Running Portworx in Production with DC/OS
DAY 1 Operations
Initial Software Setup for Production
- Deployment - Portworx can be deployed either as a DC/OS framework or as a systemd service directly in the system Refer to the install instructions for running Portworx as a DC/OS framework page. This deploys Portworx as a OCI container and as a framework in DC/OS
- Deploy Portworx directly as OCI container
- Ensure all nodes in the cluster have NTP running and the times are synchronized across all the nodes that will form the Portworx cluster
- All nodes in the cluster should have achieved quorum and
pxctl statusshould display the cluster as
- etcd - Setup etcd as a 3-node etcd cluster outside the container orchestrator to ensure maximum stability. Refer to the following page on how to install etcd and also configure it for maximum stability.
Configuring the Server or the Compute Infrastructure
- Check and ensure a minimum 2 cores and 4GB of RAM are allocated for Portworx.
The base operating system of the server supports linux kernel 3.10+ . Newer 4.x linux kernels have many performance and stability related fixes and is recommended.
Configuring the Networking Infrastructure
Make sure the following ports are open in all the servers. 9001-9022
Configure separate networks for Data and Management networks to isolate the traffic
Data network is specified giving the ‘-d’ switch and Management networks with the ‘-m’ switch. Refer to scheduler guides for specifics to enable it in your scheduler.
With multiple NICs, create a bonded ethernet port for data interface for improved availability and performance.
Configuring and Provisioning Underlying Storage
Selecting drives for an installation
Storage can be provided to Portworx explicitly by passing in a list of block devices.
lsblk -awill display a list of devices on the system. This is accomplished by the ‘-s’ flag as a runtime parameter. It can also be provided implicity by passing in the ‘-a’ flag. In this mode, Portworx will pick up all the available drives that are not in use. When combined with ‘-f’, Portworx will pick up drives even if they have a filesystem on them (mounted drives are still excluded). Note that not all nodes need to contribute storage; a node can operate in the storageless mode with the ‘-z’ switch. Refer to scheduler guides for specifics for your scheduler.
HW RAID - If there are many drives in a server and drive failure tolerance is required per server, enable HW RAID (if available) and give the block device from a HW RAID volume for Portworx to manage.
Portworx classifies drive media into different performance levels and groups them in separate pools for volume data. These levels are called
io_priorityand they offer the levels
io_priorityof a pool is determined automatically by Portworx. If the intention is to run low latency transactional workloads like databases on Portworx, then Portworx, Inc. recommends having NVMe or other SAS/SATA SSDs in the system. Pool priority can be managed as documented here
This page offers more information on different io_prioirty levels
Working with drives with AWS Auto scaling group
Portworx supports automatic management of EBS volumes. If you are using AWS ASG to manage Portworx nodes,then you should to use the ASG feature
Portworx Node Topology
Portworx replicated volumes distributes data across failure domains. For on-premises installations, this ensures that a power failure to a rack does not result in data unavailability. For cloud deployments this ensures data availability across zones.
Topology in cloud environments
Portworx auto-detects availabilty zones and regions and provisions replicas across
different zones. For e.g., see below for the partial output of
Status: PX is operational License: Trial (expires in 23 days) Node ID: a17f382d-b2ef-41b8-81fc-d9b86d56b5d1 IP: 172.31.51.89 Local Storage Pool: 2 pools POOL IO_PRIORITY RAID_LEVEL USABLE USED STATUS ZONE REGION 0 LOW raid0 64 GiB 1.1 GiB Online b us-east-1 1 LOW raid0 128 GiB 65 GiB Online b us-east-1 ... ...
This node is in us-east-1. If Portworx is started in other zones, then when a volume with greater than 1 replication factor is created, it will have the replicas automatically created in other nodes in other zones.
Topology in on-premises deployments
Failure domains in terms of RACK information can be passed in as described here
Volume Management Best Practices
Volumes - Portworx volumes are thinly provisioned by default. Make sure to monitor for capacity for
VolumeSpaceLowalerts. These alerts are triggered when the free available space goes below a threshold. See this page for more details.
For applications needing node level availability and read parallelism across nodes, it is recommended to set the volumes with replication factor 2 or replication factor 3/
Here is how one can configure a volume with replication factor 3:
pxctl volume create dbasevol --size=1 --repl=3 --iopriority=high
Portworx makes the best effort to distribute volumes evenly across all nodes and based on the
ioprioritythat is requested. When Portworx cannot find the appropriate media type that is requested to create a given
ioprioritytype, it will attempt to create the volume with the next available
Volumes can be created in different availability zones by using the
--zonesoption in the
pxctl volume createcommand:
pxctl volume create dbasevol --size=1 --repl=3 --iopriority=high --zones=us-east-1,us-east-2,us-east-3
Volumes can be created in different racks using
--racksoption and passing the rack labels when creating the volume:
pxctl volume create dbasevol --size=1 --repl=3 --iopriority=high --racks=dcrack1,dcrack2,dcrack3
Please ensure the Portworx containers are started with the corresponding rack parameters.
