Infrastructure at your Service

David Barbarin

Deploying SQL Server 2019 container on RHEL 8 with podman

Having a fresh install of RHEL8 on my lab environment, I was curious to take a look at new containerization stuff from Red Hat in the context of SQL Server 2019. Good chances are the future version of SQL Server should be available and supported on with the latest version of Red Hat but for now this blog post is purely experimental. This time I wanted to share with you some thoughts about the new Podman command.

First of all, we should be aware that since RHEL8 Red Hat decided to replace docker with CRI-O/podman in order to provide a “daemonless” container world and especially for Kubernetes. By 2016, Kubernetes project introduced the Container Runtime Interface (CRI).  Basically, with CRI, Kubernetes can be container runtime-agnostic. CRI-O that is an open source project initiated by Red Hat the same year that gives the ability to run containers directly from Kubernetes without any unnecessary code or tooling as long as the container remains OCI-compliant. Because Docker is not implemented anymore (and officially not supported) by Red Hat since RHEL8, we need a client tool for working with containers and this is where Podman steps in. To cut the story short, Podman implements almost all the Docker CLI commands and more.

So, let’s have an overview of Podman commands through the installation of a SQL Server 2019 based container. It is worth noting that Podman is not intended to be used in the context of a “standalone” container environnement and should be used with an container orchestrator like K8s or an orchestration platform like OpenShift.  That said,  let’s first create a host directory to persist the SQL Server database files.

$ sudo mkdir -p  /var/mssql/data
$ sudo chmod 755 -R /var/mssql/data

 

Then let’s download the SQL Server 2019 RHEL image. We will use the following Podman command:

$ sudo podman pull mcr.microsoft.com/mssql/rhel/server:2019-CTP3.1
Trying to pull mcr.microsoft.com/mssql/rhel/server:2019-CTP3.1...Getting image source signatures
Copying blob 079e961eee89: 70.54 MiB / 70.54 MiB [========================] 1m3s
Copying blob 1b493d38a6d3: 1.20 KiB / 1.20 KiB [==========================] 1m3s
Copying blob 89e62e5b4261: 333.24 MiB / 333.24 MiB [======================] 1m3s
Copying blob d39017c722a8: 174.82 MiB / 174.82 MiB [======================] 1m3s
Copying config dbba412361d7: 4.98 KiB / 4.98 KiB [==========================] 0s
Writing manifest to image destination
Storing signatures
dbba412361d7ca4fa426387e1d6fc3ec85e37d630bfe70e6599b5116d392394d

 

Note that if you’re already comfortable with the Docker commands, the shift to Podman will be easy thanks to the similarity between the both tools. To get information of the new fresh image, we will use the following Podman command:

$ sudo podman images
REPOSITORY                            TAG           IMAGE ID       CREATED       SIZE
mcr.microsoft.com/mssql/rhel/server   2019-CTP3.1   dbba412361d7   3 weeks ago   1.79 GB
$ sudo podman inspect dbba
…
"GraphDriver": {
            "Name": "overlay",
            "Data": {
                "LowerDir": "/var/lib/containers/storage/overlay/b2769e971a1bdb62f1c0fd9dcc0e9fe727dca83f52812abd34173b49ae55e37d/diff:/var/lib/containers/storage/overlay/4b0cbf0d9d0ff230916734a790f47ab2adba69db44a79c8eac4c814ff4183c6d/diff:/var/lib/containers/storage/overlay/9197342671da8b555f200e47df101da5b7e38f6d9573b10bd3295ca9e5c0ae28/diff",
                "MergedDir": "/var/lib/containers/storage/overlay/b372c0d6ff718d2d182af4639870dc6e4247f684d81a8b2dc2649f8517b9fc53/merged",
                "UpperDir": "/var/lib/containers/storage/overlay/b372c0d6ff718d2d182af4639870dc6e4247f684d81a8b2dc2649f8517b9fc53/diff",
                "WorkDir": "/var/lib/containers/storage/overlay/b372c0d6ff718d2d182af4639870dc6e4247f684d81a8b2dc2649f8517b9fc53/work"
            }
        },
…

 

As show above, Podman uses the CRI-O back-end store directory with the /var/lib/containers path, instead of using the Docker default storage location (/var/lib/docker).

Go ahead and let’s take a look at the Podman info command:

$ podman info
…
OCIRuntime:
    package: runc-1.0.0-54.rc5.dev.git2abd837.module+el8+2769+577ad176.x86_64
    path: /usr/bin/runc
    version: 'runc version spec: 1.0.0'
…
store:
  ConfigFile: /home/clustadmin/.config/containers/storage.conf
  ContainerStore:
    number: 0
  GraphDriverName: overlay

 

The same kind of information is provided by the Docker info command including the runtime and the graph driver name that is overlay in my case. Generally speaking, creating and getting information of a container with Podman is pretty similar to what we may use with the usual Docker commands. Here  for instance the command to spin up a SQL Server container based on the RHEL image:

