Using PL/Container

This topic covers further details on:

PL/Container Resource Management
PL/Container Functions
Configuring a Remote PL/Container

PL/Container Resource Management

The Docker containers and the Greenplum Database servers share CPU and memory resources on the same hosts. In the default case, Greenplum Database is unaware of the resources consumed by running PL/Container instances. You can use Greenplum Database resource groups to control overall CPU and memory resource usage for running PL/Container instances.

PL/Container manages resource usage at two levels - the container level and the runtime level. You can control container-level CPU and memory resources with the memory_mb and cpu_share settings that you configure for the PL/Container runtime. memory_mb governs the memory resources available to each container instance. The cpu_share setting identifies the relative weighting of a container's CPU usage compared to other containers. See plcontainer Configuration File for further details.

You cannot, by default, restrict the number of running PL/Container container instances, nor can you restrict the total amount of memory or CPU resources that they consume.

Using Resource Groups to Manage PL/Container Resources

With PL/Container 1.2.0 and later, you can use Greenplum Database resource groups to manage and limit the total CPU and memory resources of containers in PL/Container runtimes. For more information about enabling, configuring, and using Greenplum Database resource groups, refer to Using Resource Groups in the Greenplum Database Administrator Guide.

Note
If you do not explicitly configure resource groups for a PL/Container runtime, its container instances are limited only by system resources. The containers may consume resources at the expense of the Greenplum Database server.

Resource groups for external components such as PL/Container use Linux control groups (cgroups) to manage component-level use of memory and CPU resources. When you manage PL/Container resources with resource groups, you configure both a memory limit and a CPU limit that Greenplum Database applies to all container instances that share the same PL/Container runtime configuration.

When you create a resource group to manage the resources of a PL/Container runtime, you must specify MEMORY_AUDITOR=cgroup and CONCURRENCY=0 in addition to the required CPU and memory limits. For example, the following command creates a resource group named plpy_run1_rg for a PL/Container runtime:

CREATE RESOURCE GROUP plpy_run1_rg WITH (MEMORY_AUDITOR=cgroup, CONCURRENCY=0,
                                                  CPU_RATE_LIMIT=10, MEMORY_LIMIT=10);

PL/Container does not use the MEMORY_SHARED_QUOTA and MEMORY_SPILL_RATIO resource group memory limits. Refer to the CREATE RESOURCE GROUP reference page for detailed information about this SQL command.

You can create one or more resource groups to manage your running PL/Container instances. After you create a resource group for PL/Container, you assign the resource group to one or more PL/Container runtimes. You make this assignment using the groupid of the resource group. You can determine the groupid for a given resource group name from the gp_resgroup_config gp_toolkit view. For example, the following query displays the groupid of a resource group named plpy_run1_rg:

SELECT groupname, groupid FROM gp_toolkit.gp_resgroup_config
 WHERE groupname='plpy_run1_rg';
                            
 groupname   |  groupid
 --------------+----------
 plpy_run1_rg |   16391
 (1 row)

You assign a resource group to a PL/Container runtime configuration by specifying the -s resource_group_id=rg\_groupid option to the plcontainer runtime-add (new runtime) or plcontainer runtime-replace (existing runtime) commands. For example, to assign the plpy_run1_rg resource group to a new PL/Container runtime named python_run1:

plcontainer runtime-add -r python_run1 -i pivotaldata/plcontainer_python_shared:devel -l python -s resource_group_id=16391

You can also assign a resource group to a PL/Container runtime using the plcontainer runtime-edit command. For information about the plcontainer command, see plcontainer reference page.

After you assign a resource group to a PL/Container runtime, all container instances that share the same runtime configuration are subject to the memory limit and the CPU limit that you configured for the group. If you decrease the memory limit of a PL/Container resource group, queries running in containers in the group may fail with an out of memory error. If you drop a PL/Container resource group while there are running container instances, Greenplum Database terminates the running containers.

Configuring Resource Groups for PL/Container

To use Greenplum Database resource groups to manage PL/Container resources, you must explicitly configure both resource groups and PL/Container.

Perform the following procedure to configure PL/Container to use Greenplum Database resource groups for CPU and memory resource management:

If you have not already configured and enabled resource groups in your Greenplum Database deployment, configure cgroups and enable Greenplum Database resource groups as described in Using Resource Groups in the Greenplum Database Administrator Guide.

