osmo-gsm-tester/doc/manuals/chapters/install.adoc

650 lines
23 KiB
Plaintext

== Installation
=== Trial Builder
The Trial Builder is the jenkins build slave (host) building all sysroot binary
packages used later by {app-name} to run the tests. It's purpose is to build the
sysroots and provide them to {app-anme}, for instance, as jenkins job artifacts
which the {app-name} runner job can fetch.
[[jenkins_deps]]
==== Osmocom Build Dependencies
Each of the jenkins builds requires individual dependencies. This is generally
the same as for building the software outside of osmo-gsm-tester and will not
be detailed here. For the Osmocom projects, refer to
http://osmocom.org/projects/cellular-infrastructure/wiki/Build_from_Source . Be
aware of specific requirements for BTS hardware: for example, the
osmo-bts-sysmo build needs the sysmoBTS SDK installed on the build slave, which
should match the installed sysmoBTS firmware.
==== Add Build Jobs
There are various jenkins-build-* scripts in osmo-gsm-tester/contrib/, which
can be called as jenkins build jobs to build and bundle binaries as artifacts,
to be run on the osmo-gsm-tester main unit and/or BTS hardware.
Be aware of the dependencies, as hinted at in <<jenkins_deps>>.
While the various binaries could technically be built on the osmo-gsm-tester
main unit, it is recommended to use a separate build slave, to take load off
of the main unit.
Please note nowadays we set up all the osmocom jenkins jobs (including
{app-name} ones) using 'jenkins-job-builder'. You can find all the
configuration's in Osmocom's 'osmo-ci.git' files 'jobs/osmo-gsm-tester-*.yml.
Explanation below on how to set up jobs manually is left as a reference for
other projects.
On your jenkins master, set up build jobs to call these scripts -- typically
one build job per script. Look in contrib/ and create one build job for each of
the BTS types you would like to test, as well as one for the 'build-osmo-nitb'.
These are generic steps to configure a jenkins build
job for each of these build scripts, by example of the
jenkins-build-osmo-nitb.sh script; all that differs to the other scripts is the
"osmo-nitb" part:
* 'Project name': "osmo-gsm-tester_build-osmo-nitb" +
(Replace 'osmo-nitb' according to which build script this is for)
* 'Discard old builds' +
Configure this to taste, for example:
** 'Max # of build to keep': "20"
* 'Restrict where this project can be run': Choose a build slave label that
matches the main unit's architecture and distribution, typically a Debian
system, e.g.: "linux_amd64_debian8"
* 'Source Code Management':
** 'Git'
*** 'Repository URL': "git://git.osmocom.org/osmo-gsm-tester"
*** 'Branch Specifier': "*/master"
*** 'Additional Behaviors'
**** 'Check out to a sub-directory': "osmo-gsm-tester"
* 'Build Triggers' +
The decision on when to build is complex. Here are some examples:
** Once per day: +
'Build periodically': "H H * * *"
** For the Osmocom project, the purpose is to verify our software changes.
Hence we would like to test every time our code has changed:
*** We could add various git repositories to watch, and enable 'Poll SCM'.
*** On jenkins.osmocom.org, we have various jobs that build the master branches
of their respective git repositories when a new change was merged. Here, we
can thus trigger e.g. an osmo-nitb build for osmo-gsm-tester everytime the
master build has run: +
'Build after other projects are built': "OpenBSC"
*** Note that most of the Osmocom projects also need to be re-tested when their
dependencies like libosmo* have changed. Triggering on all those changes
typically causes more jenkins runs than necessary: for example, it rebuilds
once per each dependency that has rebuilt due to one libosmocore change.
There is so far no trivial way known to avoid this. It is indeed safest to
rebuild more often.
* 'Build'
** 'Execute Shell'
+
----
#!/bin/sh
set -e -x
./osmo-gsm-tester/contrib/jenkins-build-osmo-nitb.sh
----
+
(Replace 'osmo-nitb' according to which build script this is for)
* 'Post-build Actions'
** 'Archive the artifacts': "*.tgz, *.md5" +
(This step is important to be able to use the built binaries in the run job
below.)
