The Ceph project needs automated tests. Because Ceph is a highly distributed system, and has active kernel development, its testing requirements are quite different from e.g. typical LAMP web applications. Nothing out there seemed to handle our requirements, so we wrote our own framework, called Teuthology.
Teuthology runs a given set of Python functions (tasks), with an SSH connection to every host participating in the test. The SSH connection uses Paramiko, a native Python client for the SSH2 protocol, and this allows us to e.g. run multiple commands inside a single SSH connection, to speed up test execution. Tests can use gevent to perform actions concurrently or in the background.
Teuthology is not meant to be distributed as a library, therefore we depend
on the pinned dependencies listed in requirements.txt, the setup.py
will not list any and will only be there to install the package entry points
(a.k.a teuthology's scripts).
git clone https://github.com/ceph/teuthology.git cd teuthology
A boostrap script is provided for automated builds/execution of teuthology
itself. You can run it directly only if you are using Ubuntu.
Teuthology uses several Python packages that are not in the standard
library. To make the dependencies easier to get right, we use a
virtualenv to manage them. To get started, ensure you have the
virtualenv and pip programs installed; e.g. on Debian/Ubuntu:
sudo apt-get install python-dev python-virtualenv python-pip libevent-dev libmysqlclient-dev python-libvirt
and then run:
./bootstrap
Note
These instructions assume you are using homebrew
As always, create a virtualenv specific to teuthology and make sure it
is activated before proceeding (location doesn't matter, we use an example
location):
mkdir ~/.virtualenvs virtualenv --system-site-packages ~/.virtualenvs/teuthology source ~/.virtualenvs/teuthology/bin/activate
Install the system dependencies:
brew install libvirt mysql libevent
Make sure you are able to import libvirt without error:
python -c "import libvirt"
If python can't find libvirt yet, you may need to do the following:
cd /Library/Python/{pyversion}/site-packages sudo ln -s /usr/local/Cellar/libvirt/{version}/lib/python{pyversion}/site-packages/* .
Finally, install the teuthology package and requirements.txt:
python setup.py develop pip install -r requirements.txt
These instructions should help get teuthology installed properly in a system that is not OSX or Debian-based.
Install all the system dependencies needed:
- mysql client
- libevent
- libvirt (with the Python bindings)
Install Python packaging tools:
- pip
- virtualenv
In some cases, depending on the OS, you will need a python development package
with some build helpers that are required to build packages. In Ubuntu, this is
the python-dev package.
With a dedicated virtualenv activated, install the teuthology package and
requirements.txt:
python setup.py develop pip install -r requirements.txt
An integration test run takes three items of configuration:
targets: what hosts to run on; this is a dictionary mapping hosts to ssh host keys, like: "[email protected]: ssh-rsa long_hostkey_here"roles: how to use the hosts; this is a list of lists, where each entry lists all the roles to be run on a single host; for example, a single entry might say[mon.1, osd.1]tasks: how to set up the cluster and what tests to run on it; see below for examples
The format for this configuration is YAML, a structured data format that is still human-readable and editable.
For example, a full config for a test run that sets up a three-machine
cluster, mounts Ceph via ceph-fuse, and leaves you at an interactive
Python prompt for manual exploration (and enabling you to SSH in to
the nodes & use the live cluster ad hoc), might look like this:
roles:
- [mon.0, mds.0, osd.0]
- [mon.1, osd.1]
- [mon.2, client.0]
targets:
[email protected]: ssh-rsa host07_ssh_key
[email protected]: ssh-rsa host08_ssh_key
[email protected]: ssh-rsa host09_ssh_key
tasks:
- install:
- ceph:
- ceph-fuse: [client.0]
- interactive:
The number of entries under roles and targets must match.
Note the colon after every task name in the tasks section.
The install task needs to precede all other tasks.