Note
If you have previously configured and enabled resource groups in your deployment, ensure that the Greenplum Database resource group gpdb.conf cgroups configuration file includes a memory { } block as described in the previous link.
Analyze the resource usage of your Greenplum Database deployment. Determine the percentage of resource group CPU and memory resources that you want to allocate to PL/Container Docker containers.
Determine how you want to distribute the total PL/Container CPU and memory resources that you identified in the step above among the PL/Container runtimes. Identify:
- The number of PL/Container resource group(s) that you require.
- The percentage of memory and CPU resources to allocate to each resource group.
- The resource-group-to-PL/Container-runtime assignment(s).
Create the PL/Container resource groups that you identified in the step above. For example, suppose that you choose to allocate 25% of both memory and CPU Greenplum Database resources to PL/Container. If you further split these resources among 2 resource groups 60/40, the following SQL commands create the resource groups:
```
CREATE RESOURCE GROUP plr_run1_rg WITH (MEMORY_AUDITOR=cgroup, CONCURRENCY=0,
                                           CPU_RATE_LIMIT=15, MEMORY_LIMIT=15);
 CREATE RESOURCE GROUP plpy_run1_rg WITH (MEMORY_AUDITOR=cgroup, CONCURRENCY=0,
                                          CPU_RATE_LIMIT=10, MEMORY_LIMIT=10);
```

Find and note the groupid associated with each resource group that you created. For example:

SELECT groupname, groupid FROM gp_toolkit.gp_resgroup_config
WHERE groupname IN ('plpy_run1_rg', 'plr_run1_rg');

groupname   |  groupid
--------------+----------
plpy_run1_rg |   16391
plr_run1_rg  |   16393
(1 row)

Assign each resource group that you created to the desired PL/Container runtime configuration. If you have not yet created the runtime configuration, use the plcontainer runtime-add command. If the runtime already exists, use the plcontainer runtime-replace or plcontainer runtime-edit command to add the resource group assignment to the runtime configuration. For example:
```
plcontainer runtime-add -r python_run1 -i pivotaldata/plcontainer_python_shared:devel -l python -s resource_group_id=16391
plcontainer runtime-replace -r r_run1 -i pivotaldata/plcontainer_r_shared:devel -l r -s resource_group_id=16393
```
For information about the plcontainer command, see plcontainer reference page.

Notes

PL/Container logging

When PL/Container logging is enabled, you can set the log level with the Greenplum Database server configuration parameter log_min_messages. The default log level is warning. The parameter controls the PL/Container log level and also controls the Greenplum Database log level.

PL/Container logging is enabled or deactivated for each runtime ID with the setting attribute use_container_logging. The default is no logging.
The PL/Container log information is the information from the UDF that is run in the Docker container. By default, the PL/Container log information is sent to a system service. On Red Hat 7 or CentOS 7 systems, the log information is sent to the journald service.
The Greenplum Database log information is sent to log file on the Greenplum Database master.
When testing or troubleshooting a PL/Container UDF, you can change the Greenplum Database log level with the SET command. You can set the parameter in the session before you run your PL/Container UDF. This example sets the log level to debug1.
```
SET log_min_messages='debug1' ;
```
Note
The parameter log_min_messages controls both the Greenplum Database and PL/Container logging, increasing the log level might affect Greenplum Database performance even if a PL/Container UDF is not running.

PL/Container Function Limitations

Review the following limitations when creating and using PL/Container PL/Python and PL/R functions:

Greenplum Database domains are not supported.
Multi-dimensional arrays are not supported.
Python and R call stack information is not displayed when debugging a UDF.
The plpy.execute() methods nrows() and status() are not supported.
The PL/Python function plpy.SPIError() is not supported.
Running the SAVEPOINT command with plpy.execute() is not supported.
The DO command (anonymous code block) is supported only with PL/Container 3 (currently a Beta feature).
Container flow control is not supported.
Triggers are not supported.
OUT parameters are not supported.
The Python dict type cannot be returned from a PL/Python UDF. When returning the Python dict type from a UDF, you can convert the dict type to a Greenplum Database user-defined data type (UDT).