If the volumes need to be protected against accidental deletes because of background garbage collecting scripts, then the volumes need to be created with the
pxctl volume create dbasevol --size=1 --repl=3 --iopriority=high --sticky
--stickyflag can be turned on and off using the
pxctl volume updatecommands:
pxctl volume update dbasevol --sticky=off
For applications that require shared access from multiple containers running in different hosts, Portworx, Inc. recommends running shared volumes. Shared volumes can be configured as follows:
pxctl volume create wordpressvol --shared --size=100 --repl=3
This page gives more details on different volume types, how to create them and update the configuration for the volumes
Data Protection for Containers
Snapshots - Follow DR best practices and ensure volume snapshots are scheduled for instantaneous recovery in case of app failures.
Portworx supports 64 snapshots per volume
Each snapshot can be taken manually via the
pxctl snap createcommand like this:
pxctl snap create --name mysnap --label color=blue,fabric=wool myvol
Volume successfully snapped: 1152602487227170184
Alternatively, snapshots can be scheduled by creating a hourly, daily or weekly schedule. This will enable the snapshots to be automatically created without user intervention:
pxctl volume create --daily @08:00 --daily @18:00 --weekly Friday@23:30 --monthly 1@06:00 myvol
Here is more information on how to setup snapshots in Portworx Enterprise .
For DR, it is recommended to setup cloudsnaps as well. This is covered in detail in the Day 3 - Cloudsnaps section
Alerts and Monitoring for Production
Portworx, Inc. recommends setting up monitoring with Prometheus and AlertsManager to ensure monitoring of the data services infrastructure for your containers
While Prometheus can be deployed as a container within the container orchestrator, many of Portworx, Inc.’s production customers deploy Prometheus in a separate cluster that is dedicated for managing and monitoring their large scale container orchestrator infrastructure.
- The Portworx integration with Prometheus page shows how Prometheus can be set up to monitor Portworx
- Configure Grafana via this template
- Here is how Alerts Manager can be configured for looking for alerts with Alerts Manager
- List of Portworx Alerts are documented here
Day 2 Operations
Hung Node Recovery
A Portworx node may hang or appear to hang because of the following reasons:
- Underlying media being too slow to respond and thus Portworx trying to error recovery of the media
- Kernel hangs or panics that are impacting overall operations of the system
- Other applications that are not properly constrainted putting heavy memory pressure on the system
- Applications consuming a lot of CPU that are not properly constrained.
Docker Daemon issues where Docker itself has hung and thus resulting on all other containers not responding properly.
Running Portworx as a OCI container greatly alleviates any issues introduced by Docker Daemon hanging or not being responsive as Portworx runs as a OCI container and not as a docker container thus eliminating the docker dependency.
If Portworx appears to not respond, a restart of the Portworx OCI container via
Any Portworx restart within 10 mins will ensure that applications continue to run without experiencing volume unmounts/outage.
Stuck Volume Detection and Resolution
With DC/OS, it is possible that even after the application container terminates, a volume is left attached. This volume is still available for use in any other node. Portworx makes sure that if a volume is not in use by an application, it can be attached to any other node in the system.
With this attach operation, Portworx will automatically manage the volume attach status with no user intervention required and continue to serve the volume I/Os even a container attaches to the same volume from a different node.
Scaling out a cluster nodes in the Cloud and On-Prem
Scaling out a cluster in cloud
- The best way to scale a cluster is via ASG integration on AWS.
- This feature is called Stateful Autoscaling and is described here.
- Perform sizing of your data needs and determine the amount and type of storage (EBS volumes) needed per ECS instance.
- Create EBS volume templates to match the number of EBS volumes needed per EC2 instance.
- Create a stateful AMI to associate it with your auto-scaling group.
- Once everything is setup, as described in the steps above, then the cluster can be scaled up and down via ASG. Portworx will automatically manage the EBS volume creation and preserve the volumes across the cluster scaling up and down. This page desribes how Portworx handles the volume management in a auto-scaling cluster.
Scaling out a cluster on-prem
The best way to scale the cluster on-prem is by having the new nodes join the existing cluster. This page shows how to scale up a existing cluster by adding more nodes.
Using DC/OS, if Portworx is installed as a framework, you can also scale a Portworx cluster by using the DC/OS Portworx framework.
Cluster Capacity Expansion
- Cluster storage capacity can be expanded by adding more drives each node.
- Drives with similar capacity (within 1GB capacity difference) will be grouped together as a same pool.
- Drives can be added per node and Portworx will add that to the closest pool size by drive size.
- Before adding drives to the node, the node will need to be taken into maintenance mode
- Ensure the volumes in the node have replicas in other nodes.
- If the volumes have replication factor of 1, increase the replication factor.
- Ensure the services are failed over to a different node when the node is taken into maintenance mode.
- Follow the instructions in this page to add storage each node.
Server and Networking Replacements and Upgrades
- Servers running Portworx can be replaced by performing decommissioning of the server to safely remove them from the cluster.
- Ensure that all the volumes in the cluster are replicated before decommissioning the node so that the data is still available for the containers mounting the volumes after the node is decommisioned.
pxctl cluster deletecommand to manually remove the node from the cluster.