The listed targets need resolvable hostnames. If you do not have a DNS server
running, you can add entries to /etc/hosts. You also need to be able to SSH
in to the listed targets without passphrases, and the remote user needs to have
passwordless sudo access. Note that the ssh keys at the end of the
targets entries are the public ssh keys for the hosts. On Ubuntu, these
are located at /etc/ssh/ssh_host_rsa_key.pub
If you'd save the above file as example.yaml, you could run
teuthology on it by saying:
./virtualenv/bin/teuthology example.yaml
You can also pass the -v option, for more verbose execution. See
teuthology --help for more.
You can pass multiple files as arguments to teuthology. Each one
will be read as a config file, and their contents will be merged. This
allows you to e.g. share definitions of what a "simple 3 node cluster"
is. The source tree comes with roles/3-simple.yaml, so we could
skip the roles section in the above example.yaml and then
run:
./virtualenv/bin/teuthology roles/3-simple.yaml example.yaml
Before locking machines will work, you must create a .teuthology.yaml file in your home directory that sets a lock_server, i.e.:
lock_server: http://host.example.com:8080/lock
Teuthology automatically locks nodes for you if you specify the
--lock option. Without this option, you must specify machines to
run on in a targets.yaml file, and lock them using
teuthology-lock.
Note that the default owner of a machine is USER@HOST.
You can override this with the --owner option when running
teuthology or teuthology-lock.
With teuthology-lock, you can also add a description, so you can
remember which tests you were running on them. This can be done when
locking or unlocking machines, or as a separate action with the
--update option. To lock 3 machines and set a description, run:
./virtualenv/bin/teuthology-lock --lock-many 3 --desc 'test foo'
If machines become unusable for some reason, you can mark them down:
./virtualenv/bin/teuthology-lock --update --status down machine1 machine2
To see the status of all machines, use the --list option. This can
be restricted to particular machines as well:
./virtualenv/bin/teuthology-lock --list machine1 machine2
A task is a Python module in the teuthology.task package, with a
callable named task. It gets the following arguments:
ctx: a context that is available through the lifetime of the test run, and has useful attributes such ascluster, letting the task access the remote hosts. Tasks can also store their internal state here. (TODO beware namespace collisions.)config: the data structure after the colon in the config file, e.g. for the aboveceph-fuseexample, it would be a list like["client.0"].
Tasks can be simple functions, called once in the order they are
listed in tasks. But sometimes, it makes sense for a task to be
able to clean up after itself; for example, unmounting the filesystem
after a test run. A task callable that returns a Python context
manager
will have the manager added to a stack, and the stack will be unwound
at the end of the run. This means the cleanup actions are run in
reverse order, both on success and failure. A nice way of writing
context managers is the contextlib.contextmanager decorator; look
for that string in the existing tasks to see examples, and note where
they use yield.
Further details on some of the more complex tasks such as install or workunit can be obtained via python help. For example:
>>> import teuthology.task.workunit >>> help(teuthology.task.workunit)
displays a page of more documentation and more concrete examples.
Some of the more important / commonly used tasks include:
chef: Run the chef task.install: by default, the install task goes to gitbuilder and installs the results of the latest build. You can, however, add additional parameters to the test configuration to cause it to install any branch, SHA, archive or URL. The following are valid parameters.
branch: specify a branch (bobtail, cuttlefish...)flavor: specify a flavor (next, unstable...). Flavors can be thought of as subsets of branches. Sometimes (unstable, for example) they may have a predefined meaning.project: specify a project (ceph, samba...)sha1: install the build with this sha1 value.tag: specify a tag/identifying text for this build (v47.2, v48.1...)
ceph: Bring up Cephoverrides: override behavior. Typically, this includes sub-tasks being overridden. Sub-tasks can nest further information. For example, overrides of install tasks are project specific, so the following section of a yaml file would cause all ceph installation to default into using the cuttlefish branch:overrides: install: ceph: branch: cuttlefishworkunit: workunits are a way of grouping tasks and behavior on targets.sequential: group the sub-tasks into a unit where the sub-tasks run sequentially as listed.parallel: group the sub-tasks into a unit where the sub-task all run in parallel.
Sequential and parallel tasks can be nested. Tasks run sequentially if not specified.
The above list is a very incomplete description of the tasks available on teuthology. The teuthology/task subdirectory contains all the python files that implement tasks. Many of these tasks are used to run shell scripts that are defined in the ceph/ceph-qa-suite.