TIP: When you've created one build job, it is convenient to create further
build jobs by copying the first one and, e.g., simply replacing all "osmo-nitb"
with "osmo-bts-trx".
[[install_main_unit]]
=== Main Unit
The main unit is a general purpose computer that orchestrates the tests. It
runs the core network components, controls the modems and so on. This can be
anything from a dedicated production rack unit to your laptop at home.
This manual will assume that tests are run from a jenkins build slave, by a user
named 'jenkins' that belongs to group 'osmo-gsm-tester'. The user configuration
for manual test runs and/or a different user name is identical, simply replace
the user name or group.
Please, note installation steps and dependencies needed will depend on lots of
factors, like your distribution, your specific setup, which hardware you plan to
support, etc.
This section aims at being one place to document the rationale behind certain
configurations being done in one way or another. For an up to date step by step
detailed way to install and maintain the Osmocom {app-name} setup, one will want
to look at the <<ansible,ansible scripts section>>.
[[configure_jenkins_slave]]
==== Create 'jenkins' User
On the main unit, create a jenkins user:
----
useradd -m jenkins
----
==== Install Java on Main Unit
To be able to launch the Jenkins build slave, a Java RE must be available on
the main unit. For example:
----
apt-get install default-jdk
----
==== Allow SSH Access from Jenkins Master
Create an SSH keypair to be used for login on the osmo-gsm-tester. This may be
entered on the jenkins web UI; alternatively, use the jenkins server's shell:
Login on the main jenkins server shell and create an SSH keypair, for example:
----
# su jenkins
$ mkdir -p /usr/local/jenkins/keys
$ ssh-keygen
Generating public/private rsa key pair.
Enter file in which to save the key (/home/jenkins/.ssh/id_rsa): /usr/local/jenkins/keys/osmo-gsm-tester-rnd
Enter passphrase (empty for no passphrase): <enter a passphrase>
Enter same passphrase again: <enter a passphrase>
Your identification has been saved in /usr/local/jenkins/keys/osmo-gsm-tester-rnd
Your public key has been saved in /usr/local/jenkins/keys/osmo-gsm-tester-rnd.pub.
The key fingerprint is:
...
----
Copy the public key to the main unit, e.g. copy-paste:
----
cat /usr/local/jenkins/keys/osmo-gsm-tester-rnd.pub
# copy this public key
----
On the main unit:
----
mkdir ~jenkins/.ssh
cat > ~jenkins/.ssh/authorized_keys
# paste above public key and hit Ctrl-D
chown -R jenkins: ~jenkins/.ssh
----
Make sure that the user running the jenkins master accepts the main unit's host
identification. There must be an actual RSA host key available in the
known_hosts file for the jenkins master to be able to log in. Simply calling
ssh and accepting the host key as usual is not enough. Jenkins may continue to
say "Host key verification failed".
To place an RSA host key in the jenkins' known_hosts file, you may do:
On the Jenkins master:
----
main_unit_ip=10.9.8.7
ssh-keyscan -H $main_unit_ip >> ~jenkins/.ssh/known_hosts
chown jenkins: ~jenkins/.ssh/known_hosts
----
Verify that the jenkins user on the Jenkins master has SSH access to the main
unit:
----
su jenkins
main_unit_ip=10.9.8.7
ssh -i /usr/local/jenkins/keys/osmo-gsm-tester-rnd jenkins@$main_unit_ip
exit
----
[[install_add_jenkins_slave]]
==== Add Jenkins Slave
In the jenkins web UI, add a new build slave for the osmo-gsm-tester:
* 'Manage Jenkins'
** 'Manage Nodes'
*** 'New Node'
**** Enter a node name, e.g. "osmo-gsm-tester-1" +
(the "-1" is just some identification in case you'd like to add another
setup later).
**** 'Permanent Agent'
Configure the node as:
* '# of executors': 1
* 'Remote root directory': "/home/jenkins"
* 'Labels': "osmo-gsm-tester" +
(This is a general label common to all osmo-gsm-tester build slaves you may set up in the future.)