Developing PL/Container functions

When you enable PL/Container in a database of a Greenplum Database system, the language plcontainer is registered in that database. Specify plcontainer as a language in a UDF definition to create and run user-defined functions in the procedural languages supported by the PL/Container Docker images.

A UDF definition that uses PL/Container must have these items.

The first line of the UDF must be # container: ID
The LANGUAGE attribute must be plcontainer

The ID is the name that PL/Container uses to identify a Docker image. When Greenplum Database runs a UDF on a host, the Docker image on the host is used to start a Docker container that runs the UDF. In the XML configuration file plcontainer_configuration.xml, there is a runtime XML element that contains a corresponding id XML element that specifies the Docker container startup information. See plcontainer Configuration File for information about how PL/Container maps the ID to a Docker image.

The PL/Container configuration file is read only on the first invocation of a PL/Container function in each Greenplum Database session that runs PL/Container functions. You can force the configuration file to be re-read by performing a SELECT command on the view plcontainer_refresh_config during the session. For example, this SELECT command forces the configuration file to be read.

SELECT * FROM plcontainer_refresh_config;

The command runs a PL/Container function that updates the configuration on the master and segment instances and returns the status of the refresh.

 gp_segment_id | plcontainer_refresh_local_config
 ---------------+----------------------------------
 1 | ok
 0 | ok
-1 | ok
(3 rows)

Also, you can show all the configurations in the session by performing a SELECT command on the view plcontainer_show_config. For example, this SELECT command returns the PL/Container configurations.

SELECT * FROM plcontainer_show_config;

Running the command executes a PL/Container function that displays configuration information from the master and segment instances. This is an example of the start and end of the view output.

INFO:  plcontainer: Container 'plc_py_test' configuration
 INFO:  plcontainer:     image = 'pivotaldata/plcontainer_python_shared:devel'
 INFO:  plcontainer:     memory_mb = '1024'
 INFO:  plcontainer:     use container network = 'no'
 INFO:  plcontainer:     use container logging  = 'no'
 INFO:  plcontainer:     shared directory from host '/usr/local/greenplum-db/./bin/plcontainer_clients' to container '/clientdir'
 INFO:  plcontainer:     access = readonly
                
 ...
                
 INFO:  plcontainer: Container 'plc_r_example' configuration  (seg0 slice3 192.168.180.45:40000 pid=3304)
 INFO:  plcontainer:     image = 'pivotaldata/plcontainer_r_without_clients:0.2'  (seg0 slice3 192.168.180.45:40000 pid=3304)
 INFO:  plcontainer:     memory_mb = '1024'  (seg0 slice3 192.168.180.45:40000 pid=3304)
 INFO:  plcontainer:     use container network = 'no'  (seg0 slice3 192.168.180.45:40000 pid=3304)
 INFO:  plcontainer:     use container logging  = 'yes'  (seg0 slice3 192.168.180.45:40000 pid=3304)
 INFO:  plcontainer:     shared directory from host '/usr/local/greenplum-db/bin/plcontainer_clients' to container '/clientdir'  (seg0 slice3 192.168.180.45:40000 pid=3304)
 INFO:  plcontainer:         access = readonly  (seg0 slice3 192.168.180.45:40000 pid=3304)
 gp_segment_id | plcontainer_show_local_config
 ---------------+-------------------------------
  0 | ok
 -1 | ok
  1 | ok

The PL/Container function plcontainer_containers_summary() displays information about the currently running Docker containers.

SELECT * FROM plcontainer_containers_summary();

If a normal (non-superuser) Greenplum Database user runs the function, the function displays information only for containers created by the user. If a Greenplum Database superuser runs the function, information for all containers created by Greenplum Database users is displayed. This is sample output when 2 containers are running.

 SEGMENT_ID |                           CONTAINER_ID                           |   UP_TIME    |  OWNER  | MEMORY_USAGE(KB)
 ------------+------------------------------------------------------------------+--------------+---------+------------------
 1          | 693a6cb691f1d2881ec0160a44dae2547a0d5b799875d4ec106c09c97da422ea | Up 8 seconds | gpadmin | 12940
 1          | bc9a0c04019c266f6d8269ffe35769d118bfb96ec634549b2b1bd2401ea20158 | Up 2 minutes | gpadmin | 13628
 (2 rows)

When Greenplum Database runs a PL/Container UDF, Query Executer (QE) processes start Docker containers and reuse them as needed. After a certain amount of idle time, a QE process quits and destroys its Docker containers. You can control the amount of idle time with the Greenplum Database server configuration parameter gp_vmem_idle_resource_timeout. Controlling the idle time might help with Docker container reuse and avoid the overhead of creating and starting a Docker container.

