Install at an ESRF beamline#

For online data processing with Ewoks, a beamline needs at least one python environment with the scientific libraries installed (conda or anything else) and Ewoks to be configured in Beacon. When workflows are triggered from BLISS, the BLISS conda environment needs the ewoksjob python package.

Worker environment#

Create a python environment (conda, venv, …) as blissadm.

When using conda, create a conda environment for the worker (called ewoksworker in this example) like this

conda create --name ewoksworker python=3.9

Any python version supported by ewoksjob will do. Activate the environment

conda activate ewoksworker

Install basic worker dependencies

pip install ewoksjob[blissworker]

Install the project(s) that implement the actual worker tasks (depends on the beamline).

Some beamlines may want multiple environments for different scientific software.

Supervisor#

The supervisor config should be created in /users/blissadm/local/daemon/config/supervisor.d/ewoks.conf.

Each change made to the configuration must be applied by the supervisorctl shell command:

  • supervisorctl reread: to load changes in the config.

  • supervisorctl update: to apply changes in the config.

This is an example that registers a job monitor (you only ever need one) and one worker (you may need more than one):

[group:ewoks]
programs=ewoksmonitor, ewoksworker
priority=900

[program:ewoksmonitor]
command=bash -c "source /users/blissadm/conda/miniconda/bin/activate ewoksworker && ewoksjob monitor"
directory=/users/opid00/
user=opid00
environment=BEACON_HOST="id00:25000",FLOWER_UNAUTHENTICATED_API=true
startsecs=5
autostart=true
redirect_stderr=true
stdout_logfile=/var/log/%(program_name)s.log
stdout_logfile_maxbytes=1MB
stdout_logfile_backups=10
stdout_capture_maxbytes=1MB

[program:ewoksworker]
command=bash -c "source /users/blissadm/conda/miniconda/bin/activate ewoksworker && ewoksjob worker"
directory=/users/opid00/
user=opid00
environment=BEACON_HOST="id00:25000"
startsecs=5
autostart=true
redirect_stderr=true
stdout_logfile=/var/log/%(program_name)s.log
stdout_logfile_maxbytes=1MB
stdout_logfile_backups=10
stdout_capture_maxbytes=1MB

Make sure to replace opid00 and id00 with the proper string.

The filename /users/blissadm/conda/miniconda/bin/activate may differ as well. Type conda activate base;type activate in a terminal with the correct user to see where the conda activation script is located.

The working directory is important if the beamline wants to quickly add workflow tasks (i.e. without pip-installing a python project). In this case directory=/users/opid00/ewoks is probably more appropriate. All .py files in this directory and its subdirectories can contain ewoks tasks.

Instead of BEACON_HOST=”id00:25000” you could provide an explicit URL of the celery configuration with EWOKS_CONFIG_URI.

[program:ewoksmonitor]
...
environment=EWOKS_CONFIG_URI="beacon:///id00:25000/ewoks/config.yml"

[program:ewoksworker]
...
environment=EWOKS_CONFIG_URI="beacon:///id00:25000/ewoks/config.yml"

BLISS environment#

Activate the BLISS conda environment

. blissenv -d

Install the client dependencies for bliss >=1.11

pip install ewoksjob

For older versions of bliss do this instead

conda install celery  # Bliss dependencies were managed by conda
pip install ewoksjob, blissdata

This is only needed when workflows are triggered directly from BLISS.

When triggering workflows, either the BEACON_HOST or EWOKS_CONFIG_URI environment variables need to be provided. When triggering directly from BLISS, the BEACON_HOST environment variable is already set so nothing extra to do.

Beacon configuration#

Ewoks must be configured in the beamline configuration (Beacon). For example

# /users/blissadm/local/beamline_configuration/ewoks/__init__.yml

bliss_ignored: true

# /users/blissadm/local/beamline_configuration/ewoks/config.yml

celery:
    broker_url: "redis://id00:25001/2"
    result_backend: "redis://id00:25001/3"
    result_serializer: "pickle"
    accept_content: ["application/json", "application/x-python-serialize"]
    result_expires: 600
    task_remote_tracebacks: true
    enable_utc: false

Make sure to replace id00 with the host where the Redis database is running on (never use localhost).

The Celery documentation describes the different parameters available.

Test installation#

Run a test workflow in a BLISS session

DEMO_SESSION [1]: from ewoksjob.client import submit_test
DEMO_SESSION [2]: submit_test().get()
         Out [2]: {'return_value': True}

Slurm#

A worker that redirects jobs to slurm can be started with the –pool=slurm option

[group:ewoks]
programs=ewoksmonitor, ewoksworker, ewoksworker_slurm
priority=900

[program:ewoksworker_slurm]
command=bash -c "source /users/blissadm/conda/miniconda/bin/activate ewoksworker && exec ewoksjob worker --pool=slurm -Q slurm -n slurm@id00 --slurm-pre-script='module load ...' --slurm-log-directory='/tmp_14_days/{user_name}/slurm_logs' -sp time_limit=240"
directory=/users/opid00/
user=opid00
environment=
    BEACON_HOST="id00:25000",
    SLURM_URL="http://...",
    SLURM_USER="opid00",
    SLURM_TOKEN="eyJhbGciO..."
startsecs=2
autostart=true
redirect_stderr=true
stdout_logfile=/var/log/%(program_name)s.log
stdout_logfile_maxbytes=1MB
stdout_logfile_backups=10
stdout_capture_maxbytes=1MB

More details on the parameters can be found in the ewoksjob documentation. The environment specified with –slurm-pre-script needs to be setup like any other worker environment (see Worker environment).

Ewoks server#

To provide ewoks as a web service with a web GUI you can install this in the worker environment

pip install ewoksserver[frontend]

Configure the server in the supervisor configuration file

[group:ewoks]
programs=..., ewoksserver
priority=900

...

[program:ewoksserver]
command=bash -c "source /users/blissadm/conda/miniconda/bin/activate ewoksworker && python -u -m ewoksserver.server -c /users/blissadm/local/daemon/config/supervisor.d/ewoksserverconfig.py --host=0.0.0.0"
directory=/users/opid00/ewoks/
user=opid00
environment=BEACON_HOST="id00:25000"
startsecs=5
autostart=true
redirect_stderr=true
stdout_logfile=/var/log/%(program_name)s.log
stdout_logfile_maxbytes=1MB
stdout_logfile_backups=10
stdout_capture_maxbytes=1MB

Finally the server configuration should be saved in addition to the supervisor configuration

# /users/blissadm/local/daemon/config/supervisor.d/ewoksserverconfig.py

RESOURCE_DIRECTORY = "/users/opid01/ewoks/resources"

EWOKS = {
    "handlers": [
        {
            "class": "ewoksjob.events.handlers.RedisEwoksEventHandler",
            "arguments": [
                {
                    "name": "url",
                    "value": "redis://id00:25001/4",
                },
                {"name": "ttl", "value": 86400},
            ],
        }
    ]
}

CELERY = {}

Make sure to replace id00 with the beamline name.

Note: since CELERY is defined but empty, the server will use BEACON_HOST to fetch the celery configuration. This ensures that ewoks server executes workflows by sending them to the workers instead of executing them locally in a subprocess.