* 'Usage': 'Only build jobs with label expressions matching this node'
* 'Launch method': 'Launch slave agents via SSH'
** 'Host': your main unit's IP address
** 'Credentials': choose 'Add' / 'Jenkins'
*** 'Domain': 'Global credentials (unrestricted)'
*** 'Kind': 'SSH Username with private key'
*** 'Scope': 'Global'
*** 'Username': "jenkins" +
(as created on the main unit above)
*** 'Private Key': 'From a file on Jenkins master'
**** 'File': "/usr/local/jenkins/keys/osmo-gsm-tester-rnd"
*** 'Passphrase': enter same passphrase as above
*** 'ID': "osmo-gsm-tester-1"
*** 'Name': "jenkins for SSH to osmo-gsm-tester-1"
The build slave should be able to start now.
==== Add Run Job
This is the jenkins job that runs the tests on the GSM hardware:
* It sources the artifacts from jenkins' build jobs.
* It runs on the osmo-gsm-tester main unit.
Sample script to run {app-name} as a jenkins job can be found in
'osmo-gsm-tester.git' file 'contrib/jenkins-run.sh'.
Please note nowadays we set up all the osmocom jenkins jobs (including
{app-name} ones) using 'jenkins-job-builder'. You can find all the
configuration's in Osmocom's 'osmo-ci.git' files 'jobs/osmo-gsm-tester-*.yml.
Explanation below on how to set up jobs manually is left as a reference for
other projects.
Here is the configuration for the run job:
* 'Project name': "osmo-gsm-tester_run"
* 'Discard old builds' +
Configure this to taste, for example:
** 'Max # of build to keep': "20"
* 'Restrict where this project can be run': "osmo-gsm-tester" +
(to match the 'Label' configured in <<install_add_jenkins_slave>>).
* 'Source Code Management':
** 'Git'
*** 'Repository URL': "git://git.osmocom.org/osmo-gsm-tester"
*** 'Branch Specifier': "*/master"
*** 'Additional Behaviors'
**** 'Check out to a sub-directory': "osmo-gsm-tester"
**** 'Clean before checkout'
* 'Build Triggers' +
The decision on when to build is complex. For this run job, it is suggested
to rebuild:
** after each of above build jobs that produced new artifacts: +
'Build after other projects are built': "osmo-gsm-tester_build-osmo-nitb,
osmo-gsm-tester_build-osmo-bts-sysmo, osmo-gsm-tester_build-osmo-bts-trx" +
(Add each build job name you configured above)
** as well as once per day: +
'Build periodically': "H H * * *"
** and, in addition, whenever the osmo-gsm-tester scripts have been modified: +
'Poll SCM': "H/5 * * * *" +
(i.e. look every five minutes whether the upstream git has changed)
* 'Build'
** Copy artifacts from each build job you have set up:
*** 'Copy artifacts from another project'
**** 'Project name': "osmo-gsm-tester_build-osmo-nitb"
**** 'Which build': 'Latest successful build'
**** enable 'Stable build only'
**** 'Artifacts to copy': "*.tgz, *.md5"
*** Add a separate similar 'Copy artifacts...' section for each build job you
have set up.
** 'Execute Shell'
+
----
#!/bin/sh
set -e -x
# debug: provoke a failure
#export OSMO_GSM_TESTER_OPTS="-s debug -t fail"
PATH="$PWD/osmo-gsm-tester/src:$PATH" \
./osmo-gsm-tester/contrib/jenkins-run.sh
----
+
Details:
*** The 'jenkins-run.sh' script assumes to find the 'osmo-gsm-tester.py' in the
'$PATH'. To use the most recent osmo-gsm-tester code here, we direct
'$PATH' to the actual workspace checkout. This could also run from a sytem
wide install, in which case you could omit the explicit PATH to
"$PWD/osmo-gsm-tester/src".
*** This assumes that there are configuration files for osmo-gsm-tester placed
on the system (see <<config_paths>>).