Caution
Changing gp_vmem_idle_resource_timeout value, might affect performance due to resource issues. The parameter also controls the freeing of Greenplum Database resources other than Docker containers.

Basic Function Examples

The values in the # container lines of the examples, plc_python_shared and plc_r_shared, are the id XML elements defined in the plcontainer_config.xml file. The id element is mapped to the image element that specifies the Docker image to be started. If you configured PL/Container with a different ID, change the value of the # container line. For information about configuring PL/Container and viewing the configuration settings, see plcontainer Configuration File.

This is an example of PL/Python function that runs using the plc_python_shared container that contains Python 2:

CREATE OR REPLACE FUNCTION pylog100() RETURNS double precision AS $$
 # container: plc_python_shared
 import math
 return math.log10(100)
 $$ LANGUAGE plcontainer;

This is an example of a similar function using the plc_r_shared container:

CREATE OR REPLACE FUNCTION rlog100() RETURNS text AS $$
# container: plc_r_shared
return(log10(100))
$$ LANGUAGE plcontainer;

If the # container line in a UDF specifies an ID that is not in the PL/Container configuration file, Greenplum Database returns an error when you try to run the UDF.

About PL/Python 2 Functions in PL/Container

In the Python 2 language container, the module plpy is implemented. The module contains these methods:

plpy.execute(stmt) - Runs the query string stmt and returns query result in a list of dictionary objects. To be able to access the result fields ensure your query returns named fields.
plpy.prepare(stmt[, argtypes]) - Prepares the execution plan for a query. It is called with a query string and a list of parameter types, if you have parameter references in the query.
plpy.execute(plan[, argtypes]) - Runs a prepared plan.
plpy.debug(msg) - Sends a DEBUG2 message to the Greenplum Database log.
plpy.log(msg) - Sends a LOG message to the Greenplum Database log.
plpy.info(msg) - Sends an INFO message to the Greenplum Database log.
plpy.notice(msg) - Sends a NOTICE message to the Greenplum Database log.
plpy.warning(msg) - Sends a WARNING message to the Greenplum Database log.
plpy.error(msg) - Sends an ERROR message to the Greenplum Database log. An ERROR message raised in Greenplum Database causes the query execution process to stop and the transaction to rollback.
plpy.fatal(msg) - Sends a FATAL message to the Greenplum Database log. A FATAL message causes Greenplum Database session to be closed and transaction to be rolled back.
plpy.subtransaction() - Manages plpy.execute calls in an explicit subtransaction. See Explicit Subtransactions in the PostgreSQL documentation for additional information about plpy.subtransaction().

If an error of level ERROR or FATAL is raised in a nested Python function call, the message includes the list of enclosing functions.

The Python language container supports these string quoting functions that are useful when constructing ad-hoc queries.

plpy.quote_literal(string) - Returns the string quoted to be used as a string literal in an SQL statement string. Embedded single-quotes and backslashes are properly doubled. quote_literal() returns null on null input (empty input). If the argument might be null, quote_nullable() might be more appropriate.
plpy.quote_nullable(string) - Returns the string quoted to be used as a string literal in an SQL statement string. If the argument is null, returns NULL. Embedded single-quotes and backslashes are properly doubled.
plpy.quote_ident(string) - Returns the string quoted to be used as an identifier in an SQL statement string. Quotes are added only if necessary (for example, if the string contains non-identifier characters or would be case-folded). Embedded quotes are properly doubled.

When returning text from a PL/Python function, PL/Container converts a Python unicode object to text in the database encoding. If the conversion cannot be performed, an error is returned.

PL/Container does not support this Greenplum Database PL/Python feature:

Multi-dimensional arrays.

