Nowcast Production Deployments

In October 2016 the production deployment of the nowcast system was changed to use the SalishSeaNowcast package that is based on the NEMO_Nowcast framework framework. The production deployment uses 2 systems:

1. The nemo_nowcast.message_broker, nemo_nowcast.manager, nemo_nowcast.log_aggregator, most of the pre- and post-processing workers run on the Salish Sea Model Results Server, :m`, where the deployment is in the /SalishSeaCast/ directory tree.

2. The daily forecast2 (preliminary forecast), nowcast, forecast, and nowcast-green NEMO-3.6 model runs are computed on a cluster of virtual machines provided by Ocean Networks Canada on the Compute Canada arbutus.cloud cluster. The shared storage for those VMs is provided by an NFS-mounted volume of arbutus.cloud Ceph object storage. The nowcast deployment is in the /nemoShare/MEOPAR/nowcast-sys/ directory tree.

   In April 2017, daily wwatch3-nowcast, daily wwatch3-forecast, and wwatch3-forecast2 (preliminary wave forecast) WaveWatch III® v5.16 wave model runs for the Strait of Georgia and Juan de Fuca Strait were added to the computations on arbutus.cloud. The wwatch3-nowcast and wwatch3-forecast runs are executed in sequence after the daily nowcast-green NEMO-3.6 runs. The wwatch3-forecast2 runs are executed after the daily forecast2 NEMO-3.6 runs.

   In January 2018, daily fvcom-nowcast and fvcom-forecast FVCOM v4.1-beta model runs for Vancouver Harbour and the Lower Fraser River were added to the computations on the arbutus.cloud. They are executed in sequence after the daily nowcast NEMO-3.6 runs. In January 2019, the resolution of the Vancouver Harbour and Lower Fraser River FVCOM v4.1-beta model was increased. Those runs are designated fvcom-nowcast-x2 and fvcom-forecast-x2. In March 2019, an even higher resolution Vancouver Harbour and Lower Fraser River model configuration was added to the system, running daily nowcast runs as fvcom-nowcast-r12.

These sections describe the setup of the nowcast system on skookum and arbutus.cloud, and their operation.

• skookum Deployment
  • Git Repositories
  • Python Packages
  • Environment Variables
  • Nowcast Runs Directories
  • ECCC MSC Datamart Mirror Directories
  • Logging Directories
  • Static Web Site Assets Directories
  • Persistent Dask Cluster for make_averaged_dataset Worker
  • ssh Keys and Configuration
• arbutus.cloud Deployment
  • Web Interface
  • ssh Access
  • Provisioning and Configuration
  • Hosts Mappings
  • Git Repositories
  • Build XIOS-2
  • Build NEMO-3.6
  • Build WAVEWATCH III ^®
  • Build FVCOM-4.1
  • Python Packages
  • Environment Variables
  • NEMO Runs Directory
  • WaveWatch Runs Directories
  • FVCOM Runs Directory
  • Managing Compute Nodes
• Nowcast System Operations
  • ssh Hosts and Keys Configuration
  • Cold Start
  • System Management
  • Automatic Deployment of Changes to salishsea-site App

In May 2018 production runs of a nowcast-green configuration with AGRIF sub-grids for Baynes Sound and Haro Strait were added to the system. Those runs are executed on a reserved chassis on orcinus. The setup on orcinus, as well are the sub-grid initialization preparation with the NEMO-AGRIF nesting tools are described in:

• orcinus Deployment for nowcast-agrif Runs
  • Create Directory Trees
  • Clone Git Repositories
  • Build XIOS-2
  • Build NEMO-3.6
  • Install Python Packages
  • Populate Run Preparation Directory Tree
  • Sub-grid Initialization Preparation with Nesting Tools

In February 2019 we got access to the UBC EOAS optimum cluster. We use it primarily for long hindcast runs, but also some research runs. The setup on optimum is described in:

• optimum Deployment for Hindcast Runs
  • Environment Variables
  • Module Loads
  • Create Directory Trees
  • Clone Git Repositories
  • Build XIOS-2
  • Build NEMO-3.6 and REBUILD_NEMO
  • Install Python Packages
  • Populate Run Preparation Directory

See also the #optimum-cluster Slack channel.

With the update of the production to run the V21-11 model version in January 2024, we decided to end the daily nowcast-dev development model runs on salish. Development is now generally done in research runs on graham. salish is now mostly used for analysis tasks, post-processing of NEMO model results files to produce day-average and month-average dataset files, and Lagrangian particle tracking analysis with ariane and OceanParcels.