If machines were locked as part of the run (with the --lock switch),
teuthology normally leaves them locked when there is any task failure
for investigation of the machine state. When developing new teuthology
tasks, sometimes this behavior is not useful. The unlock_on_failure
global option can be set to True to make the unlocking happen unconditionally.
Sometimes when a bug triggers, instead of automatic cleanup, you want to explore the system as is. Adding a top-level:
interactive-on-error: true
as a config file for teuthology will make that possible. With that
option, any task that fails, will have the interactive task
called after it. This means that before any cleanup happens, you get a
chance to inspect the system -- both through Teuthology and via extra
SSH connections -- and the cleanup completes only when you choose so.
Just exit the interactive Python session to continue the cleanup.
Note that this only catches exceptions between the tasks. If a task
calls multiple subtasks, e.g. with contextutil.nested, those
cleanups will be performed. Later on, we can let tasks communicate
the subtasks they wish to invoke to the top-level runner, avoiding
this issue.
The interactive task presents a prompt for you to interact with the
teuthology configuration. The ctx variable is available to explore,
and a pprint.PrettyPrinter().pprint object is added for convenience as
'pp', so you can do things like pp(dict-of-interest) to see a formatted
view of the dict.
Teuthology currently places most test files and mount points in a
sandbox directory, defaulting to /home/$USER/cephtest. To change
the location of the sandbox directory, the following option can be
specified in $HOME/.teuthology.yaml:
test_path: <directory>
Teuthology also supports virtual machines, which can function like physical machines but differ in the following ways:
A new entry, vpshost, has been added to the teuthology database of available machines. For physical machines, this value is null. For virtual machines, this entry is the name of the physical machine that that virtual machine resides on.
There are fixed "slots" for virtual machines that appear in the teuthology database. These slots have a machine type of vps and can be locked like any other machine. The existence of a vpshost field is how teuthology knows whether or not a database entry represents a physical or a virtual machine.
The following needs to be set in ~/.config/libvirt/libvirt.conf in order to get the right virtual machine associations for the Inktank lab:
uri_aliases = [
'mira001=qemu+ssh://[email protected]/system?no_tty=1',
'mira003=qemu+ssh://[email protected]/system?no_tty=1',
'mira004=qemu+ssh://[email protected]/system?no_tty=1',
'mira006=qemu+ssh://[email protected]/system?no_tty=1',
'mira007=qemu+ssh://[email protected]/system?no_tty=1',
'mira008=qemu+ssh://[email protected]/system?no_tty=1',
'mira009=qemu+ssh://[email protected]/system?no_tty=1',
'mira010=qemu+ssh://[email protected]/system?no_tty=1',
'mira011=qemu+ssh://[email protected]/system?no_tty=1',
'mira013=qemu+ssh://[email protected]/system?no_tty=1',
'mira014=qemu+ssh://[email protected]/system?no_tty=1',
'mira015=qemu+ssh://[email protected]/system?no_tty=1',
'mira017=qemu+ssh://[email protected]/system?no_tty=1',
'mira018=qemu+ssh://[email protected]/system?no_tty=1',
'mira020=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi01=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi02=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi03=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi04=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi05=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi06=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi07=qemu+ssh://[email protected]/system?no_tty=1',
'vercoi08=qemu+ssh://[email protected]/system?no_tty=1',
'senta01=qemu+ssh://[email protected]/system?no_tty=1',
'senta02=qemu+ssh://[email protected]/system?no_tty=1',
'senta03=qemu+ssh://[email protected]/system?no_tty=1',
'senta04=qemu+ssh://[email protected]/system?no_tty=1',
]
When a virtual machine is locked, downburst is run on that machine to install a
new image. This allows the user to set different virtual OSes to be installed
on the newly created virtual machine. Currently the default virtual machine is
ubuntu (precise). A different vm installation can be set using the
--os-type option in teuthology.lock.
When a virtual machine is unlocked, downburst destroys the image on the machine.