Also, the Python module has two global dictionary objects that retain the data between function calls. They are named GD and SD. GD is used to share the data between all the function running within the same container, while SD is used for sharing the data between multiple calls of each separate function. Be aware that accessing the data is possible only within the same session, when the container process lives on a segment or master. Be aware that for idle sessions Greenplum Database terminates segment processes, which means the related containers would be shut down and the data from GD and SD lost.

For information about PL/Python, see PL/Python Language.

For information about the plpy methods, see https://www.postgresql.org/docs/9.4/plpython-database.htm.

About PL/Python 3 Functions in PL/Container

PL/Container for Greenplum Database 5 supports Python version 3.6+. PL/Container for Greenplum Database 6 supports Python 3.7+.

If you want to use PL/Container to run the same function body in both Python2 and Python3, you must create 2 different user-defined functions.

Keep in mind that UDFs that you created for Python 2 may not run in PL/Container with Python 3. The following Python references may be useful:

Changes to Python - What’s New in Python 3
Porting from Python 2 to 3 - Porting Python 2 Code to Python 3

Developing CUDA API Functions with PL/Container

Beginning with version 2.2, PL/Container supports developing Compute Unified Device Architecture (CUDA) API functions that utilize NVIDIA GPU hardware. This is accomplished by using the NVIDIA Container Toolkit nvidia-docker image and the pycuda python library. This procedure explains how to set up PL/Container for developing these functions.

Prerequisites

To develop CUDA functions with PL/Container you require:

A Docker installation having Docker engine version v19.03 or newer
PL/Container version 2.2.0 or newer
At least one NVIDIA GPU with the required GPU driver installed on your host

See the Getting Started section of the NVIDIA Container Toolkit GitHub project for information about installing the NVIDIA driver or Docker engine for your Linux distribution.

Follow the Installation Guide for the NVIDIA Container Toolkit GitHub project to install the nvidia-docker container.

Verify that the Docker image can use your installed GPU(s) by running a command similar to:

$ docker run --rm --gpus=all -it nvidia/cuda:11.7.0-devel-ubuntu20.04 nvidia-smi –L

(Substitute the actual nvidia-docker image name and tag that you installed.) The command output should show that GPU hardware is utilized. For example:

GPU 0: NVIDIA GeForce RTX 2070 (UUID: GPU-d4d626a3-bbc9-ef88-98dc-44423ad081bf)

Record the name of the GPU device ID (0 in the above example) or the device UUID (GPU-d4d626a3-bbc9-ef88-98dc-44423ad081bf) that you want to assign to the PL/Container image.

Install and Customize the PL/Container Image

Note
For more information about download prerequisites, troubleshooting, and instructions, see Download Broadcom products and software.

Download the plcontainer-python3-image-2.2.0-gp6.tar.gz file from Broadcom Support Portal under the desired Greenplum release.

Load the downloaded PL/Container image into Docker:

$ docker image load < plcontainer-python3-image-2.2.0-gp6.tar.gz

Customize the PL/Container image to add the required CUDA runtime and pycuda library. The following example Dockerfile contents show how to add CUDA 11.7 and pycuda 2021.1 to the PL/Container image. Use a text editor to create the Dockerfile:

FROM pivotaldata/plcontainer_python3_shared:devel 

ENV XKBLAYOUT=en 
ENV DEBIAN_FRONTEND=noninteractive 

# Install CUDA from https://developer.nvidia.com/cuda-downloads 
# By downloading and using the software, you agree to fully comply with the terms and conditions of the CUDA EULA. 
RUN true &&\ 
    wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin && \ 
    mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600 && \ 
    wget https://developer.download.nvidia.com/compute/cuda/11.7.0/local_installers/cuda-repo-ubuntu1804-11-7-local_11.7.0-515.43.04-1_amd64.deb && \ 
    dpkg -i cuda-repo-ubuntu1804-11-7-local_11.7.0-515.43.04-1_amd64.deb && \ 
    cp /var/cuda-repo-ubuntu1804-11-7-local/cuda-*-keyring.gpg /usr/share/keyrings/ && \ 
    apt-get update && \ 
    apt-get -y install cuda && \ 
    rm cuda-repo-ubuntu1804-11-7-local_11.7.0-515.43.04-1_amd64.deb &&\ 
    rm -rf /var/lib/apt/lists/* 