$ sudo podman run -d -e 'ACCEPT_EULA=Y' -e \
> 'MSSQL_SA_PASSWORD=Password1'  \
> --name 'sqltest' \
> -p 1460:1433 \
> -v /var/mssql/data:/var/opt/mssql/data:Z \
> mcr.microsoft.com/mssql/rhel/server:2019-CTP3.1
4f5128d36e44b1f55d23e38cbf8819041f84592008d0ebb2b24ff59065314aa4
$ sudo podman ps
CONTAINER ID  IMAGE                                            COMMAND               CREATED        STATUS            PORTS                   NAMES
4f5128d36e44  mcr.microsoft.com/mssql/rhel/server:2019-CTP3.1  /opt/mssql/bin/sq...  4 seconds ago  Up 3 seconds ago  0.0.0.0:1460->1433/tcp  sqltest

 

Here comes the interesting part. Looking at the pstree output we may notice that there is not dependencies with any (docker) daemon with CRI-O implementation. Usually with the Docker implementation we retrieve the containerd daemon and the related shim for the process within the tree. 

$ pstree
systemd─┬─NetworkManager───2*[{NetworkManager}]
        ├─…
        ├─conmon─┬─sqlservr─┬─sqlservr───138*[{sqlservr}]
        │        │          └─{sqlservr}

 

By using the runc command below, we may notice the MSSQL container (identified by the ID here) is actually running through CRI-O and runc runtime.

$ sudo runc list -q
4f5128d36e44b1f55d23e38cbf8819041f84592008d0ebb2b24ff59065314aa4

 

Let’s have a look at the existing namespace. The 9449 PID corresponds to the SQL Server process running in isolation mode through Linux namespaces.

$ sudo lsns 
…
4026532116 net         2  9449 root   /opt/mssql/bin/sqlservr
4026532187 mnt         2  9449 root   /opt/mssql/bin/sqlservr
4026532188 uts         2  9449 root   /opt/mssql/bin/sqlservr
4026532189 ipc         2  9449 root   /opt/mssql/bin/sqlservr
4026532190 pid         2  9449 root   /opt/mssql/bin/sqlservr

$ ps aux | grep sqlservr
root       9449  0.1  0.6 152072 25336 ?        Ssl  05:08   0:00 /opt/mssql/bin/sqlservr
root       9465  5.9 18.9 9012096 724648 ?      Sl   05:08   0:20 /opt/mssql/bin/sqlservr
clustad+   9712  0.0  0.0  12112  1064 pts/0    S+   05:14   0:00 grep --color=auto sqlservr

 

We can double check that the process belongs to the SQL Server container by using the nsenter command:

sudo nsenter -t 17182 --mount --uts --ipc --net --pid sh
sh-4.2# ps aux
USER        PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root          1  0.0  0.7 152076 28044 ?        Ssl  Jul23   0:00 /opt/mssql/bin/sqlservr
root          9  2.2 19.7 9034224 754820 ?      Sl   Jul23   0:28 /opt/mssql/bin/sqlservr
root        319  0.0  0.0  13908  3400 ?        S    00:01   0:00 sh
root        326  0.0  0.1  53832  3900 ?        R+   00:02   0:00 ps aux

 

Well, we used different Podman commands to spin up a container that meets the OCI specification like Docker. For a sake of curiosity, let’s build a custom image from a Dockerfile. In fact, this is a custom image we developed for customers to meet our best practices requirements. 

$ ls -l
total 40
drwxrwxr-x. 2 clustadmin clustadmin   70 Jul 24 02:06 BestPractices
drwxrwxr-x. 2 clustadmin clustadmin   80 Jul 24 02:06 DMK
-rw-rw-r--. 1 clustadmin clustadmin  614 Jul 24 02:06 docker-compose.yml
-rw-rw-r--. 1 clustadmin clustadmin 2509 Jul 24 02:06 Dockerfile
-rw-rw-r--. 1 clustadmin clustadmin 3723 Jul 24 02:06 entrypoint.sh
-rw-rw-r--. 1 clustadmin clustadmin 1364 Jul 24 02:06 example.docker-swarm-compose.yml
-rw-rw-r--. 1 clustadmin clustadmin  504 Jul 24 02:06 healthcheck.sh
-rw-rw-r--. 1 clustadmin clustadmin   86 Jul 24 02:06 mssql.conf
-rw-rw-r--. 1 clustadmin clustadmin 4497 Jul 24 02:06 postconfig.sh
-rw-rw-r--. 1 clustadmin clustadmin 2528 Jul 24 02:06 Readme.md
drwxrwxr-x. 2 clustadmin clustadmin   92 Jul 24 02:06 scripts

 

To build an image from a Dockerfile the corresponding Podman command is as follow:

$ sudo podman build -t dbi_mssql_linux:2019-CTP3.1 .
…
--> 5db120fba51f3adc7482ec7a9fed5cc4194f13e97b855d9439a1386096797c39
STEP 65: FROM 5db120fba51f3adc7482ec7a9fed5cc4194f13e97b855d9439a1386096797c39
STEP 66: EXPOSE ${MSSQL_TCP_PORT}
--> 8b5e8234af47adb26f80d64abe46715637bd48290b4a6d7711ddf55c393cd5a8
STEP 67: FROM 8b5e8234af47adb26f80d64abe46715637bd48290b4a6d7711ddf55c393cd5a8
STEP 68: ENTRYPOINT ["/usr/local/bin/entrypoint.sh"]
--> 11045806b8af7cf2f67e5a279692e6c9e25212105bcd104ed17b235cdaea97fe
STEP 69: FROM 11045806b8af7cf2f67e5a279692e6c9e25212105bcd104ed17b235cdaea97fe
STEP 70: CMD ["tail -f /dev/null"]
--> bcb8c26d503010eb3e5d72da4b8065aa76aff5d35fac4d7958324ac3d97d5489
STEP 71: FROM bcb8c26d503010eb3e5d72da4b8065aa76aff5d35fac4d7958324ac3d97d5489
STEP 72: HEALTHCHECK --interval=15s CMD [ "/usr/local/bin/healthcheck.sh" ]
--> e7eedf0576f73c95b19adf51c49459b00449da497cf7ae417e597dd39a9e4c8f
STEP 73: COMMIT dbi_mssql_linux:2019-CTP3.1

 

The image built is now available in the local repository:

$ sudo podman images
REPOSITORY                            TAG           IMAGE ID       CREATED         SIZE
localhost/dbi_mssql_linux             2019-CTP3.1   e7eedf0576f7   2 minutes ago   1.79 GB
mcr.microsoft.com/mssql/rhel/server   2019-CTP3.1   dbba412361d7   3 weeks ago     1.79 GB

 

The next step will consist in spinning up a SQL Server container based on this new image. Note that I used a custom parameter DMK=Y to drive the creation of the DMK maintenance tool in our case which including the deployment of a custom dbi_tools database ans related objects that carry out the database maintenance.

$ sudo podman run -d -e 'ACCEPT_EULA=Y' \
> -e 'MSSQL_SA_PASSWORD=Password1' -e 'DMK=Y'  \
> --name 'sqltest2' \
> -p 1470:1433 \
> localhost/dbi_mssql_linux:2019-CTP3.1
d057e0ca41f08a948de4206e9aa07b53450c2830590f2429e50458681d230f6b

 

Let’s check if the dbi_tools has been created during the container runtime phase:

$ sudo podman exec -ti d057 /opt/mssql-tools/bin/sqlcmd -S localhost -U sa -P Password1 -Q"SELECT name from sys.databases"
name
--------------------------------------------------------------------------------------------------------------------------------
master
tempdb
model
msdb
dbi_tools

 

Finally, to make the transition with a future blog post, the Podman tool comes with extra commands (under development) that is not available with Docker CLI. The following example generates a YAML deployment file and the corresponding service from an existing container. Please note however that containers with volumes are not supported yet.

The container definition is a follows:

$ sudo podman run -d -e 'ACCEPT_EULA=Y' -e \
'MSSQL_SA_PASSWORD=Password1'  \
--name 'sqltestwithnovolumes' \
-p 1480:1433 \
mcr.microsoft.com/mssql/rhel/server:2019-CTP3.1
7e99581eaec4c91d7c13af4525bfb3805d5b56e675fdb53d0061c231294cd442

 

And we get the corresponding YAML file generated by the Podman command:

$ sudo podman generate kube -s 7e99
# Generation of Kubernetes YAML is still under development!
#
# Save the output of this file and use kubectl create -f to import
# it into Kubernetes.
#
# Created with podman-1.0.2-dev
apiVersion: v1
kind: Pod
metadata:
  creationTimestamp: 2019-07-24T03:52:18Z
  labels:
    app: sqltestwithnovolumes
  name: sqltestwithnovolumes
spec:
  containers:
  - command:
    - /opt/mssql/bin/sqlservr
    env:
    - name: PATH
      value: /usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
    - name: TERM
      value: xterm
    - name: HOSTNAME
    - name: container
      value: oci
    - name: ACCEPT_EULA
      value: "Y"
    - name: MSSQL_SA_PASSWORD
      value: Password1
    image: mcr.microsoft.com/mssql/rhel/server:2019-CTP3.1
    name: sqltestwithnovolumes
    ports:
    - containerPort: 1433
      hostPort: 1480
      protocol: TCP
    resources: {}
    securityContext:
      allowPrivilegeEscalation: true
      capabilities: {}
      privileged: false
      readOnlyRootFilesystem: false
    workingDir: /
status: {}
---
apiVersion: v1
kind: Service
metadata:
  creationTimestamp: 2019-07-24T03:52:18Z
  labels:
    app: sqltestwithnovolumes
  name: sqltestwithnovolumes
spec:
  ports:
  - name: "1433"
    nodePort: 30309
    port: 1433
    protocol: TCP
    targetPort: 0
  selector:
    app: sqltestwithnovolumes
  type: NodePort
status:
  loadBalancer: {}

 

By default the service type NodePort has been created by the command. This latest command needs further testing for sure!

See you

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

David Barbarin
David Barbarin

Principal Consultant & Microsoft Technology Leader