*** If you'd like to check the behavior of test failures, you can uncomment the
line below "# debug" to produce a build failure on every run. Note that
this test typically produces a quite empty run result, since it launches no
NITB nor BTS.
* 'Post-build Actions'
** 'Archive the artifacts'
*** 'Files to archive': "*-run.tgz, *-bin.tgz" +
This stores the complete test report with config files, logs, stdout/stderr
output, pcaps as well as the binaries used for the test run in artifacts.
This allows analysis of older builds, instead of only the most recent build
(which cleans up the jenkins workspace every time). The 'trial-N-run.tgz'
and 'trial-N-bin.tgz' archives are produced by the 'jenkins-run.sh' script,
both for successful and failing runs.
==== Install osmo-gsm-tester
This assumes you have already created the jenkins user (see <<configure_jenkins_slave>>).
Dependencies needed will depend on lots of factors, like your distribution, your
specific setup, which hardware you plan to support, etc.
On a Debian/Ubuntu based system, these commands install the packages needed to
run the osmo-gsm-tester.py code, i.e. install these on your main unit:
----
apt-get install \
dbus \
tcpdump \
sqlite3 \
python3 \
python3-setuptools \
python3-yaml \
python3-mako \
python3-gi \
python3-numpy \
python3-wheel \
ofono \
patchelf \
sudo \
libcap2-bin \
python3-pip \
udhcpc \
iperf3 \
locales
pip3 install \
"git+https://github.com/podshumok/python-smpplib.git@master#egg=smpplib" \
pydbus \
pyusb \
pysispm
----
IMPORTANT: ofono may need to be installed from source to contain the most
recent fixes needed to operate your modems. This depends on the modem hardware
used and the tests run. Please see <<hardware_modems>>.
==== User Permissions
On the main unit, create a group for all users that should be allowed to use
the osmo-gsm-tester, and add users (here 'jenkins') to this group.
----
groupadd osmo-gsm-tester
gpasswd -a jenkins osmo-gsm-tester
----
NOTE: you may also need to add users to the 'usrp' group, see
<<user_config_uhd>>.
A user added to a group needs to re-login for the group permissions to take
effect.
===== Paths
Assuming that you are using the example config, prepare a system wide state
location in '/var/tmp':
----
mkdir -p /var/tmp/osmo-gsm-tester/state
chown -R :osmo-gsm-tester /var/tmp/osmo-gsm-tester
chmod -R g+rwxs /var/tmp/osmo-gsm-tester
setfacl -d -m group:osmo-gsm-tester:rwx /var/tmp/osmo-gsm-tester/state
----
IMPORTANT: the state directory needs to be shared between all users potentially
running the osmo-gsm-tester to resolve resource allocations. Above 'setfacl'
command sets the access control to keep all created files group writable.
With the jenkins build as described here, the trials will live in the build
slave's workspace. Other modes of operation (a daemon scheduling concurrent
runs, *TODO*) may use a system wide directory to manage trials to run:
----
mkdir -p /var/tmp/osmo-gsm-tester/trials
chown -R :osmo-gsm-tester /var/tmp/osmo-gsm-tester
chmod -R g+rwxs /var/tmp/osmo-gsm-tester
----
===== Allow DBus Access to ofono
Put a DBus configuration file in place that allows the 'osmo-gsm-tester' group
to access the org.ofono DBus path:
----
# cat > /etc/dbus-1/system.d/osmo-gsm-tester.conf <<END
<!-- Additional rules for the osmo-gsm-tester to access org.ofono from user
land -->
<!DOCTYPE busconfig PUBLIC "-//freedesktop//DTD D-BUS Bus Configuration 1.0//EN"
"http://www.freedesktop.org/standards/dbus/1.0/busconfig.dtd">
<busconfig>
<policy group="osmo-gsm-tester">
<allow send_destination="org.ofono"/>
</policy>
</busconfig>
END
----
(No restart of dbus nor ofono necessary.)
