# 36 RabbitMQ

This section provides additional details about how PEcAn uses RabbitMQ to manage communication between its Docker containers.

In PEcAn, we use the Python pika client to post and retrieve messages from RabbitMQ. As such, every Docker container that communicates with RabbitMQ contains two Python scripts: sender.py and reciever.py. Both are located in the docker directory in the PEcAn source code root.

## 36.1 Producer – sender.py

The sender.py script is in charge of posting messages to RabbitMQ. In the RabbitMQ documentation, it is known as a “producer”. It runs once for every message posted to RabbitMQ, and then immediately exits (unlike the receiver.py, which runs continuously – see below).

Its usage is as follows:

python3 sender.py <URI> <queue> <message>

The arguments are:

• <URI> – The unique identifier of the RabbitMQ instance, similar to a URL. The format is amqp://username:password@host/vhost. By default, this is amqp://guest:guest@rabbitmq/%2F (the %2F here is the hexadecimal encoding for the / character).

• <queue> – The name of the queue on which to post the message.

• <message> – The contents of the message to post, in JSON format. A typical message posted by PEcAn looks like the following:

{ "folder" : "/path/to/PEcAn_WORKFLOWID", "workflow" : "WORKFLOWID" }

The PEcAn.remote::start_rabbitmq function is a wrapper for this script that provides an easy way to post a folder message to RabbitMQ from R.

Unlike sender.py, receiver.py runs like a daemon, constantly listening for messages. In the RabbitMQ documentation, it is known as a “consumer”. In PEcAn, you can tell that it is ready to receive messages if the corresponding logs (e.g. docker-compose logs executor) show the following message:

[*] Waiting for messages. To exit press CTRL+C.

Our reciever is configured by three environment variables:

• RABBITMQ_URI – This defines the URI where RabbitMQ is running. See corresponding argument in the producer

• RABBITMQ_QUEUE – This is the name of the queue on which the consumer will listen for messages, just as in the producer.

• APPLICATION – This specifies the name (including the path) of the default executable to run when receiving a message. At the moment, it should be an executable that runs in the directory specified by the message’s folder variable. In the case of PEcAn models, this is usually ./job.sh, such that the folder corresponds to the run directory associated with a particular runID (i.e. where the job.sh is located). For the PEcAn workflow itself, this is set to R CMD BATCH workflow.R, such that the folder is the root directory of the workflow (in the executor Docker container, something like /data/workflows/PEcAn_<workflowID>). This default executable is overridden if the message contains a custom_application key. If included, the string specified by the custom_application key will be run as a command exactly as is on the container, from the directory specified by folder. For instance, in the example below, the container will print “Hello there!” instead of running its default application.

{"custom_application": "echo 'Hello there!'", "folder": "/path/to/my/dir"}

NOTE that in RabbitMQ messages, the folder key is always required.

## 36.3 RabbitMQ and the PEcAn web interface

RabbitMQ is configured by the following variables in config.php:

• $rabbitmq_host – The RabbitMQ server hostname (default: rabbitmq, because that is the name of the rabbitmq service in docker-compose.yml) •$rabbitmq_port – The port on which RabbitMQ listens for messages (default: 5672)
• $rabbitmq_vhost – The path of the RabbitMQ Virtual Host (default: /). •$rabbitmq_queue – The name of the RabbitMQ queue associated with the PEcAn workflow (default: pecan)
• $rabbitmq_username – The RabbitMQ username (default: guest) •$rabbitmq_password – The RabbitMQ password (default: guest)

In addition, for running models via RabbitMQ, you will also need to add an entry like the following to the config.php $hostlist:$hostlist=array($fqdn => array("rabbitmq" => "amqp://guest:guest@rabbitmq/%2F"), ...) This will set the hostname to the name of the current machine (defined by the$fqdn variable earlier in the config.php file) to an array with one entry, whose key is rabbitmq and whose value is the RabbitMQ URI (amqp://...).

These values are converted into the appropriate entries in the pecan.xml in web/04-runpecan.php.