Temporary yaml files are used to downburst a virtual machine. A typical yaml file will look like this:
downburst:
cpus: 1
disk-size: 30G
distro: centos
networks:
- {source: front}
ram: 4G
These values are used by downburst to create the virtual machine.
When locking a file, a downburst meta-data yaml file can be specified by using the downburst-conf parameter on the command line.
To find the downburst executable, teuthology first checks the PATH environment variable. If not defined, teuthology next checks for src/downburst/virtualenv/bin/downburst executables in the user's home directory, /home/ubuntu, and /home/teuthology. This can all be overridden if the user specifies a downburst field in the user's .teuthology.yaml file.
Because teuthology reinstalls a new machine, a new hostkey is generated. After
locking, once a connection is established to the new machine,
teuthology-lock with the --list or --list-targets options will
display the new keys. When vps machines are locked using the --lock-many
option, a message is displayed indicating that --list-targets should be run
later.
Once teuthology starts after a new vm is installed, teuthology
checks for the existence of /ceph-qa-ready. If this file is not
present, ceph-qa-chef is run when teuthology first comes up.
It is assumed that downburst is on the user's $PATH.
Most of the current teuthology test suite execution scripts automatically
download their tests from the master branch of the appropriate github
repository. People who want to run experimental test suites usually modify the
download method in the teuthology/task script to use some other branch or
repository. This should be generalized in later teuthology releases.
Teuthology QA suites can be found in src/ceph-qa-suite. Make sure that this
directory exists in your source tree before running the test suites.
Each suite name is determined by the name of the directory in ceph-qa-suite
that contains that suite. The directory contains subdirectories and yaml files,
which, when assembled, produce valid tests that can be run. The test suite
application generates combinations of these files and thus ends up running a
set of tests based off the data in the directory for the suite.
To run a suite, enter:
./schedule_suite.sh <suite> <ceph> <kernel> <email> <flavor> <teuth> <mtype> <template>
where:
suite: the name of the suite (the directory in ceph-qa-suite).ceph: ceph branch to be used.kernel: version of the kernel to be used.email: email address to send the results to.flavor: flavor of the testteuth: version of teuthology to runmtype: machine type of the runtemplates: template file used for further modifying the suite (optional)
For example, consider:
schedule_suite.sh rbd wip-fix cuttlefish [email protected] master cuttlefish plana
The above command runs the rbd suite using wip-fix as the ceph branch, a straight cuttlefish kernel, and the master flavor of cuttlefish teuthology. It will run on plana machines.
In order for a queued task to be run, a teuthworker thread on
teuthology.front.sepia.ceph.com needs to remove the task from the queue.
On teuthology.front.sepia.ceph.com, run ps aux | grep teuthology-worker
to view currently running tasks. If no processes are reading from the test
version that you are running, additonal teuthworker tasks need to be started.
To start these tasks:
copy your build tree to
/home/teuthworkeronteuthology.front.sepia.ceph.com.Give it a unique name (in this example, xxx)
start up some number of worker threads (as many as machines you are testing with, there are 60 running for the default queue):
/home/virtualenv/bin/python /var/lib/teuthworker/xxx/virtualenv/bin/teuthworker /var/lib/teuthworker/archive --tube xxx --log-dir /var/lib/teuthworker/archive/worker_logs Note: The threads on teuthology.front.sepia.ceph.com are started via ~/teuthworker/start.sh. You can use that file as a model for your own threads, or add to this file if you want your threads to be more permanent.
Once the suite completes, an email message is sent to the users specified, and
a large amount of information is left on teuthology.front.sepia.ceph.com in
/var/lib/teuthworker/archive.
This is symbolically linked to /a for convenience. A new directory is created
whose name consists of a concatenation of the date and time that the suite was
started, the name of the suite, the ceph branch tested, the kernel used, and
the flavor. For every test run there is a directory whose name is the pid
number of the pid of that test. Each of these directory contains a copy of the
teuthology.log for that process. Other information from the suite is
stored in files in the directory, and task-specific yaml files and other logs
are saved in the subdirectories.
These logs are also publically available at
http://qa-proxy.ceph.com/teuthology/.