[[install_slave_unit]]
=== Slave Unit(s)
The slave units are the hosts used by {app-name} to run proceses on. It may be
the <<install_main_unit,Main Unit>> itself and processes will be run locally, or
it may be a remote host were processes are run usually through SSH.
This guide assumes slaves unit(s) use same configuration as the Main Unit, that
is, it runs under 'jenkins' user which is a member of the 'osmo-gsm-tester' user
group. In order to do so, follow the instruction under the
<<install_main_unit,Main Unit>> section above. Keep in mind the 'jenkins' user
on the Main Unit will need to be able to log in through SSH as the slave unit
'jenkins' user to run the processes. No direct access from Jenkins Master node
is required here.
[[install_capture_packets]]
==== Capture Packets
In order to allow collecting pcap traces of the network communication for later
reference, allow the osmo-gsm-tester group to capture packets using the 'tcpdump'
program:
----
chgrp osmo-gsm-tester /usr/sbin/tcpdump
chmod 750 /usr/sbin/tcpdump
setcap cap_net_raw,cap_net_admin=eip /usr/sbin/tcpdump
----
Put 'tcpdump' in the '$PATH' -- assuming that 'tcpdump' is available for root:
----
ln -s `which tcpdump` /usr/local/bin/tcpdump
----
TIP: Why a symlink in '/usr/local/bin'? On Debian, 'tcpdump' lives in
'/usr/sbin', which is not part of the '$PATH' for non-root users. To avoid
hardcoding non-portable paths in the osmo-gsm-tester source, 'tcpdump' must be
available in the '$PATH'. There are various trivial ways to modify '$PATH' for
login shells, but the jenkins build slave typically runs in a *non-login*
shell; modifying non-login shell enviroments is not trivially possible without
also interfering with files installed from debian packages. Probably the
easiest way to allow all users and all shells to find the 'tcpdump' binary is
to actually place a symbolic link in a directory that is already part of the
non-login shell's '$PATH'. Above example places such in '/usr/local/bin'.
Verify that a non-login shell can find 'tcpdump':
----
su jenkins -c 'which tcpdump'
# should print: "/usr/local/bin/tcpdump"
----
WARNING: When logged in via SSH on your main unit, running 'tcpdump' to capture
packets may result in a feedback loop: SSH activity to send tcpdump's output to
your terminal is in turn is picked up in the tcpdump trace, and so forth. When
testing 'tcpdump' access, make sure to have proper filter expressions in place.
==== Allow Core Files
In case a binary run for the test crashes, a core file of the crash should be
written. This requires a limit rule. Create a file with the required rule:
----
sudo -s
echo "@osmo-gsm-tester - core unlimited" > /etc/security/limits.d/osmo-gsm-tester_allow-core.conf
----
Re-login the user to make these changes take effect.
Set the *kernel.core_pattern* sysctl to *core* (usually the default). For each
binary run by osmo-gsm-tester, a core file will then appear in the same dir that
contains stdout and stderr for that process (because this dir is set as CWD).
----
sysctl -w kernel.core_pattern=core
----
TIP: Files required to be installed under '/etc/security/limits.d/' can be found
under 'osmo-gsm-tester.git/utils/limits.d/', so one can simply cp them from
there.
==== Allow Realtime Priority
Certain binaries should be run with real-time priority, like 'osmo-bts-trx'.
Add this permission on the main unit:
----
sudo -s
echo "@osmo-gsm-tester - rtprio 99" > /etc/security/limits.d/osmo-gsm-tester_allow-rtprio.conf
----
Re-login the user to make these changes take effect.
TIP: Files required to be installed under '/etc/security/limits.d/' can be found
under 'osmo-gsm-tester.git/utils/limits.d/', so one can simply cp them from
there.
===== Allow capabilities: 'CAP_NET_RAW', 'CAP_NET_ADMIN', 'CAP_SYS_ADMIN'
Certain binaries require 'CAP_NET_RAW' to be set, like 'osmo-bts-octphy' as it
uses a 'AF_PACKET' socket. Similarly, others (like osmo-ggsn) require
'CAP_NET_ADMIN' to be able to create tun devices, and so on.