## 36.4 RabbitMQ in the PEcAn XML

RabbitMQ is a special case of remote execution, so it is configured by the host node. An example RabbitMQ configuration is as follows:

<host>
<rabbitmq>
<uri>amqp://guest:guest@rabbitmq/%2F</uri>
<queue>sipnet_136</queue>
</rabbitmq>
</host>

Here, uri and queue have the same general meanings as described in “producer”. Note that queue here refers to the target model. In PEcAn, RabbitMQ model queues are named as MODELTYPE_REVISION, so the example above refers to the SIPNET model version 136. Another example is ED2_git, referring to the latest git version of the ED2 model.

## 36.5 RabbitMQ configuration in Dockerfiles

As described in the “consumer” section, our standard RabbitMQ receiver script is configured using three environment variables: RABBITMQ_URI, RABBITMQ_QUEUE, and APPLICATION. Therefore, configuring a container to work with PEcAn’s RabbitMQ instance requires setting these three variables in the Dockerfile using an ENV statement.

For example, this excerpt from docker/base/Dockerfile.executor (for the pecan/executor image responsible for the PEcAn workflow) sets these variables as follows:

ENV RABBITMQ_URI="amqp://guest:guest@rabbitmq/%2F" \
RABBITMQ_QUEUE="pecan" \
APPLICATION="R CMD BATCH workflow.R"

Similarly, this excerpt from docker/models/Dockerfile.sipnet (which builds the SIPNET model image) is a typical example for a model image. Note the use of ARG here to specify a default version model version of 136 while allowing this to be configurable (via --build-arg MODEL_VERSION=X) at build time:

ARG MODEL_VERSION=136

ENV APPLICATION="./job.sh" \
MODEL_TYPE="SIPNET" \
MODEL_VERSION="${MODEL_VERSION}" ENV RABBITMQ_QUEUE="${MODEL_TYPE}_${MODEL_VERSION}" WARNING: Dockerfile environment variables set via ENV are assigned all at once; they do not evaluate successively, left to right. Consider the following block: # Don't do this! ENV MODEL_TYPE="SIPNET" \ MODEL_VERSION=136 \ RABBITMQ_QUEUE=${MODEL_TYPE}_\${MODEL_VERSION}   # <- Doesn't know about MODEL_TYPE or MODEL_VERSION!

In this block, the expansion for setting RABBITMQ_QUEUE is not aware of the current values of MODEL_TYPE or MODEL_VERSION, and will therefore be set incorrectly to just _ (unless they have been set previously, in which case it will be aware only of their earlier values). As such, variables depending on other variables must be set in a separate, subsequent ENV statement than the variables they depend on.

## 36.6 Case study: PEcAn web interface

The following describes in general terms what happens during a typical run of the PEcAn web interface with RabbitMQ.

1. The user initializes all containers with docker-compose up. All the services that interact with RabbitMQ (executor and all models) run receiver.py in the foreground, waiting for messages to tell them what to do.

2. The user browses to http://localhost:8000/pecan/ and steps through the web interface. All the pages up to the 04-runpecan.php run on the web container, and are primarily for setting up the pecan.xml file.

3. Once the user starts the PEcAn workflow at 04-runpecan.php, the underlying PHP code connects to RabbitMQ (based on the URI provided in config.php) and posts the following message to the pecan queue:

{"folder": "/workflows/PEcAn_WORKFLOWID", "workflowid": "WORKFLOWID"}
1. The executor service, which is listening on the pecan queue, hears this message and executes its APPLICATION (R CMD BATCH workflow.R) in the working directory specified in the message’s folder. The executor service then performs the pre-execution steps (e.g. trait meta-analysis, conversions) itself. Then, to actually execute the model, executor posts the following message to the target model’s queue:
{"folder": "/workflows/PEcAn_WORKFLOWID/run/RUNID"}
1. The target model service, which is listening on its dedicated queue, hears this message and runs its APPLICATION, which is job.sh, in the directory indicated by the message. Upon exiting (normally), the model service writes its status into a file called rabbitmq.out in the same directory.

2. The executor container continuously looks for the rabbitmq.out file as an indication of the model run’s status. Once it sees this file, it reads the status and proceeds with the post-execution parts of the workflow. (NOTE that this isn’t perfect. If the model running process exits abnormally, the rabbitmq.out file may not be created, which can cause the executor container to hang. If this happens, the solution is to restart the executor container with docker-compose restart executor).