ENV PATH="/usr/local/cuda-11.7/bin/:${PATH}" 
ENV LD_LIBRARY_PATH="/usr/local/cuda-11.7/lib64:${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}" 
ENV CUDA_HOME="/usr/local/cuda-11.7" 

RUN true && \ 
    python3.7 -m pip --no-cache-dir install typing-extensions==3.10.0.0 && \ 
    python3.7 -m pip --no-cache-dir install Mako==1.2.0 && \ 
    python3.7 -m pip --no-cache-dir install platformdirs==2.5.2 && \ 
    python3.7 -m pip --no-cache-dir install pytools==2022.1.2 && \ 
    python3.7 -m pip --no-cache-dir install pycuda==2021.1

Build the a customized container using your Dockerfile:
```
$ docker build . -t localhost/plcontainer_python3_cuda_shared:latest
```
Note
The remaining instructions use the example image tag localhost/plcontainer_python3_cuda_shared:latest. Substitute the actual tag name as needed.

Import the image runtime to PL/Container:

$ plcontainer runtime-add -r plc_python_cuda_shared -I localhost/plcontainer_python3_cuda_shared:latest -l python3

Edit the image runtime to assign a GPU. The following example adds GPU device ID 0 as the GPU, and gpadmin as the designated role. Substitute either the GPU device ID or the device UUID that you recorded earlier:

$ plcontainer runtime-edit

<runtime> 
    <id>plc_python_cuda_shared</id> 
    <image>localhost/plcontainer_python3_cuda_shared:latest</image> 
    <command>/clientdir/py3client.sh</command> 
    <setting roles="gpadmin"/> 
    <shared_directory access="ro" container="/clientdir" host="/usr/local/greenplum-db/bin/plcontainer_clients"/> 
    <device_request type="gpu"> 
        <deviceid>0</deviceid> 
    </device_request> 
</runtime>

Create and Run a Sample CUDA Function

Connect to a Greenplum database where PL/Container is installed:
```
$ psql -d mytest -h master_host -p 5432 -U `gpadmin`
```

Create a sample PL/Container function that uses the container you customized (plc_python_cuda_shared in this example). This simple function multiplies randomized, single-precision numbers by sending them to the CUDA constructor of pycuda.compiler.SourceModule:

CREATE FUNCTION hello_cuda() RETURNS float4[] AS $$ 
# container: plc_python_cuda_shared 

import pycuda.driver as drv 
import pycuda.tools 
import pycuda.autoinit 
import numpy 
import numpy.linalg as la 
from pycuda.compiler import SourceModule 

mod = SourceModule(""" 
__global__ void multiply_them(float *dest, float *a, float *b) 
{ 
  const int i = threadIdx.x; 
  dest[i] = a[i] * b[i]; 
} 
""") 

multiply_them = mod.get_function("multiply_them") 

a = numpy.random.randn(400).astype(numpy.float32) 
b = numpy.random.randn(400).astype(numpy.float32) 

dest = numpy.zeros_like(a) 
multiply_them( 
        drv.Out(dest), drv.In(a), drv.In(b), 
        block=(400,1,1)) 

return [float(i) for i in (dest-a*b)] 

$$ LANGUAGE plcontainer;

Run the sample function and verify its output:

$ WITH a AS (SELECT unnest(hello) AS cuda FROM hello_cuda() AS hello) SELECT sum(cuda) FROM a;

psql>   +-----+ 
psql>   | sum | 
psql>   |-----| 
psql>   | 0.0 | 
psql>   +-----+ 
psql>   SELECT 1 
psql>   Time: 0.012s

$ SELECT * FROM hello_cuda();

psql>   +-----------------------+ 
psql>   |       hello_cuda      | 
psql>   |-----------------------| 
psql>   | {0, 0.... many 0 ...} | 
psql>   +-----------------------+ 
psql>   SELECT 1 
psql>   Time: 0.012s