To be able to set the following capability without being root, osmo-gsm-tester
uses sudo to gain permissions to set the capability.
This is the script that osmo-gsm-tester expects on the host running the process:
----
echo /usr/local/bin/osmo-gsm-tester_setcap_net_raw.sh <<EOF
#!/bin/bash
/sbin/setcap cap_net_raw+ep $1
EOF
chmod +x /usr/local/bin/osmo-gsm-tester_setcap_net_raw.sh
----
Now, again on the same host, we need to provide sudo access to this script for
osmo-gsm-tester:
----
echo "%osmo-gsm-tester ALL=(root) NOPASSWD: /usr/local/bin/osmo-gsm-tester_setcap_net_raw.sh" > /etc/sudoers.d/osmo-gsm-tester_setcap_net_raw
chmod 0440 /etc/sudoers.d/osmo-gsm-tester_setcap_net_raw
----
The script file name 'osmo-gsm-tester_setcap_net_raw.sh' is important, as
osmo-gsm-tester expects to find a script with this name in '$PATH' at run time.
TIP: Files required to be installed under '/etc/sudoers.d/' can be found
under 'osmo-gsm-tester.git/utils/sudoers.d/', so one can simply cp them from
there.
TIP: Files required to be installed under '/usr/local/bin/' can be found
under 'osmo-gsm-tester.git/utils/bin/', so one can simply cp them from
there.
[[user_config_uhd]]
==== UHD
Grant permission to use the UHD driver to run USRP devices for osmo-bts-trx, by
adding the jenkins user to the 'usrp' group:
----
gpasswd -a jenkins usrp
----
To run osmo-bts-trx with a USRP attached, you may need to install a UHD driver.
Please refer to http://osmocom.org/projects/osmotrx/wiki/OsmoTRX#UHD for
details; the following is an example for the B200 family USRP devices:
----
apt-get install libuhd-dev uhd-host
/usr/lib/uhd/utils/uhd_images_downloader.py
----
==== Log Rotation
To avoid clogging up /var/log, it makes sense to choose a sane maximum log size:
----
echo maxsize 10M > /etc/logrotate.d/maxsize
----
==== Install Scripts
IMPORTANT: When using the jenkins build slave as configured above, *there is no
need to install the osmo-gsm-tester sources on the main unit*. The jenkins job
will do so implicitly by checking out the latest osmo-gsm-tester sources in the
workspace for every run. If you're using only the jenkins build slave, you may
skip this section.
If you prefer to use a fixed installation of the osmo-gsm-tester sources
instead of the jenkins workspace, you can:
. From the run job configured above, remove the line that says
+
----
PATH="$PWD/osmo-gsm-tester/src:$PATH" \
----
+
so that this uses a system wide installation instead.
. Install the sources e.g. in '/usr/local/src' as indicated below.
On the main unit, to install the latest in '/usr/local/src':
----
apt-get install git
mkdir -p /usr/local/src
cd /usr/local/src
git clone git://git.osmocom.org/osmo-gsm-tester
----
To allow all users to run 'osmo-gsm-tester.py', from login as well as non-login
shells, the easiest solution is to place a symlink in '/usr/local/bin':
----
ln -s /usr/local/src/osmo-gsm-tester/src/osmo-gsm-tester.py /usr/local/bin/
----
(See also the tip in <<install_capture_packets>> for a more detailed
explanation.)
The example configuration provided in the source is suitable for running as-is,
*if* your hardware setup matches (you could technically use that directly by a
symlink e.g. from '/usr/local/etc/osmo-gsm-tester' to the 'example' dir). If in
doubt, rather copy the example, point 'paths.conf' at the 'suites' dir, and
adjust your own configuration as needed. For example:
----
cd /etc
cp -R /usr/local/src/osmo-gsm-tester/example osmo-gsm-tester
sed -i 's#\./suites#/usr/local/src/osmo-gsm-tester/suites#' osmo-gsm-tester/paths.conf
----
NOTE: The configuration will be looked up in various places, see
<<config_paths>>.