oliver Smith
OBS: sync_obs_projects: support security.d.o
The Debian 13 meta config in the OpenSUSE OBS now lists security.debian.org. Adjust the script that syncs the config to support this URL too, rewrite it to HTTPS just like the other mirror URL.
Fix for: changing url to https: http://security.debian.org/debian-security?dist=trixie-security&component=main Traceback (most recent call last): File "/obs/sync_obs_projects.py", line 322, in <module> main() File "/obs/sync_obs_projects.py", line 312, in main rewrite_meta(project) File "/obs/sync_obs_projects.py", line 190, in rewrite_meta assert url.startswith("http://ftp.de.debian.org/debian"), \ AssertionError: unexpected mirror URL
Commit
bc5dbb3093b47a5dfdca8f956d85e2cd5a7bcba1
by Oliver Smith
jenkins-gerrit: binpkgs: ignore decode errors
The output of recent osmo-s1gw jobs for building binary packages contains invalid utf-8 characters, even though it should be utf-8 according to the HTTP header: content-type: text/plain;charset=utf-8
We only need to get the distro name from the output here, so simply ignore any decode errors.
Fix for: $ ./comment_generate.py https://jenkins.osmocom.org/jenkins/job/gerrit-osmo-s1gw/417/ -o gerrit_report.json -t result Reading https://jenkins.osmocom.org/jenkins/job/gerrit-osmo-s1gw/417//consoleText Reading https://jenkins.osmocom.org/jenkins/job/gerrit-binpkgs/17255/consoleText Reading https://jenkins.osmocom.org/jenkins/job/gerrit-binpkgs/17260/consoleText Traceback (most recent call last): File "/home/user/code/osmo-dev/src/osmo-ci/scripts/jenkins-gerrit/./comment_generate.py", line 270, in <module> main() File "/home/user/code/osmo-dev/src/osmo-ci/scripts/jenkins-gerrit/./comment_generate.py", line 257, in main comment = get_comment_result(args.build_url, args.notify_on_success) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/user/code/osmo-dev/src/osmo-ci/scripts/jenkins-gerrit/./comment_generate.py", line 198, in get_comment_result pipeline = parse_pipeline(build_url) ^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/user/code/osmo-dev/src/osmo-ci/scripts/jenkins-gerrit/./comment_generate.py", line 89, in parse_pipeline stage = stage_from_job_name(job_name, job_url) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/user/code/osmo-dev/src/osmo-ci/scripts/jenkins-gerrit/./comment_generate.py", line 59, in stage_from_job_name return stage_binpkgs_from_url(job_url) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/user/code/osmo-dev/src/osmo-ci/scripts/jenkins-gerrit/./comment_generate.py", line 46, in stage_binpkgs_from_url content = response.read().decode("utf-8") ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ UnicodeDecodeError: 'utf-8' codec can't decode byte 0xef in position 237446: invalid continuation byte
Commit
cf88fab8cda55362233afbbd54aba0b99a263d44
by Vadim Yanitskiy
jobs/gerrit-verifications: osmo-ttcn3-hacks: do make clean
It's easy to break the build verification [for everybody] by submitting a patch that creates a symlink to a non-existent file. The workspace is not wiped, so the stale symlink remains and results in failures like this:
Commit
bf3ef9d95af939d581f10f8a35e952e1a8131b21
by Oliver Smith
OBS: --version-append: never override version
We have some branches set up that when pushed to, directly result in building OBS packages. We append a string to the package version there, such as ~rhizomatica.testing. The expectation is that all packages built from that branch have this string in the version.
Recently an edge case was found: when amending the commit of the last tagged release, e.g. osmo-sgsn 1.12.0, then the generated version is based on the previous git tag (as the last git tag's commit is not part of the current branch anymore) while having the newer version in debian/changelog. This then results in the newer version being used, without appending the version string:
osmo-sgsn: WARNING: version from changelog (1.12.0) is higher than version based on git tag (1.11.1.20-e000e~rhizomatica.production), using version from changelog (git tag not pushed yet?)
While IMHO it is a bad idea to just amend the commit of the last release instead of making a separate commit, this results in unexpected behavior so adjust the script to not override the version if --version-append is used.
Commit
ffef33c677f7bf9c76a99a6dfa7ab7eb29a49c23
by Oliver Smith
jobs: run ttcn3-remsim-test-asan with bookworm
The default distribution to run ttcn3 testsuites and SUTs with has recently been changed from debian 12 (bookworm) to 13 (trixie). Since then we have seen 27 additional failures with osmocom:nightly:asan (while osmocom:nightly and osmocom:latest work fine). As the reason for this is currently unclear, revert to using debian 12 for this job for now, so this doesn't mask other errors that the job would find otherwise.
Commit
60bf602f6f8b0f4d06329b14767c4c49f8d42ded
by Oliver Smith
OBS: disable_manuals: adjust for osmocom-bb
With the related change, we build manauals in the debian package of osmocom-bb. The osmocom:nightly:asan feed gets built with manuals disabled. Fix that building the osmocom-bb package in the asan feed currently fails with:
[ 160s] /usr/bin/make -C doc/manuals [ 160s] make[2]: Entering directory '/usr/src/packages/BUILD/doc/manuals' [ 160s] OSMO_GSM_MANUALS_DIR: /usr/src/packages/osmo-gsm-manuals (fallback) [ 160s] ERROR: OSMO_GSM_MANUALS_DIR does not exist! [ 160s] Install osmo-gsm-manuals or set OSMO_GSM_MANUALS_DIR. [ 160s] Makefile:6: /build/Makefile.asciidoc.inc: No such file or directory [ 160s] make[2]: *** No rule to make target '/build/Makefile.asciidoc.inc'. Stop.
Commit
bd2cf1f881aea5dd5ca4a0f8822c5467762d1166
by Oliver Smith
OBS: stop building limesuite
Building limesuite is not needed anymore, as it is available at least since debian 10 in the debian and ubuntu repositories. We have been building the upstream version without changes.
Commit
77eaf6179803cf8c9bd71df39beb375e46205a52
by Oliver Smith
testenv-coredump-helper: remove
After looking at this again, it makes more sense to keep /proc/sys/kernel/core_pattern as "core" on build4 and build5, which we use for jenkins jobs instead of installing systemd's coredumpctl and making all coredumps go through that. The motivation for using coredumpctl in testenv was that lots of developers may have it already installed (though I'm not sure about that anymore), and that it gives a nice API for retrieving the related executable name to a coredump instead of having to parse the output of e.g. "file".
Using coredumpctl had the disadvantage that coredumps for all other jobs, such as the master-builds are also caught by that and cannot be easily placed in the workspace directory. I have started implementing this by extending testenv-coredump-helper to communicate through a socket and mounting that socket inside docker containers running contrib/jenkins.sh (host -> lxc -> docker), and it works, but this complexity is not useful here.
Instead the related patch will make testenv pick the core files when coredumpctl is not available, and read the path from "file", making testenv-coredump-helper obsolete.
Commit
dfa9fcca49ae8b8ce6c691196c3883206933e503
by Oliver Smith
scripts/wrapper_core_bt_on_error: new script
Add a wrapper script that runs a given program, and checks the exit code. If it is not 0, and a core file appears, then load the coredump in gdb and display its backtrace.
![[Jenkins]](/jenkins/static/166fa6d8/images/svgs/logo.svg){kind=logo}