- Follow the instructions in this page to delete nodes in the cluster
- Once the node is decommissioned, components like network adapters, storage adapters that need to be replaced can be replaced.
- The server can be replaced as well.
- Once the replacement is done, the node can be joined back to the cluster by going through the steps described in the scaling-out the cluster section.
Portworx Software Upgrades
- Work with Portworx, Inc.’s support team before planning major upgrades. Ensure all volumes have the latest snapshots before performing upgrades.
- Ensure there are cloudsnaps that are taken for all the volumes.
- If you are using the the Portworx DC/OS framework for deploying Portworx and running Portworx as OCI format container, follow this link to perform the upgrades.
- If you are running Portworx as a systemd service, follow this link.
- Work with Portworx, Inc.’s support team before planning major upgrades. Ensure all volumes have the latest snapshots before performing upgrade.
- Ensure there are cloudsnaps that are taken.
- Due to a bug in DCOS the Portworx service file will get removed after an upgrade. Please run
dcos portworx plan force-restart deploy portworx-deployafter the DCOS upgrade to re-deploy the service file.
- After the migration, relaunch Portworx and ensure that the entire cluster is online by running
- Check if the DC/OS services via marathon and any other frameworks can mount the Portworx volumes from the marathon UI or the DC/OS UI.
OS upgrades and Docker Upgrades
- Work with Portworx, Inc.’s support team before planning major upgrades. Ensure all volumes have the latest snapshots before performing the upgrade.
- Ensure kernel-devel packages are installed after a OS migration.
- If Portworx is run as a OCI container, Docker Upgrades and Restarts do not impact Portworx runtime. Portworx, Inc. recommends you run Portworx as a OCI container.
Day 3 Operations
Handling Lost or Stale Nodes on the Cloud and On-Prem
- Lost or Stale Nodes can be removed from the Portworx cluster for force-decommissioning the node from the cluster.
- The command used to remove a node is
pxctl cluster delete --force.
- For e.g., if a specific node is offline but it no longer exists, use
pxctl cluster delete --force node-idto remove the node from the cluster.
Volume Data Recovery
Disaster Recovery with Cloudsnaps
- It is recommended to setup cloudsnaps for volume backup and recovery to handle DR scenarios.
- Cloudsnaps are also good way to perform cluster to cluster data migration.
- Cloudsnaps can work with Amazon S3, Azure Blob, Google Cloud Storage or any S3 compatible object store.
- Cloudsnaps store the volume snaps in the cloud and on import, can roll up all the snaps and import a point-in-time copy of the volume into the cluster.
- It is recommended to take atleast one cloudsnap a day for each volume in production in the cluster.
- Cloudsnaps can be scheduled via the Portworx CLI for hourly, daily, weekly or monthly snaps.
- Cloudsnaps can also be scheduled to happen at a particular time. It is recommended to schedule cloudsnaps at a time when the application data traffic is light to ensure faster back-ups.
- Follow DR best practices and setup periodic cloudsnaps so in case of a disaster, Portworx volumes can be restored from an offsite backup.
- Any drive in a given node can be replaced by another drive in the same node.
- In order to perform a drive replacement, the Portworx node must be put into
Step 1: Enter Maintenance mode
pxctl service maintenance --enter
This is a disruptive operation, PX will restart in maintenance mode. Are you sure you want to proceed ? (Y/N): y PX is not running on this host.
Step 2: Replace old drive with a new drive
Ensure the replacement drive is already available in the system.
For e.g., replace drive /dev/sde with /dev/sdc:
pxctl service drive replace --source /dev/sde --target /dev/sdc --operation start
"Replace operation is in progress"
Check the status:
pxctl service drive replace --source /dev/sde --target /dev/sdc --operation status
"Started on 16.Dec 22:17:06, finished on 16.Dec 22:17:06, 0 write errs, 0 uncorr. read errs\n"
Step 3: Exit Maintenance mode
pxctl service maintenance --exit
PX is now operational
Step 4: Check if the drive has been successfully replaced
pxctl service drive show
PX drive configuration: Pool ID: 0 IO_Priority: LOW Size: 15 TiB Status: Online Has meta data: No Drives: 1: /dev/sdc, 3.0 GiB allocated of 7.3 TiB, Online 2: /dev/sdb, 0 B allocated of 7.3 TiB, Online Pool ID: 1 IO_Priority: HIGH Size: 1.7 TiB Status: Online Has meta data: Yes Drives: 1: /dev/sdj, 1.0 GiB allocated of 1.7 TiB, Online
- If there is no spare drive available in the system, then the following steps need to be performed:
- Decommission the Portworx node (Refer to
pxctl cluster delete).
- Ensure all volumes have replicas in other nodes if you still need to access the data.
- Replace the bad drive(s) with new drive(s).
- Add the node to the cluster as a new node. (refer to adding cluster nodes).
- Ensure the cluster is operational and the new node has been added to the cluster via
pxctl cluster statusand
pxctl cluster list.
- Decommission the Portworx node (Refer to