About PL/R Functions in PL/Container

In the R language container, the module pg.spi is implemented. The module contains these methods:

pg.spi.exec(stmt) - Runs the query string stmt and returns query result in R data.frame. To be able to access the result fields make sure your query returns named fields.
pg.spi.prepare(stmt[, argtypes]) - Prepares the execution plan for a query. It is called with a query string and a list of parameter types if you have parameter references in the query.
pg.spi.execp(plan[, argtypes]) - Runs a prepared plan.
pg.spi.debug(msg) - Sends a DEBUG2 message to the Greenplum Database log.
pg.spi.log(msg) - Sends a LOG message to the Greenplum Database log.
pg.spi.info(msg) - Sends an INFO message to the Greenplum Database log.
pg.spi.notice(msg) - Sends a NOTICE message to the Greenplum Database log.
pg.spi.warning(msg) - Sends a WARNING message to the Greenplum Database log.
pg.spi.error(msg) - Sends an ERROR message to the Greenplum Database log. An ERROR message raised in Greenplum Database causes the query execution process to stop and the transaction to rollback.
pg.spi.fatal(msg) - Sends a FATAL message to the Greenplum Database log. A FATAL message causes Greenplum Database session to be closed and transaction to be rolled back.

PL/Container does not support this PL/R feature:

Multi-dimensional arrays.

For information about PL/R, see PL/R Language.

For information about the pg.spi methods, see http://www.joeconway.com/plr/doc/plr-spi-rsupport-funcs-normal.html

Configuring a Remote PL/Container

You may configure one or more hosts outside your Greenplum cluster to use as a remote container host. The PL/Container workload can be dispatched to this host for execution and it will return the results, reducing the computing overload of the Greenplum hosts.

Prerequisites

You are using PL/Container version 2.4.0.
You are using a Docker installation with a Docker engine version v19.03 or newer.
You have root or sudo permission on the remote host.

Configure the Remote Host

Install docker on the remote host. This step may vary depending on your operating system. For example, for RHEL 7:

sudo yum install -y yum-utils 
sudo yum-config-manager --add-repo https://download.docker.com/linux/centos/docker-ce.repo 
sudo yum install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin 
sudo systemctl enable  --now docker

Enable the remote API for Docker:

sudo systemctl edit docker.service 
# add the following to the start of the file:
 
[Service] 
ExecStart= 
ExecStart=/usr/bin/dockerd -H fd:// -H tcp://0.0.0.0:2375 
# restart docker service 
sudo systemctl restart docker

Set up the remote host. This example assumes that you have created the gpadmin user, enabled password-less ssh access, and that python3 and rsync are installed in the remote host.

ssh gpadmin@<remoteip> "sudo mkdir $GPHOME && sudo chown gpadmin:gpadmin $GPHOME"

From the Greenplum coordinator, copy the plcontainer client to the remote host.

plcontainer remote-setup --hosts <remoteip>

If you are configuring multiple hosts, you may run the command against multiple remote hosts:

plcontainer remote-setup --hosts <remoteip_1>, <remoteip_2>, <remoteip_3>

Load the Docker Image to the Remote Host

From the coordinator host, load the Docker image into the remote host. You may run the command against multiple remote hosts:

plcontainer image-add --hosts <remoteip_1>, <remoteip_2>, <remoteip_3> -f <image_file>

Configure a Backend Node

Run the following command from the coordinator host:

plcontainer runtime-edit

This command provides the PL/Container configuration XML file. Add the backend section, as depicted in the below example, specifying the remote host IP address and port. Then edit the existing runtime section to use the newly added backend.

<?xml version="1.0" ?> 
<configuration> 
	<backend name="calculate_cluster" type="remote_docker"> 
		<address>{THE REMOTE ADDRESS}</address> 
		<port>2375</port> 
	</backend> 
	<runtime> 
		<id>plc_python_cuda_shared</id> 
		<image>localhost/plcontainer_python3_cuda_shared:latest</image> 
		<command>/clientdir/py3client.sh</command> 
		<shared_directory access="ro" container="/clientdir" host="/home/sa/GPDB/install/bin/plcontainer_clients"/> 
		<backend name="calculate_cluster" /> 
	</runtime> 
</configuration>

If you are using multiple remote hosts, you must create separate backend sections. Because you can only set one backend per runtime, you must also create a separate runtime section per backend.

Verify the Configuration

Run the following from the psql command line:

CREATE FUNCTION dummyPython() RETURNS text AS $$ 
# container: plc_python_cuda_shared 
return 'hello from Python' 
$$ LANGUAGE plcontainer; 
 
SELECT * from dummyPython()

If the function runs successfully, it is running on the remote host.